feat(httpapi): harden control plane and speed up prefix refresh

Очередь job получила lease и fencing, телеметрия и прогресс deploy сохраняются, ревизия пишет manifest входов. Обновление префиксов не переписывает неизменённый снимок, большие наборы идут через COPY, а устаревшая ревизия не попадает в deploy. Бандл ноды кэшируется, prune и постановка job больше не держат горячий путь.

Co-authored-by: Cursor <[email protected]>

.gitea/workflows/quality.yaml

@@ -322,6 +322,66 @@ jobs:
             go build -o "$out/$name" "./$d"
           done
 
+  postgres-integration:
+    needs: [changes]
+    if: needs.changes.outputs.go == 'true'
+    runs-on: ubuntu-latest
+    services:
+      postgres:
+        image: postgres:16
+        env:
+          POSTGRES_USER: evobgp
+          POSTGRES_PASSWORD: evobgp_ci
+          POSTGRES_DB: evobgp_test
+        ports:
+          - 5432:5432
+        options: >-
+          --health-cmd "pg_isready -U evobgp -d evobgp_test"
+          --health-interval 5s
+          --health-timeout 5s
+          --health-retries 12
+    steps:
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+      - uses: actions/setup-go@d35c59abb061a4a6fb18e82ac0862c26744d6ab5 # v5.5.0
+        with:
+          go-version: "1.24"
+          cache: false
+      - name: Export cache paths
+        run: sh scripts/ci/export-cache-env.sh
+      - id: go-hash
+        run: echo "key=$(sha256sum go.sum | awk '{print $1}')" >> "$GITHUB_OUTPUT"
+      - uses: actions/cache@5a3ec84eff668545956fd18022155c47e93e2684 # v4.2.3
+        with:
+          path: |
+            ${{ env.GOMODCACHE }}
+            ${{ env.GOCACHE }}
+          key: go-${{ runner.os }}-1.24-postgres-${{ steps.go-hash.outputs.key }}
+          restore-keys: |
+            go-${{ runner.os }}-1.24-postgres-
+      - name: Run migrations and verify PostgreSQL integration tests executed
+        env:
+          GOFLAGS: -mod=readonly
+          EVOBGP_TEST_DATABASE_URL: postgres://evobgp:[email protected]:5432/evobgp_test?sslmode=disable
+        run: bash scripts/ci/postgres-integration.sh
+
+  legacy-compatibility:
+    needs: [changes]
+    if: needs.changes.outputs.go == 'true'
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        legacy_version: ["2.56.2", "1.0.0"]
+    steps:
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+      - uses: actions/setup-go@d35c59abb061a4a6fb18e82ac0862c26744d6ab5 # v5.5.0
+        with:
+          go-version: "1.24"
+          cache: false
+      - name: Build pinned legacy executables and test current control plane
+        env:
+          EVOBGP_LEGACY_VERSION: ${{ matrix.legacy_version }}
+        run: bash scripts/ci/legacy-node-compatibility.sh
+
   bird2:
     runs-on: ubuntu-latest
     needs: [changes, go]
@@ -335,6 +395,10 @@ jobs:
        needs.changes.outputs.docker_go == 'true')
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+      - uses: actions/setup-go@d35c59abb061a4a6fb18e82ac0862c26744d6ab5 # v5.5.0
+        with:
+          go-version: "1.24"
+          cache: false
       - name: Install bird2 (репозиторий Ubuntu runner, как в образе evobgp-bird2)
         run: |
           set -euxo pipefail
@@ -358,6 +422,11 @@ jobs:
             echo "==> $conf"
             bird -c "$WS/$conf" -p
           done
+      - name: Parse staged birddeploy generation with real BIRD
+        env:
+          GOFLAGS: -mod=readonly
+          EVOBGP_TEST_BIRD_BIN: bird
+        run: go test ./internal/birddeploy -run '^TestRealBirdParseGeneration$' -count=1 -v
 
   commitlint:
     if: inputs.is_pull_request
@@ -405,6 +474,8 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+      - name: Offline mirror behavior tests
+        run: bash scripts/tests/mirror-base-images.test.sh
       - uses: docker/setup-buildx-action@e468171a9de216ec08956ac3ada2f0791b6bd435 # v3.11.1
         with:
           name: evobgp

.gitignore

@@ -32,3 +32,6 @@ deploy/compose/runtime-logs/
 .cursor/mcp.json
 .mcp.json
 .codegraph/daemon.pid
+
+# Local backend resource snapshots (host metadata, not source artifacts)
+artifacts/backend-baseline/

cmd/evobgp-agent/main.go

@@ -6,10 +6,14 @@ import (
 	"fmt"
 	"log"
 	"os"
+	"os/signal"
+	"strings"
 	"sync"
+	"syscall"
 	"time"
 
 	"evobgp/internal/agentserver"
+	"evobgp/internal/agentwatch"
 	"evobgp/internal/birdfmt"
 )
 
@@ -18,19 +22,22 @@ func main() {
 	birdc := flag.String("birdc", "", "path to birdc binary (default: birdc from PATH)")
 	socket := flag.String("socket", "", "optional birdc control socket (-s)")
 	timeout := flag.Duration("timeout", 30*time.Second, "timeout for bird/birdc")
-	watchEvery := flag.Duration("watch-interval", 30*time.Second, "for watch: interval between birdc configure")
+	watchEvery := flag.Duration("watch-interval", 30*time.Second, "for watch: interval between configuration checks")
 	listen := flag.String("listen", "", "for serve: listen address (default :8443 or EVOBGP_AGENT_LISTEN)")
 	flag.Usage = func() {
 		fmt.Fprintf(os.Stderr, "Usage: %s [flags] <command>\n", os.Args[0])
 		fmt.Fprintf(os.Stderr, "Commands:\n")
 		fmt.Fprintf(os.Stderr, "  parse-check <path/to/bird.conf>   run bird -c <path> -p (syntax check)\n")
 		fmt.Fprintf(os.Stderr, "  configure                         run birdc configure (reload running BIRD)\n")
-		fmt.Fprintf(os.Stderr, "  watch                             periodically run birdc configure (compose sidecar)\n")
+		fmt.Fprintf(os.Stderr, "  watch                             reload changed BIRD configuration and periodically reconcile\n")
 		fmt.Fprintf(os.Stderr, "  serve                             Panel→Node HTTP API (POST /v1/agent/sync)\n")
 		flag.PrintDefaults()
 	}
-	flag.Parse()
+	args, err := parseAgentArgs(flag.CommandLine, os.Args[1:])
-	args := flag.Args()
+	if err != nil {
+		fmt.Fprintln(os.Stderr, err)
+		os.Exit(2)
+	}
 	if len(args) < 1 {
 		flag.Usage()
 		os.Exit(2)
@@ -47,10 +54,12 @@ func main() {
 	switch args[0] {
 	case "serve":
 		runServe(*listen, *timeout)
-	case "parse-check", "configure", "watch":
+	case "watch":
+		runWatch(ctl, *watchEvery, *timeout)
+	case "parse-check", "configure":
 		ctx, cancel := context.WithTimeout(context.Background(), *timeout)
 		defer cancel()
-		runBirdCommand(ctx, args, ctl, *watchEvery, *timeout)
+		runBirdCommand(ctx, args, ctl)
 	default:
 		fmt.Fprintf(os.Stderr, "unknown command: %s\n", args[0])
 		flag.Usage()
@@ -58,7 +67,47 @@ func main() {
 	}
 }
 
-func runBirdCommand(ctx context.Context, args []string, ctl *birdfmt.BirdCtl, watchEvery, timeout time.Duration) {
+// parseAgentArgs accepts flags before or immediately after the subcommand.
+// A second parse preserves previously supplied flag values and rejects unknown
+// flags instead of silently passing them to watch or serve as positional args.
+func parseAgentArgs(fs *flag.FlagSet, argv []string) ([]string, error) {
+	if err := fs.Parse(argv); err != nil {
+		return nil, err
+	}
+	args := fs.Args()
+	if len(args) == 0 {
+		return args, nil
+	}
+	command := args[0]
+	globalStop := false
+	for i, consumed := 0, len(argv)-len(args); i < consumed; i++ {
+		if argv[i] == "--" {
+			globalStop = true
+			break
+		}
+		name, _, hasValue := strings.Cut(strings.TrimLeft(argv[i], "-"), "=")
+		if !hasValue {
+			if f := fs.Lookup(name); f != nil {
+				boolean, ok := f.Value.(interface{ IsBoolFlag() bool })
+				if !ok || !boolean.IsBoolFlag() {
+					i++
+				}
+			}
+		}
+	}
+	if !globalStop {
+		if err := fs.Parse(args[1:]); err != nil {
+			return nil, err
+		}
+		args = append([]string{command}, fs.Args()...)
+	}
+	if (command == "watch" || command == "serve" || command == "configure") && len(args) != 1 {
+		return nil, fmt.Errorf("%s takes no positional arguments", command)
+	}
+	return args, nil
+}
+
+func runBirdCommand(ctx context.Context, args []string, ctl *birdfmt.BirdCtl) {
 	switch args[0] {
 	case "parse-check":
 		if len(args) != 2 {
@@ -78,21 +127,33 @@ func runBirdCommand(ctx context.Context, args []string, ctl *birdfmt.BirdCtl, wa
 			fmt.Fprintln(os.Stderr, err)
 			os.Exit(1)
 		}
-	case "watch":
+	}
-		if watchEvery <= 0 {
+}
-			fmt.Fprintln(os.Stderr, "watch-interval must be > 0")
+
+func runWatch(ctl *birdfmt.BirdCtl, every, timeout time.Duration) {
+	if every <= 0 || timeout <= 0 {
+		fmt.Fprintln(os.Stderr, "watch-interval and timeout must be > 0")
 		os.Exit(2)
 	}
-		log.Printf("evobgp-agent watch: birdc configure every %s (socket=%q)", watchEvery, ctl.Socket)
+	reconcile := 5 * time.Minute
-		for {
+	if raw := os.Getenv("EVOBGP_AGENT_WATCH_RECONCILE_INTERVAL"); raw != "" {
-			cctx, cancel := context.WithTimeout(context.Background(), timeout)
+		var err error
-			err := ctl.Configure(cctx)
+		reconcile, err = time.ParseDuration(raw)
-			cancel()
+		if err != nil || reconcile <= 0 {
-			if err != nil {
+			fmt.Fprintln(os.Stderr, "EVOBGP_AGENT_WATCH_RECONCILE_INTERVAL must be a positive duration")
-				log.Printf("evobgp-agent watch: configure: %v", err)
+			os.Exit(2)
 		}
-			time.Sleep(watchEvery)
 	}
+	active := os.Getenv("EVOBGP_BIRD_ACTIVE_DIR")
+	if active == "" {
+		active = "/etc/bird"
+	}
+	ctx, stop := signal.NotifyContext(context.Background(), os.Interrupt, syscall.SIGTERM)
+	defer stop()
+	log.Printf("evobgp-agent watch: checking %s every %s, forced reconciliation every %s (socket=%q)", active, every, reconcile, ctl.Socket)
+	if err := agentwatch.Run(ctx, ctl, agentwatch.Config{ActiveDir: active, Every: every, ReconcileEvery: reconcile, CommandTimeout: timeout}); err != nil {
+		fmt.Fprintln(os.Stderr, err)
+		os.Exit(1)
 	}
 }
 

cmd/evobgp-agent/main_test.go

@@ -0,0 +1,49 @@
+package main
+
+import (
+	"flag"
+	"io"
+	"reflect"
+	"testing"
+	"time"
+)
+
+func TestAgentFlagsBeforeAndAfterCommand(t *testing.T) {
+	for _, tc := range []struct {
+		name           string
+		args           []string
+		want           []string
+		socket, listen string
+		every          time.Duration
+	}{
+		{"legacy flags first", []string{"-socket=/custom/bird.ctl", "-watch-interval=45s", "watch"}, []string{"watch"}, "/custom/bird.ctl", "", 45 * time.Second},
+		{"compose command first", []string{"watch", "-socket=/run/bird/bird.ctl", "-watch-interval=30s"}, []string{"watch"}, "/run/bird/bird.ctl", "", 30 * time.Second},
+		{"serve", []string{"serve", "-listen=127.0.0.1:9999"}, []string{"serve"}, "", "127.0.0.1:9999", time.Minute},
+		{"mixed", []string{"-socket=/first", "watch", "-watch-interval=5s"}, []string{"watch"}, "/first", "", 5 * time.Second},
+		{"parse check", []string{"parse-check", "bird.conf"}, []string{"parse-check", "bird.conf"}, "", "", time.Minute},
+		{"global delimiter", []string{"--", "parse-check", "-bird.conf"}, []string{"parse-check", "-bird.conf"}, "", "", time.Minute},
+		{"flags then delimiter", []string{"-socket=/first", "--", "parse-check", "-bird.conf"}, []string{"parse-check", "-bird.conf"}, "/first", "", time.Minute},
+		{"delimiter as value", []string{"-socket", "--", "watch", "-watch-interval=5s"}, []string{"watch"}, "--", "", 5 * time.Second},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			fs := flag.NewFlagSet("agent", flag.ContinueOnError)
+			fs.SetOutput(io.Discard)
+			socket := fs.String("socket", "", "")
+			listen := fs.String("listen", "", "")
+			every := fs.Duration("watch-interval", time.Minute, "")
+			args, err := parseAgentArgs(fs, tc.args)
+			if err != nil || !reflect.DeepEqual(args, tc.want) || *socket != tc.socket || *listen != tc.listen || *every != tc.every {
+				t.Fatalf("args=%v socket=%s listen=%s every=%s err=%v", args, *socket, *listen, *every, err)
+			}
+		})
+	}
+}
+func TestAgentRejectsIgnoredFlagsAndUnexpectedArguments(t *testing.T) {
+	for _, argv := range [][]string{{"watch", "-unknown=1"}, {"serve", "-unknown=1"}, {"watch", "unused"}, {"serve", "unused"}, {"configure", "unused"}} {
+		fs := flag.NewFlagSet("agent", flag.ContinueOnError)
+		fs.SetOutput(io.Discard)
+		if _, err := parseAgentArgs(fs, argv); err == nil {
+			t.Fatalf("ignored argv: %v", argv)
+		}
+	}
+}

cmd/evobgp-scheduler/main.go

@@ -41,23 +41,41 @@ func main() {
 
 	apiBase := strings.TrimSpace(os.Getenv("EVOBGP_CONTROL_PLANE_URL"))
 	apiTok := strings.TrimSpace(os.Getenv("EVOBGP_SCHEDULER_BEARER"))
-	if os.Getenv("EVOBGP_SCHEDULER_STANDALONE") == "1" && apiBase == "" {
+	mode := strings.ToLower(strings.TrimSpace(os.Getenv("EVOBGP_SCHEDULER_MODE")))
-		log.Fatal("evobgp-scheduler: EVOBGP_SCHEDULER_STANDALONE=1 requires EVOBGP_CONTROL_PLANE_URL and EVOBGP_SCHEDULER_BEARER (split deploy must not use in-process jobs.Registry)")
+	if mode == "" {
+		switch {
+		case apiBase != "" || apiTok != "":
+			mode = "http"
+		case pool != nil:
+			mode = "db"
+		default:
+			mode = "local"
 		}
-	deps := &scheduler.Deps{
-		Store: st,
-		HTTP:  &http.Client{Timeout: 45 * time.Second},
 	}
-	if (apiBase != "") != (apiTok != "") {
+	deps := &scheduler.Deps{Store: st, HTTP: &http.Client{Timeout: 45 * time.Second}}
-		log.Fatal("evobgp-scheduler: set both EVOBGP_CONTROL_PLANE_URL and EVOBGP_SCHEDULER_BEARER for HTTP mode, or leave both empty for in-process job registry")
+	switch mode {
+	case "db":
+		if pool == nil {
+			log.Fatal("evobgp-scheduler: db mode requires EVOBGP_DATABASE_URL")
 		}
-	if apiBase != "" {
-		deps.APIBase = apiBase
-		deps.APIToken = apiTok
-		log.Printf("evobgp-scheduler: control plane HTTP mode (%s)", apiBase)
-	} else {
 		deps.Jobs = reg
-		log.Printf("evobgp-scheduler: in-process job registry mode (use EVOBGP_CONTROL_PLANE_URL + EVOBGP_SCHEDULER_BEARER for split containers)")
+		log.Print("evobgp-scheduler: trusted DB enqueue mode; API workers consume the shared durable queue")
+	case "http":
+		tenant := strings.TrimSpace(os.Getenv("EVOBGP_SCHEDULER_TENANT_ID"))
+		if apiBase == "" || apiTok == "" || tenant == "" {
+			log.Fatal("evobgp-scheduler: http mode requires EVOBGP_CONTROL_PLANE_URL, EVOBGP_SCHEDULER_BEARER and EVOBGP_SCHEDULER_TENANT_ID")
+		}
+		deps.APIBase, deps.APIToken, deps.APITenantID = apiBase, apiTok, tenant
+		log.Print("evobgp-scheduler: single-tenant HTTP enqueue mode")
+	case "local":
+		if os.Getenv("EVOBGP_SCHEDULER_STANDALONE") == "1" || pool != nil {
+			log.Fatal("evobgp-scheduler: local mode is for in-process development only; use db for a shared PostgreSQL queue")
+		}
+		deps.Jobs = reg
+		defer func() { reg.RequestCancelAll(); _ = reg.Drain(context.Background()) }()
+		log.Print("evobgp-scheduler: local development mode")
+	default:
+		log.Fatal("evobgp-scheduler: EVOBGP_SCHEDULER_MODE must be db, http or local")
 	}
 	scheduler.Run(ctx, deps)
 }

docs/openapi.yaml

@@ -10,6 +10,20 @@ info:
     **Соглашения:** префикс `/v1`; идентификаторы - UUID v7 или ULID (строки); время - ISO 8601 UTC.
     Ошибки - `application/problem+json` ([RFC 9457](https://www.rfc-editor.org/rfc/rfc9457)).
     Пагинация списков - `cursor` + `limit`; ответ содержит `items`, `next_cursor`, `has_more`.
+    **Общий предел request body:** после аутентификации каждый непубличный `/v1/...`
+    запрос проверяется целиком до выполнения handler: максимум 8 MiB (8388608 bytes),
+    независимо от HTTP-метода, Content-Type и наличия Content-Length. Превышение,
+    включая whitespace после JSON, возвращает `413 application/problem+json`
+    до мутаций. Ошибка чтения тела возвращает `400`, отмена/таймаут чтения — `408`.
+    Пустые node GET/pull и streaming SSE responses не меняются. CSV дополнительно
+    сохраняет ограничения числа строк и полей своей операции.
+    Запросы с телом занимают ограниченный слот API instance до окончания handler:
+    по умолчанию 4 одновременно, `EVOBGP_API_BODY_MAX_CONCURRENT` задаёт 1..64.
+    Свободных слотов нет — немедленный `429` с `Retry-After: 1`, без очереди ожидания.
+    Пустые запросы (`http.NoBody`) слота не занимают.
+    Если CRUD-изменение сохранено, но автоматическая refresh/reconcile задача не поставлена,
+    успешный ответ содержит `X-EvoBGP-Job-Status: enqueue-failed`; ошибка также попадает в аудит.
+    Изменение уже сохранено: следует явно запустить refresh/reconcile, а не повторять создание объекта.
 
     **Аутентификация (dual):**
     - **API key** — `Authorization: Bearer <token>` из `EVOBGP_API_KEYS` / таблицы `api_key` (роли `viewer`/`editor`/`operator`/`node`/`firewall`).
@@ -226,6 +240,16 @@ components:
         Основные значения: `module_refresh`, `peer_reconcile`, `deploy_apply`, `revision_rollback`, `bird_reload`.
 
   responses:
+    ExpensiveOperationLimited:
+      description: Локальная квота процесса API — по умолчанию 4 активных preview/lookup, 2 на tenant; 120 lookup и по 30 CDN/revision preview в минуту на tenant. Квоты умножаются на число реплик. Повторить после Retry-After, без очереди ожидания.
+      headers:
+        Retry-After:
+          description: Минимальная задержка повторной попытки в секундах; не гарантия свободного места.
+          schema: {type: integer, minimum: 1}
+      content:
+        application/problem+json:
+          schema:
+            $ref: "#/components/schemas/Problem"
     Unauthorized:
       description: Нет или невалидный Bearer-токен.
       content:
@@ -263,7 +287,10 @@ components:
           schema:
             $ref: "#/components/schemas/Problem"
     DefaultProblem:
-      description: Ошибка (см. тело Problem).
+      description: >
+        Ошибка (см. тело Problem). Общие для authenticated `/v1` ответы middleware:
+        413 при request body свыше 8 MiB, 400 при ошибке чтения тела, 408 при отмене/таймауте чтения.
+        429 с Retry-After: 1 при занятых слотах обработки request body.
       content:
         application/problem+json:
           schema:
@@ -903,6 +930,21 @@ components:
         connected_speaker_label:
           type: string
           description: Человекочитаемая метка ноды из live-опроса.
+        last_known_session_state:
+          type: ["string", "null"]
+          description: Последнее сохранённое состояние; не является текущим наблюдением.
+        session_state:
+          type: ["string", "null"]
+          description: >
+            При live-ответе текущее наблюдение: BGP state, absent после успешного
+            пустого опроса или null, если актуальное состояние неизвестно.
+        live_observation:
+          $ref: "#/components/schemas/LiveObservation"
+          description: >
+            Качество BGP-наблюдений при GET /v1/peers?live=1. Для назначенного
+            пира учитывается только bgp_speaker_id; для общего — все speakers.
+            Смешанный успех и сбой не дают quality=ok или session_state=absent.
+            Timestamps относятся только к ожидаемым speakers.
         session_on_speakers:
           type: array
           description: Состояние протокола пира на каждой опрошенной ноде.
@@ -915,6 +957,10 @@ components:
                 type: string
               state:
                 type: string
+              poll_error:
+                type: string
+              observation:
+                $ref: "#/components/schemas/LiveObservation"
         established_on_speakers:
           type: array
           description: Ноды, где сессия в состоянии Established (один пир может быть на нескольких).
@@ -1001,7 +1047,11 @@ components:
           description: Человекочитаемая метка ноды (agent domain или CP master).
         agent_ok:
           type: boolean
-          description: true если agent /v1/agent/health успешен (master — local birdc poll).
+          description: true только после успешного agent /v1/agent/health (master — local birdc poll).
+        agent_observation:
+          $ref: "#/components/schemas/LiveObservation"
+        bgp_observation:
+          $ref: "#/components/schemas/LiveObservation"
         agent_error:
           type: string
         agent_last_sync_at:
@@ -1015,15 +1065,49 @@ components:
         bgp_poll_error:
           type: string
         bgp_sessions_total:
-          type: integer
+          type: ["integer", "null"]
+          description: null при отсутствии успешного опроса.
         bgp_established:
-          type: integer
+          type: ["integer", "null"]
+          description: null при отсутствии успешного опроса.
         sessions:
           type: array
           items:
             $ref: "#/components/schemas/BgpSessionLive"
       additionalProperties: true
 
+    LiveObservation:
+      type: object
+      description: >
+        Метаданные конкретного опроса. observed_at — время завершения опроса;
+        received_at — время получения результата CP; last_success_at — время
+        последнего успешного опроса. Старые агенты не обязаны передавать эти поля:
+        CP формирует их по собственным часам. При отсутствии опроса timestamps null.
+      required: [observed_at, received_at, last_success_at, source, quality]
+      properties:
+        observed_at:
+          type: ["string", "null"]
+          format: date-time
+        received_at:
+          type: ["string", "null"]
+          format: date-time
+        last_success_at:
+          type: ["string", "null"]
+          format: date-time
+        expires_at:
+          type: ["string", "null"]
+          format: date-time
+          description: Срок актуальности snapshot по часам CP и настроенному интервалу collector; клиент не должен считать cached quality=ok актуальным после этого времени. Для агрегата — самый ранний срок источников.
+        source:
+          type: string
+          enum: [none, cp_birdc, agent_bird_protocols, agent_health, mixed]
+        quality:
+          type: string
+          enum: [ok, stale, unreachable, unknown, disabled]
+        error_code:
+          type: string
+          enum: [not_configured, not_polled, poll_failed, reported_unhealthy, snapshot_expired, configuration_changed, ambiguous_authority]
+
     BgpSessionLive:
       type: object
       properties:
@@ -1052,9 +1136,12 @@ components:
         ok:
           type: boolean
         session_count:
-          type: integer
+          type: ["integer", "null"]
+          description: null, если успешного опроса не было.
         poll_error:
           type: string
+        observation:
+          $ref: "#/components/schemas/LiveObservation"
       additionalProperties: true
 
     PostgresOverview:
@@ -2108,6 +2195,8 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/LookupResponse"
+        "429":
+          $ref: "#/components/responses/ExpensiveOperationLimited"
         "400":
           $ref: "#/components/responses/BadRequest"
         "401":
@@ -2115,6 +2204,29 @@ paths:
         default:
           $ref: "#/components/responses/DefaultProblem"
 
+  /v1/telemetry/refresh:
+    parameters:
+      - $ref: "#/components/parameters/TenantId"
+    post:
+      tags: [Speakers]
+      summary: Запросить фоновое обновление телеметрии tenant
+      description: Требует bgp:network:read. Объединяет повторные запросы в пределах 10 секунд; HTTP handler не опрашивает агентов или BIRD. GET speakers/peers/discoveries читает сохранённые snapshots, включая live=1. Запрос во время выполняющегося poll сохраняется для следующего цикла. Не изменяет конфигурацию speaker и не требует нового протокола агента.
+      operationId: requestTelemetryRefresh
+      responses:
+        "202":
+          description: Запрос принят либо объединён с уже принятым.
+          content:
+            application/json:
+              schema:
+                type: object
+                required: [requested]
+                properties:
+                  requested:
+                    type: boolean
+                    description: true при постановке нового запроса; false при объединении или отсутствии целей.
+        default:
+          $ref: "#/components/responses/DefaultProblem"
+
   /v1/router-lists/catalog:
     get:
       tags: [Modules]
@@ -2789,6 +2901,9 @@ paths:
         - `ipRange,community`
 
         В поле `community` можно передавать либо ID community, либо её значение.
+        Предел запроса — 8 MiB (8388608 байт), 10000 строк данных без заголовка и 64 поля в записи.
+        Превышение возвращает 413 до создания записей; большой импорт следует разбить на части.
+        Ошибка в данных или БД после начала создания записей может оставить частичный импорт.
       operationId: importModuleEntriesCsv
       parameters:
         - $ref: "#/components/parameters/IdempotencyKey"
@@ -2812,6 +2927,12 @@ paths:
                     minimum: 0
                   module_type:
                     type: string
+        "413":
+          description: Превышен лимит байт, строк или полей CSV; записи не созданы.
+          content:
+            application/problem+json:
+              schema:
+                $ref: "#/components/schemas/Problem"
         "404":
           $ref: "#/components/responses/NotFound"
         "422":
@@ -2898,6 +3019,8 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/CdnPreviewResponse"
+        "429":
+          $ref: "#/components/responses/ExpensiveOperationLimited"
         "401":
           $ref: "#/components/responses/Unauthorized"
         "403":
@@ -3146,7 +3269,7 @@ paths:
             type: string
             enum: ["1"]
           description: >
-            Опрос birdc на CP и GET /v1/agent/bird/protocols на репликах; обогащает session_state и connected_speaker_*.
+            Сохранённые результаты фонового опроса birdc на CP и agent protocols на репликах; обогащает session_state и connected_speaker_*. GET не запускает опрос; live=1 сохранён для совместимости.
       responses:
         "200":
           description: Успешно.
@@ -3167,7 +3290,7 @@ paths:
                   live_speaker_poll:
                     type: array
                     description: >
-                      При live=1 — результат опроса каждой ноды (CP birdc + agent protocols).
+                      Сохранённый результат фонового опроса каждой ноды (CP birdc + agent protocols); live=1 не запускает новый опрос.
                     items:
                       $ref: "#/components/schemas/LiveSpeakerPoll"
         default:
@@ -3204,7 +3327,7 @@ paths:
       summary: Список обнаруженных (неодобренных) пиров
       description: >
         Dynamic BGP-сессии (`evobgp_dyn_*`), которых ещё нет в `bgp_peer`.
-        По умолчанию возвращает `status=pending`. При листинге выполняет live-опрос birdc/agent и upsert pending.
+        По умолчанию возвращает `status=pending`. Читает сохранённые discoveries. Фоновый collector выполняет опрос и fenced upsert; GET не опрашивает ноды и не изменяет данные.
       operationId: listDiscoveredPeers
       parameters:
         - $ref: "#/components/parameters/TenantId"
@@ -3380,7 +3503,7 @@ paths:
             type: string
             enum: ["1"]
           description: >
-            Live-опрос agent /v1/agent/health и BGP protocols на репликах; CP — local birdc.
+            Сохранённые результаты фоновых опросов agent health/BGP и local birdc. GET не запускает опрос.
             Обогащает каждый item полем `live`.
       responses:
         "200":
@@ -3640,6 +3763,8 @@ paths:
       description: Только чтение, без apply.
       operationId: getRevisionPreview
       responses:
+        "429":
+          $ref: "#/components/responses/ExpensiveOperationLimited"
         "200":
           description: Текст или структурированное представление - формат задаётся реализацией.
           content:
@@ -3881,6 +4006,58 @@ paths:
         default:
           $ref: "#/components/responses/DefaultProblem"
 
+  /v1/jobs/{job_id}/report:
+    parameters:
+      - $ref: "#/components/parameters/TenantId"
+      - $ref: "#/components/parameters/JobId"
+    get:
+      tags: [Jobs]
+      summary: Отчёт задачи и сохранённый прогресс доставки
+      description: Требует bgp:operations:read. Прогресс хранится независимо от meta задачи. Для reconciliation без собственного прогресса возвращается прогресс исходной задачи в том же tenant; после начала восстановления приоритет имеет собственный прогресс. agent_reported не означает проверку состояния BIRD.
+      operationId: getJobReport
+      responses:
+        "200":
+          description: Успешно.
+          content:
+            application/json:
+              schema:
+                type: object
+                required: [job_id, kind, status, deployment_progress]
+                properties:
+                  job_id: {type: string}
+                  kind: {type: string}
+                  status: {type: string}
+                  meta: {type: [object, "null"], additionalProperties: true}
+                  error: {type: [string, "null"]}
+                  created_at: {type: string, format: date-time}
+                  revision_id: {type: string}
+                  log_entries: {type: array, items: {type: object, additionalProperties: true}}
+                  log_total: {type: integer}
+                  source_deployment_job_id: {type: string}
+                  deployment_progress:
+                    type: array
+                    items:
+                      type: object
+                      required: [job_id, speaker_id, revision_id, generation, state, confirmation, dispatch_attempts, retry_round]
+                      properties:
+                        job_id: {type: string}
+                        speaker_id: {type: string}
+                        revision_id: {type: string}
+                        generation: {type: integer, format: int64}
+                        state: {type: string}
+                        confirmation: {type: string}
+                        error: {type: string}
+                        dispatch_attempts: {type: integer}
+                        retry_round: {type: integer}
+                        health_received_at: {type: [string, "null"], format: date-time}
+                        health_revision_id: {type: string}
+                        health_source_at: {type: string}
+                        health_error: {type: string}
+        "404":
+          $ref: "#/components/responses/NotFound"
+        default:
+          $ref: "#/components/responses/DefaultProblem"
+
   /v1/jobs/{job_id}/cancel:
     parameters:
       - $ref: "#/components/parameters/TenantId"
@@ -3904,6 +4081,33 @@ paths:
         default:
           $ref: "#/components/responses/DefaultProblem"
 
+  /v1/jobs/{job_id}/retry:
+    parameters:
+      - $ref: "#/components/parameters/TenantId"
+      - $ref: "#/components/parameters/JobId"
+    post:
+      tags: [Jobs]
+      summary: Повторить завершившуюся ошибкой задачу
+      description: Операторский повтор сохраняет ID и исходный запрос, сбрасывая бюджет попыток.
+      operationId: postJobRetry
+      responses:
+        "202":
+          description: Повтор поставлен в очередь.
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/Job"
+        "409":
+          $ref: "#/components/responses/Conflict"
+        "429":
+          description: Очередь задач заполнена.
+          content:
+            application/problem+json:
+              schema:
+                $ref: "#/components/schemas/Problem"
+        default:
+          $ref: "#/components/responses/DefaultProblem"
+
   /v1/speakers/{speaker_id}/revisions/latest:
     parameters:
       - $ref: "#/components/parameters/SpeakerId"

internal/agentwatch/snapshot.go

@@ -0,0 +1,312 @@
+package agentwatch
+
+import (
+	"context"
+	"crypto/sha256"
+	"encoding/binary"
+	"errors"
+	"fmt"
+	"hash"
+	"io"
+	"os"
+	"path/filepath"
+	"regexp"
+	"sort"
+	"strings"
+)
+
+const (
+	maxConfigFiles      = 512
+	maxConfigBytes      = 64 << 20
+	maxFileBytes        = 16 << 20
+	maxIncludeDepth     = 16
+	maxDirectoryEntries = 4096
+)
+
+var errMainNotReady = errors.New("main bird.conf is not ready")
+
+// Only standalone literal include statements are hashable. Other valid BIRD
+// grammar falls back to the legacy configure-on-every-tick path.
+var includeLine = regexp.MustCompile(`(?m)^[\t ]*include[\t ]+"([^"\r\n]+)"[\t ]*;[\t ]*(?:#[^\r\n]*)?\r?$`)
+
+type configScanner struct {
+	ctx   context.Context
+	root  string
+	h     hash.Hash
+	seen  map[string]bool
+	files int
+	bytes int64
+}
+
+func hashConfigTree(ctx context.Context, activeDir string) ([sha256.Size]byte, error) {
+	var zero [sha256.Size]byte
+	root, err := filepath.Abs(activeDir)
+	if err != nil {
+		return zero, fmt.Errorf("agentwatch: active directory: %w", err)
+	}
+	root, err = filepath.EvalSymlinks(root)
+	if err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			return zero, fmt.Errorf("agentwatch: %w: %v", errMainNotReady, err)
+		}
+		return zero, fmt.Errorf("agentwatch: resolve active directory: %w", err)
+	}
+	s := &configScanner{ctx: ctx, root: filepath.Clean(root), h: sha256.New(), seen: make(map[string]bool)}
+	if err := s.visit(filepath.Join(s.root, "bird.conf"), 0); err != nil {
+		return zero, err
+	}
+	var digest [sha256.Size]byte
+	copy(digest[:], s.h.Sum(nil))
+	return digest, nil
+}
+
+func (s *configScanner) visit(path string, depth int) error {
+	if err := s.ctx.Err(); err != nil {
+		return err
+	}
+	if depth > maxIncludeDepth {
+		return fmt.Errorf("agentwatch: include depth exceeds %d", maxIncludeDepth)
+	}
+	path = filepath.Clean(path)
+	rel, err := filepath.Rel(s.root, path)
+	if err != nil || rel == ".." || strings.HasPrefix(rel, ".."+string(os.PathSeparator)) {
+		return fmt.Errorf("agentwatch: include %q escapes active directory", path)
+	}
+	if s.seen[path] {
+		return nil
+	}
+	info, err := os.Lstat(path)
+	if err != nil {
+		if depth == 0 && errors.Is(err, os.ErrNotExist) {
+			return fmt.Errorf("agentwatch: %w: %v", errMainNotReady, err)
+		}
+		return fmt.Errorf("agentwatch: inspect %q: %w", path, err)
+	}
+	if !info.Mode().IsRegular() {
+		return fmt.Errorf("agentwatch: include %q is not a regular file", path)
+	}
+	resolved, err := filepath.EvalSymlinks(path)
+	if err != nil {
+		return fmt.Errorf("agentwatch: resolve %q: %w", path, err)
+	}
+	if _, err := s.withinRoot(resolved); err != nil {
+		return err
+	}
+	if info.Size() > maxFileBytes {
+		return fmt.Errorf("agentwatch: include %q exceeds %d bytes", path, maxFileBytes)
+	}
+	if s.files >= maxConfigFiles || s.bytes+info.Size() > maxConfigBytes {
+		return fmt.Errorf("agentwatch: configuration tree exceeds %d files or %d bytes", maxConfigFiles, maxConfigBytes)
+	}
+	f, err := os.Open(path)
+	if err != nil {
+		return fmt.Errorf("agentwatch: open %q: %w", path, err)
+	}
+	defer func() { _ = f.Close() }()
+	limit := int64(maxFileBytes)
+	if remaining := int64(maxConfigBytes) - s.bytes; remaining < limit {
+		limit = remaining
+	}
+	content, err := readBounded(f, limit)
+	if err != nil {
+		return fmt.Errorf("agentwatch: read %q: %w", path, err)
+	}
+	s.seen[path] = true
+	s.files++
+	s.bytes += int64(len(content))
+	writeHash(s.h, filepath.ToSlash(rel), content)
+	patterns, err := includePatterns(string(content))
+	if err != nil {
+		return fmt.Errorf("agentwatch: %s: %w", rel, err)
+	}
+	for _, pattern := range patterns {
+		if strings.ContainsAny(pattern, `[\`) {
+			return fmt.Errorf("agentwatch: unsupported escaped include %q", pattern)
+		}
+		candidate := pattern
+		if !filepath.IsAbs(candidate) {
+			candidate = filepath.Join(filepath.Dir(path), filepath.FromSlash(candidate))
+		}
+		candidate = filepath.Clean(candidate)
+		// The pattern itself changes the digest even when its glob currently matches nothing.
+		writeHash(s.h, "include", []byte(filepath.ToSlash(candidate)))
+		if _, err := s.withinRoot(candidate); err != nil {
+			return err
+		}
+		if strings.ContainsAny(candidate, "*?") {
+			if strings.ContainsAny(filepath.Dir(candidate), "*?") {
+				return fmt.Errorf("agentwatch: unsupported include directory glob %q", pattern)
+			}
+			resolvedDir, err := filepath.EvalSymlinks(filepath.Dir(candidate))
+			if err != nil {
+				return fmt.Errorf("agentwatch: resolve include directory %q: %w", pattern, err)
+			}
+			if _, err := s.withinRoot(resolvedDir); err != nil {
+				return err
+			}
+			matches, err := boundedGlob(candidate)
+			if err != nil {
+				return fmt.Errorf("agentwatch: include glob %q: %w", pattern, err)
+			}
+			for _, match := range matches {
+				if err := s.visit(match, depth+1); err != nil {
+					return err
+				}
+			}
+			continue
+		}
+		if err := s.visit(candidate, depth+1); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func readBounded(r io.Reader, limit int64) ([]byte, error) {
+	content, err := io.ReadAll(io.LimitReader(r, limit+1))
+	if err != nil {
+		return nil, err
+	}
+	if int64(len(content)) > limit {
+		return nil, fmt.Errorf("configuration tree changed or exceeded size limit")
+	}
+	return content, nil
+}
+
+// boundedGlob only handles * and ? in a filename. Broader BIRD glob grammar
+// falls back to every-tick configure instead of risking a false unchanged hash.
+func boundedGlob(pattern string) ([]string, error) {
+	dir, base := filepath.Dir(pattern), filepath.Base(pattern)
+	f, err := os.Open(dir)
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = f.Close() }()
+	names, err := f.Readdirnames(maxDirectoryEntries + 1)
+	if err != nil && !errors.Is(err, io.EOF) {
+		return nil, err
+	}
+	if len(names) > maxDirectoryEntries {
+		return nil, fmt.Errorf("directory exceeds %d entries", maxDirectoryEntries)
+	}
+	matches := make([]string, 0)
+	for _, name := range names {
+		match, err := filepath.Match(base, name)
+		if err != nil {
+			return nil, err
+		}
+		if match {
+			matches = append(matches, filepath.Join(dir, name))
+			if len(matches) > maxConfigFiles {
+				return nil, fmt.Errorf("glob exceeds %d matches", maxConfigFiles)
+			}
+		}
+	}
+	sort.Strings(matches)
+	return matches, nil
+}
+
+func (s *configScanner) withinRoot(path string) (string, error) {
+	rel, err := filepath.Rel(s.root, path)
+	if err != nil || rel == ".." || strings.HasPrefix(rel, ".."+string(os.PathSeparator)) {
+		return "", fmt.Errorf("agentwatch: include %q escapes active directory", path)
+	}
+	return rel, nil
+}
+
+func writeHash(h hash.Hash, name string, content []byte) {
+	var size [8]byte
+	binary.LittleEndian.PutUint64(size[:], uint64(len(name)))
+	_, _ = h.Write(size[:])
+	_, _ = h.Write([]byte(name))
+	binary.LittleEndian.PutUint64(size[:], uint64(len(content)))
+	_, _ = h.Write(size[:])
+	_, _ = h.Write(content)
+}
+
+func includePatterns(content string) ([]string, error) {
+	tokens, unsupportedQuotedText := includeTokenOffsets(content)
+	if unsupportedQuotedText {
+		return nil, fmt.Errorf("unsupported quoted text or escape")
+	}
+	tokenSet := make(map[int]bool, len(tokens))
+	for _, token := range tokens {
+		tokenSet[token] = true
+	}
+	var patterns []string
+	for _, match := range includeLine.FindAllStringSubmatchIndex(content, -1) {
+		keyword := match[0] + strings.Index(content[match[0]:match[1]], "include")
+		if !tokenSet[keyword] {
+			continue
+		} // text inside a block comment
+		delete(tokenSet, keyword)
+		patterns = append(patterns, content[match[2]:match[3]])
+	}
+	if len(tokenSet) != 0 {
+		return nil, fmt.Errorf("unsupported include syntax")
+	}
+	return patterns, nil
+}
+
+// Recognize include keywords outside BIRD comments and quoted strings.
+func includeTokenOffsets(config string) ([]int, bool) {
+	var offsets []int
+	quote := byte(0)
+	lineComment, blockComment := false, false
+	for i := 0; i < len(config); {
+		c := config[i]
+		if lineComment {
+			if c == '\n' {
+				lineComment = false
+			}
+			i++
+			continue
+		}
+		if blockComment {
+			if c == '*' && i+1 < len(config) && config[i+1] == '/' {
+				blockComment = false
+				i += 2
+			} else {
+				i++
+			}
+			continue
+		}
+		if quote != 0 {
+			if c == '\\' {
+				return offsets, true
+			}
+			if c == quote {
+				quote = 0
+			}
+			i++
+			continue
+		}
+		if c == '#' || (c == '/' && i+1 < len(config) && config[i+1] == '/') {
+			lineComment = true
+			i++
+			continue
+		}
+		if c == '/' && i+1 < len(config) && config[i+1] == '*' {
+			blockComment = true
+			i += 2
+			continue
+		}
+		if c == '"' || c == '\'' {
+			quote = c
+			i++
+			continue
+		}
+		if (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') || c == '_' {
+			start := i
+			for i < len(config) && ((config[i] >= 'a' && config[i] <= 'z') || (config[i] >= 'A' && config[i] <= 'Z') || (config[i] >= '0' && config[i] <= '9') || config[i] == '_') {
+				i++
+			}
+			if config[start:i] == "include" {
+				offsets = append(offsets, start)
+			}
+			continue
+		}
+		i++
+	}
+	return offsets, quote != 0 || blockComment
+}

internal/agentwatch/watch.go

@@ -0,0 +1,109 @@
+// Package agentwatch reloads BIRD when its active configuration tree changes.
+package agentwatch
+
+import (
+	"context"
+	"crypto/sha256"
+	"errors"
+	"fmt"
+	"log"
+	"path/filepath"
+	"time"
+)
+
+// Control is the subset of BIRD operations required by the watch loop.
+type Control interface {
+	ParseCheck(context.Context, string) error
+	Configure(context.Context) error
+}
+
+// Config sets bounded polling and reconciliation behavior.
+type Config struct {
+	ActiveDir      string
+	Every          time.Duration
+	ReconcileEvery time.Duration
+	CommandTimeout time.Duration
+	Logf           func(string, ...any)
+}
+
+type watcher struct {
+	ctl            Control
+	cfg            Config
+	hash           [sha256.Size]byte
+	haveHash       bool
+	lastConfigured time.Time
+}
+
+// Run starts with one check, then polls until ctx is cancelled. A successful
+// birdc exit advances the remembered hash; it is not proof of BIRD ACK.
+func Run(ctx context.Context, ctl Control, cfg Config) error {
+	if ctl == nil || cfg.Every <= 0 || cfg.ReconcileEvery <= 0 || cfg.CommandTimeout <= 0 {
+		return fmt.Errorf("agentwatch: control and positive intervals/timeouts are required")
+	}
+	if cfg.ActiveDir == "" {
+		cfg.ActiveDir = "/etc/bird"
+	}
+	if cfg.Logf == nil {
+		cfg.Logf = log.Printf
+	}
+	w := &watcher{ctl: ctl, cfg: cfg}
+	if err := w.step(ctx, time.Now()); err != nil && ctx.Err() == nil {
+		cfg.Logf("evobgp-agent watch: %v", err)
+	}
+	ticker := time.NewTicker(cfg.Every)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			return nil
+		case now := <-ticker.C:
+			if err := w.step(ctx, now); err != nil && ctx.Err() == nil {
+				cfg.Logf("evobgp-agent watch: %v", err)
+			}
+		}
+	}
+}
+
+func (w *watcher) step(ctx context.Context, now time.Time) error {
+	if err := ctx.Err(); err != nil {
+		return err
+	}
+	digest, scanErr := hashConfigTree(ctx, w.cfg.ActiveDir)
+	if scanErr != nil {
+		if errors.Is(scanErr, errMainNotReady) {
+			return fmt.Errorf("configuration not ready: %w", scanErr)
+		}
+		// Unsupported/unhashable trees retain the old every-tick configure cadence.
+		w.cfg.Logf("evobgp-agent watch: cannot hash complete config tree (%v); using every-tick parse/configure", scanErr)
+	} else if w.haveHash && digest == w.hash && now.Sub(w.lastConfigured) < w.cfg.ReconcileEvery {
+		return nil
+	}
+	main := filepath.Join(w.cfg.ActiveDir, "bird.conf")
+	if err := w.command(ctx, func(cctx context.Context) error { return w.ctl.ParseCheck(cctx, main) }); err != nil {
+		return fmt.Errorf("parse check: %w", err)
+	}
+	if scanErr == nil {
+		stable, err := hashConfigTree(ctx, w.cfg.ActiveDir)
+		if err != nil {
+			return fmt.Errorf("configuration changed during parse check: %w", err)
+		}
+		if stable != digest {
+			return fmt.Errorf("configuration changed during parse check; retrying")
+		}
+	}
+	if err := w.command(ctx, w.ctl.Configure); err != nil {
+		return fmt.Errorf("configure: %w", err)
+	}
+	w.haveHash = scanErr == nil
+	if w.haveHash {
+		w.hash = digest
+	}
+	w.lastConfigured = now
+	return nil
+}
+
+func (w *watcher) command(parent context.Context, run func(context.Context) error) error {
+	ctx, cancel := context.WithTimeout(parent, w.cfg.CommandTimeout)
+	defer cancel()
+	return run(ctx)
+}

internal/agentwatch/watch_test.go

@@ -0,0 +1,311 @@
+package agentwatch
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+)
+
+type fakeControl struct {
+	parsed        int
+	configured    int
+	failConfigure int
+	onParse       func()
+}
+
+func (f *fakeControl) ParseCheck(ctx context.Context, _ string) error {
+	if err := ctx.Err(); err != nil {
+		return err
+	}
+	f.parsed++
+	if f.onParse != nil {
+		f.onParse()
+	}
+	return nil
+}
+
+func (f *fakeControl) Configure(ctx context.Context) error {
+	if err := ctx.Err(); err != nil {
+		return err
+	}
+	f.configured++
+	if f.failConfigure > 0 {
+		f.failConfigure--
+		return errors.New("birdc unavailable")
+	}
+	return nil
+}
+
+func testConfig(t *testing.T) (string, string) {
+	t.Helper()
+	root := t.TempDir()
+	if err := os.Mkdir(filepath.Join(root, "bird.d"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	main := filepath.Join(root, "bird.conf")
+	if err := os.WriteFile(main, []byte("router id 192.0.2.1;\ninclude \"bird.d/*.conf\";\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	fragment := filepath.Join(root, "bird.d", "peer.conf")
+	if err := os.WriteFile(fragment, []byte("# peer v1\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	return root, fragment
+}
+
+func testWatcher(root string, ctl Control) *watcher {
+	return &watcher{ctl: ctl, cfg: Config{ActiveDir: root, Every: 30 * time.Second, ReconcileEvery: 5 * time.Minute, CommandTimeout: time.Second, Logf: func(string, ...any) {}}}
+}
+
+func TestHashConfigTreeTracksMainIncludesAndGlobMembership(t *testing.T) {
+	root, fragment := testConfig(t)
+	first, err := hashConfigTree(context.Background(), root)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(fragment, []byte("# peer v2\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	second, err := hashConfigTree(context.Background(), root)
+	if err != nil || second == first {
+		t.Fatalf("included edit not detected: digest=%x err=%v", second, err)
+	}
+	if err := os.WriteFile(filepath.Join(root, "bird.d", "manual.conf"), []byte("# manual\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	third, err := hashConfigTree(context.Background(), root)
+	if err != nil || third == second {
+		t.Fatalf("new glob member not detected: digest=%x err=%v", third, err)
+	}
+	if err := os.WriteFile(filepath.Join(root, "bird.conf"), []byte("router id 192.0.2.2;\ninclude \"bird.d/*.conf\";\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	fourth, err := hashConfigTree(context.Background(), root)
+	if err != nil || fourth == third {
+		t.Fatalf("main edit not detected: digest=%x err=%v", fourth, err)
+	}
+}
+
+func TestWatchSkipsUnchangedRetriesErrorsAndForcesReconciliation(t *testing.T) {
+	root, fragment := testConfig(t)
+	ctl := &fakeControl{}
+	w := testWatcher(root, ctl)
+	now := time.Now()
+	if err := w.step(context.Background(), now); err != nil {
+		t.Fatal(err)
+	}
+	if err := w.step(context.Background(), now.Add(30*time.Second)); err != nil {
+		t.Fatal(err)
+	}
+	if ctl.configured != 1 || ctl.parsed != 1 {
+		t.Fatalf("unchanged tree configured again: %+v", ctl)
+	}
+	if err := os.WriteFile(fragment, []byte("# changed\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	ctl.failConfigure = 1
+	if err := w.step(context.Background(), now.Add(time.Minute)); err == nil {
+		t.Fatal("expected configure failure")
+	}
+	if !w.haveHash || ctl.configured != 2 {
+		t.Fatalf("failure changed remembered state: %+v", w)
+	}
+	if err := w.step(context.Background(), now.Add(90*time.Second)); err != nil {
+		t.Fatal(err)
+	}
+	if ctl.configured != 3 {
+		t.Fatalf("failed change was not retried: %+v", ctl)
+	}
+	if err := w.step(context.Background(), now.Add(7*time.Minute)); err != nil {
+		t.Fatal(err)
+	}
+	if ctl.configured != 4 {
+		t.Fatalf("forced reconciliation missing: %+v", ctl)
+	}
+	// A fresh watcher must reconcile once even when another process already loaded the same bytes.
+	restarted := testWatcher(root, ctl)
+	if err := restarted.step(context.Background(), now.Add(8*time.Minute)); err != nil {
+		t.Fatal(err)
+	}
+	if ctl.configured != 5 {
+		t.Fatalf("restart did not reconcile: %+v", ctl)
+	}
+}
+
+func TestWatchRejectsChangingTreeBeforeConfigure(t *testing.T) {
+	root, fragment := testConfig(t)
+	ctl := &fakeControl{onParse: func() { _ = os.WriteFile(fragment, []byte("# raced\n"), 0o644) }}
+	w := testWatcher(root, ctl)
+	if err := w.step(context.Background(), time.Now()); err == nil || !strings.Contains(err.Error(), "changed during parse check") {
+		t.Fatalf("want race retry, got %v", err)
+	}
+	if ctl.configured != 0 {
+		t.Fatal("configured a tree that changed after parse check")
+	}
+	ctl.onParse = nil
+	if err := w.step(context.Background(), time.Now()); err != nil {
+		t.Fatal(err)
+	}
+	if ctl.configured != 1 {
+		t.Fatal("stable tree was not configured")
+	}
+}
+
+func TestUnsupportedIncludeUsesLegacyEveryTickCadence(t *testing.T) {
+	root, _ := testConfig(t)
+	if err := os.WriteFile(filepath.Join(root, "bird.conf"), []byte("include variable;\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	ctl := &fakeControl{}
+	w := testWatcher(root, ctl)
+	now := time.Now()
+	for i := 0; i < 3; i++ {
+		if err := w.step(context.Background(), now.Add(time.Duration(i)*30*time.Second)); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if ctl.parsed != 3 || ctl.configured != 3 || w.haveHash {
+		t.Fatalf("unsupported tree was skipped: %+v", ctl)
+	}
+}
+
+func TestMissingIncludedFileFallsBackButMissingMainWaits(t *testing.T) {
+	root, fragment := testConfig(t)
+	if err := os.WriteFile(filepath.Join(root, "bird.conf"), []byte("include \"bird.d/peer.conf\";\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Remove(fragment); err != nil {
+		t.Fatal(err)
+	}
+	ctl := &fakeControl{}
+	w := testWatcher(root, ctl)
+	now := time.Now()
+	for i := 0; i < 2; i++ {
+		if err := w.step(context.Background(), now.Add(time.Duration(i)*30*time.Second)); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if ctl.configured != 2 {
+		t.Fatalf("missing include skipped legacy cadence: %+v", ctl)
+	}
+	if err := os.Remove(filepath.Join(root, "bird.conf")); err != nil {
+		t.Fatal(err)
+	}
+	if err := w.step(context.Background(), now.Add(time.Minute)); !errors.Is(err, errMainNotReady) {
+		t.Fatalf("missing main should wait, got %v", err)
+	}
+	if ctl.configured != 2 {
+		t.Fatal("configured without main bird.conf")
+	}
+}
+
+func TestHashConfigTreeBoundsAndEscapes(t *testing.T) {
+	root, fragment := testConfig(t)
+	if err := os.WriteFile(fragment, make([]byte, maxFileBytes+1), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := hashConfigTree(context.Background(), root); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("want size limit, got %v", err)
+	}
+	if err := os.WriteFile(fragment, []byte("# small\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	outside := filepath.Join(t.TempDir(), "outside.conf")
+	if err := os.WriteFile(outside, []byte("# outside\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(root, "bird.conf"), []byte("include \""+filepath.ToSlash(outside)+"\";\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := hashConfigTree(context.Background(), root); err == nil || !strings.Contains(err.Error(), "escapes active directory") {
+		t.Fatalf("want outside include rejection, got %v", err)
+	}
+}
+
+func TestHashConfigTreeRejectsSymlinkInclude(t *testing.T) {
+	root, fragment := testConfig(t)
+	if err := os.Remove(fragment); err != nil {
+		t.Fatal(err)
+	}
+	outside := filepath.Join(t.TempDir(), "outside.conf")
+	if err := os.WriteFile(outside, []byte("# outside\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Symlink(outside, fragment); err != nil {
+		t.Skipf("symlinks unavailable: %v", err)
+	}
+	if _, err := hashConfigTree(context.Background(), root); err == nil || !strings.Contains(err.Error(), "not a regular file") {
+		t.Fatalf("want symlink rejection, got %v", err)
+	}
+}
+
+func TestIncludeCommentsAndUnsupportedForms(t *testing.T) {
+	patterns, err := includePatterns("# include \"ignored.conf\";\n/*\ninclude \"also-ignored.conf\";\n*/\ninclude \"bird.d/*.conf\";\n")
+	if err != nil || len(patterns) != 1 || patterns[0] != "bird.d/*.conf" {
+		t.Fatalf("comment handling failed: patterns=%v err=%v", patterns, err)
+	}
+	if _, err := includePatterns("if true then include \"dynamic.conf\";\n"); err == nil {
+		t.Fatal("inline include should fall back to legacy cadence")
+	}
+	if _, err := includePatterns("include \"bird.d/[!a].conf\";\n"); err != nil {
+		// Bracket glob is rejected by the scanner after lexical parsing.
+		t.Fatalf("literal include statement should parse before glob validation: %v", err)
+	}
+	root, _ := testConfig(t)
+	if err := os.WriteFile(filepath.Join(root, "bird.conf"), []byte("include \"bird.d/[!a].conf\";\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := hashConfigTree(context.Background(), root); err == nil || !strings.Contains(err.Error(), "unsupported") {
+		t.Fatalf("bracket glob should fall back, got %v", err)
+	}
+	if _, err := includePatterns("define note = \"single\\\";\ninclude \"bird.d/real.conf\";\n"); err == nil {
+		t.Fatal("backslash in quoted text must force conservative fallback")
+	}
+}
+
+func TestBoundedGlobLimitsMatches(t *testing.T) {
+	root := t.TempDir()
+	for i := 0; i <= maxConfigFiles; i++ {
+		name := filepath.Join(root, fmt.Sprintf("peer-%03d.conf", i))
+		if err := os.WriteFile(name, nil, 0o644); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if _, err := boundedGlob(filepath.Join(root, "*.conf")); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("want bounded glob rejection, got %v", err)
+	}
+}
+
+func TestReadBoundedStopsAfterLimit(t *testing.T) {
+	r := strings.NewReader("abcdefghij")
+	if _, err := readBounded(r, 3); err == nil {
+		t.Fatal("growing file exceeded limit")
+	}
+	if got := r.Len(); got != 6 {
+		t.Fatalf("reader consumed beyond limit+1: %d bytes remain", got)
+	}
+}
+
+func TestRunStopsOnCancellation(t *testing.T) {
+	root, _ := testConfig(t)
+	ctx, cancel := context.WithCancel(context.Background())
+	ctl := &fakeControl{}
+	done := make(chan error, 1)
+	go func() {
+		done <- Run(ctx, ctl, Config{ActiveDir: root, Every: time.Hour, ReconcileEvery: 5 * time.Minute, CommandTimeout: time.Second, Logf: func(string, ...any) {}})
+	}()
+	cancel()
+	select {
+	case err := <-done:
+		if err != nil {
+			t.Fatal(err)
+		}
+	case <-time.After(2 * time.Second):
+		t.Fatal("watch did not stop after cancellation")
+	}
+}

internal/asnresolve/ripestat.go

@@ -23,6 +23,8 @@ const DefaultASOverviewURL = "https://stat.ripe.net/data/as-overview/data.json"
 
 // AnnouncedPrefixes returns currently announced IPv4/IPv6 prefixes for the ASN (best-effort via RIPEstat).
 func AnnouncedPrefixes(ctx context.Context, hc *http.Client, asn int64) ([]netip.Prefix, error) {
+	ctx, cancel := ripeRequestContext(ctx)
+	defer cancel()
 	if hc == nil {
 		hc = httpclient.New(httpclient.DefaultTimeout)
 	}
@@ -43,13 +45,14 @@ func AnnouncedPrefixes(ctx context.Context, hc *http.Client, asn int64) ([]netip
 		return nil, fmt.Errorf("ripestat fetch AS%d: %w", asn, err)
 	}
 	defer func() { _ = resp.Body.Close() }()
-	body, err := io.ReadAll(io.LimitReader(resp.Body, 32<<20))
-	if err != nil {
-		return nil, err
-	}
 	if resp.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(io.LimitReader(resp.Body, 4<<10))
 		return nil, fmt.Errorf("ripestat AS%d: HTTP %s: %s", asn, resp.Status, truncateForErr(body, 200))
 	}
+	body, err := httpclient.ReadResponseBody(resp, 32<<20)
+	if err != nil {
+		return nil, fmt.Errorf("ripestat AS%d response: %w", asn, err)
+	}
 
 	var wrap struct {
 		Status string `json:"status"`
@@ -85,6 +88,8 @@ func AnnouncedPrefixes(ctx context.Context, hc *http.Client, asn int64) ([]netip
 
 // ASHolderName returns the holder / organization label for the ASN from RIPEstat as-overview (best-effort).
 func ASHolderName(ctx context.Context, hc *http.Client, asn int64) (string, error) {
+	ctx, cancel := ripeRequestContext(ctx)
+	defer cancel()
 	if hc == nil {
 		hc = httpclient.New(httpclient.DefaultTimeout)
 	}
@@ -105,13 +110,14 @@ func ASHolderName(ctx context.Context, hc *http.Client, asn int64) (string, erro
 		return "", fmt.Errorf("ripestat as-overview AS%d: %w", asn, err)
 	}
 	defer func() { _ = resp.Body.Close() }()
-	body, err := io.ReadAll(io.LimitReader(resp.Body, 4<<20))
-	if err != nil {
-		return "", err
-	}
 	if resp.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(io.LimitReader(resp.Body, 4<<10))
 		return "", fmt.Errorf("ripestat as-overview AS%d: HTTP %s: %s", asn, resp.Status, truncateForErr(body, 200))
 	}
+	body, err := httpclient.ReadResponseBody(resp, 4<<20)
+	if err != nil {
+		return "", fmt.Errorf("ripestat as-overview AS%d response: %w", asn, err)
+	}
 
 	var wrap struct {
 		Status string `json:"status"`
@@ -135,3 +141,10 @@ func truncateForErr(b []byte, n int) string {
 	}
 	return s
 }
+
+func ripeRequestContext(ctx context.Context) (context.Context, context.CancelFunc) {
+	if _, ok := ctx.Deadline(); ok {
+		return context.WithCancel(ctx)
+	}
+	return context.WithTimeout(ctx, httpclient.DefaultTimeout)
+}

internal/asnresolve/ripestat_limits_test.go

@@ -0,0 +1,103 @@
+package asnresolve
+
+import (
+	"compress/gzip"
+	"context"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"evobgp/internal/httpclient"
+)
+
+type ripeRoundTrip func(*http.Request) (*http.Response, error)
+
+func (f ripeRoundTrip) RoundTrip(r *http.Request) (*http.Response, error) { return f(r) }
+
+type ripeCountingBody struct {
+	io.Reader
+	read   int
+	closed bool
+}
+
+func (b *ripeCountingBody) Read(p []byte) (int, error) {
+	n, e := b.Reader.Read(p)
+	b.read += n
+	return n, e
+}
+func (b *ripeCountingBody) Close() error { b.closed = true; return nil }
+
+func TestRIPECompleteResponseLimits(t *testing.T) {
+	cases := []struct {
+		name   string
+		limit  int
+		prefix string
+		call   func(context.Context, *http.Client) error
+	}{
+		{"announced", 32 << 20, `{"status":"ok","data":{"prefixes":[]}}`, func(ctx context.Context, hc *http.Client) error { _, e := AnnouncedPrefixes(ctx, hc, 64496); return e }},
+		{"holder", 4 << 20, `{"status":"ok","data":{"holder":"Example"}}`, func(ctx context.Context, hc *http.Client) error { _, e := ASHolderName(ctx, hc, 64496); return e }},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			for _, known := range []bool{true, false} {
+				for _, extra := range []int{0, 1} {
+					body := &ripeCountingBody{Reader: strings.NewReader(tc.prefix + strings.Repeat(" ", tc.limit-len(tc.prefix)+extra))}
+					hc := &http.Client{Transport: ripeRoundTrip(func(r *http.Request) (*http.Response, error) {
+						if _, ok := r.Context().Deadline(); !ok {
+							t.Error("missing request deadline")
+						}
+						length := int64(-1)
+						if known {
+							length = int64(tc.limit + extra)
+						}
+						return &http.Response{StatusCode: 200, Body: body, ContentLength: length, Header: make(http.Header)}, nil
+					})}
+					err := tc.call(context.Background(), hc)
+					if extra == 0 && err != nil {
+						t.Fatalf("exact limit rejected known=%v: %v", known, err)
+					}
+					if extra == 1 && !errors.Is(err, httpclient.ErrResponseTooLarge) {
+						t.Fatalf("oversized valid prefix accepted known=%v: %v", known, err)
+					}
+					if body.read > tc.limit+1 || !body.closed {
+						t.Fatalf("read=%d closed=%v", body.read, body.closed)
+					}
+				}
+			}
+		})
+	}
+}
+
+func TestRIPEErrorBodySmallAndContextPreserved(t *testing.T) {
+	body := &ripeCountingBody{Reader: strings.NewReader(strings.Repeat("x", 1<<20))}
+	hc := &http.Client{Transport: ripeRoundTrip(func(*http.Request) (*http.Response, error) {
+		return &http.Response{StatusCode: 400, Status: "400 Bad Request", Body: body, Header: make(http.Header)}, nil
+	})}
+	if _, err := AnnouncedPrefixes(context.Background(), hc, 1); err == nil || body.read > 4<<10 || !body.closed {
+		t.Fatalf("non200 unbounded read=%d close=%v err=%v", body.read, body.closed, err)
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 20*time.Millisecond)
+	defer cancel()
+	hc.Transport = ripeRoundTrip(func(r *http.Request) (*http.Response, error) { <-r.Context().Done(); return nil, r.Context().Err() })
+	if _, err := ASHolderName(ctx, hc, 1); !errors.Is(err, context.DeadlineExceeded) {
+		t.Fatalf("caller deadline lost: %v", err)
+	}
+}
+
+func TestRIPEGzipExpandedLimit(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Encoding", "gzip")
+		gz := gzip.NewWriter(w)
+		_, _ = io.WriteString(gz, `{"status":"ok","data":{"holder":"Example"}}`+strings.Repeat(" ", 4<<20))
+		_ = gz.Close()
+	}))
+	defer server.Close()
+	t.Setenv("EVOBGP_RIPESTAT_AS_OVERVIEW_URL", server.URL)
+	if _, err := ASHolderName(context.Background(), server.Client(), 1); !errors.Is(err, httpclient.ErrResponseTooLarge) {
+		t.Fatalf("gzip bypassed decoded limit: %v", err)
+	}
+}

internal/birddeploy/apply.go

@@ -1,152 +1,433 @@
-// Package birddeploy applies revision preview fragments to a shared BIRD volume with parse-check and LKG (plan §13).
 package birddeploy
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"io/fs"
 	"os"
 	"path/filepath"
+	"regexp"
+	"sort"
 	"strings"
+	"time"
 
 	"evobgp/internal/birdfmt"
 	"evobgp/internal/store"
 )
 
-// Config drives two-phase deploy from a revision in the store.
+const recoveryTimeout = 30 * time.Second
+
+var includeLine = regexp.MustCompile(`(?m)^([ \t]*include[ \t]+")([^"]+)("[ \t]*;[ \t]*(?:#[^\n]*)?$)`)
+
+// Config identifies the BIRD instance. StagingDir remains for caller compatibility;
+// generations live inside ActiveDir so referenced files survive process restart.
 type Config struct {
-	ActiveDir  string // e.g. /etc/bird (live config tree)
+	ActiveDir  string
-	StagingDir string // e.g. /tmp/evobgp-staging (same host/volume as ActiveDir)
+	StagingDir string
 	BirdBin    string
 	BirdcBin   string
-	Socket     string // birdc -s
+	Socket     string
+	writeFile  func(string, []byte) error // test-only fault injection; nil uses synced writes
 }
 
-// ApplyRevision writes preview fragments to staging, runs bird -p, swaps into active, runs birdc configure; restores LKG on failure.
+// ApplyRevision installs a checked generation through one atomic bird.conf rename.
 func ApplyRevision(ctx context.Context, ctl *birdfmt.BirdCtl, rev *store.Revision, cfg Config) error {
 	if rev == nil || len(rev.PreviewFragments) == 0 {
 		return fmt.Errorf("birddeploy: no preview fragments")
 	}
-	active := strings.TrimSpace(cfg.ActiveDir)
+	if strings.TrimSpace(cfg.ActiveDir) == "" {
-	staging := strings.TrimSpace(cfg.StagingDir)
+		return fmt.Errorf("birddeploy: ActiveDir required")
-	if active == "" || staging == "" {
-		return fmt.Errorf("birddeploy: ActiveDir and StagingDir required")
 	}
-	if err := os.MkdirAll(staging, 0o755); err != nil {
+	active, err := filepath.Abs(cfg.ActiveDir)
-		return err
+	if err != nil {
+		return fmt.Errorf("birddeploy: resolve active directory: %w", err)
 	}
-	if err := os.MkdirAll(filepath.Join(active, "bird.d"), 0o755); err != nil {
+	if err := os.MkdirAll(active, 0o755); err != nil {
-		return err
+		return fmt.Errorf("birddeploy: create active directory: %w", err)
 	}
-	lkg := filepath.Join(active, ".evobgp_lkg")
+	lock, err := lockTarget(ctx, filepath.Join(active, ".evobgp-apply.lock"))
-
+	if err != nil {
-	// Phase 1: write staging tree
+		return fmt.Errorf("birddeploy: acquire apply lock: %w", err)
-	for rel, content := range rev.PreviewFragments {
-		rel = strings.TrimPrefix(rel, "/")
-		if !isDeployableBirdFragment(rel) {
-			continue
 	}
-		dst := filepath.Join(staging, rel)
+	defer func() {
-		if err := os.MkdirAll(filepath.Dir(dst), 0o755); err != nil {
+		if err := lock.Close(); err != nil {
-			return err
+			fmt.Fprintf(os.Stderr, "birddeploy: release apply lock: %v\n", err)
 		}
-		if err := os.WriteFile(dst, []byte(content), 0o644); err != nil {
+	}()
-			return err
+	if err := ctx.Err(); err != nil {
-		}
+		return fmt.Errorf("birddeploy: apply cancelled: %w", err)
-	}
-	mainConf := filepath.Join(staging, "bird.conf")
-	if _, err := os.Stat(mainConf); err != nil {
-		return fmt.Errorf("birddeploy: staging missing bird.conf: %w", err)
 	}
 	bc := ctl
 	if bc == nil {
 		bc = &birdfmt.BirdCtl{Bird: cfg.BirdBin, Birdc: cfg.BirdcBin, Socket: cfg.Socket}
 	}
-	if err := bc.ParseCheck(ctx, mainConf); err != nil {
+	writeFile := cfg.writeFile
-		return fmt.Errorf("birddeploy: bird -p failed: %w", err)
+	if writeFile == nil {
+		writeFile = writeSynced
 	}
-
+	mainPath := filepath.Join(active, "bird.conf")
-	// Snapshot LKG (best-effort copy of current active bird.conf + bird.d)
+	mainInfo, err := os.Lstat(mainPath)
-	_ = os.RemoveAll(lkg)
+	if err != nil {
-	_ = snapshotBirdTree(active, lkg)
+		return fmt.Errorf("birddeploy: inspect current bird.conf: %w", err)
-
-	// Phase 2: atomic swap into active
-	for rel, content := range rev.PreviewFragments {
-		rel = strings.TrimPrefix(rel, "/")
-		if !isDeployableBirdFragment(rel) {
-			continue
 	}
-		dst := filepath.Join(active, rel)
+	if !mainInfo.Mode().IsRegular() {
-		if err := os.MkdirAll(filepath.Dir(dst), 0o755); err != nil {
+		return fmt.Errorf("birddeploy: current bird.conf must be a regular file; refusing to replace %s", mainInfo.Mode())
-			_ = restoreLKG(lkg, active)
+	}
+	oldMain, err := os.ReadFile(mainPath)
+	if err != nil {
+		return fmt.Errorf("birddeploy: read current bird.conf: %w", err)
+	}
+	if err := bc.ParseCheck(ctx, mainPath); err != nil {
+		return fmt.Errorf("birddeploy: current configuration is not a usable LKG: %w", err)
+	}
+	gen, err := os.MkdirTemp(active, ".evobgp-generation-")
+	if err != nil {
+		return fmt.Errorf("birddeploy: create generation: %w", err)
+	}
+	activated := false
+	defer func() {
+		if !activated {
+			if err := os.RemoveAll(gen); err != nil {
+				fmt.Fprintf(os.Stderr, "birddeploy: remove unactivated generation %s: %v\n", gen, err)
+			}
+		}
+	}()
+	if err := writeGenerationWith(active, gen, rev.PreviewFragments, writeFile); err != nil {
 		return err
 	}
-		tmp := dst + ".tmp"
+	newMain, err := os.ReadFile(filepath.Join(gen, "bird.conf"))
-		if err := os.WriteFile(tmp, []byte(content), 0o644); err != nil {
+	if err != nil {
-			_ = restoreLKG(lkg, active)
+		return fmt.Errorf("birddeploy: read generated bird.conf: %w", err)
-			return err
 	}
-		if err := os.Rename(tmp, dst); err != nil {
+	if err := bc.ParseCheck(ctx, filepath.Join(gen, "bird.conf")); err != nil {
-			_ = restoreLKG(lkg, active)
+		return fmt.Errorf("birddeploy: staged bird -p failed: %w", err)
-			return err
 	}
+	if err := writeFile(filepath.Join(gen, "previous-bird.conf"), oldMain); err != nil {
+		return fmt.Errorf("birddeploy: save LKG: %w", err)
+	}
+	if err := syncDir(gen); err != nil {
+		return fmt.Errorf("birddeploy: sync LKG generation: %w", err)
+	}
+	if err := ctx.Err(); err != nil {
+		return fmt.Errorf("birddeploy: apply cancelled: %w", err)
+	}
+	swapped, err := replaceMain(active, newMain)
+	activated = swapped
+	if err != nil {
+		if swapped {
+			return recoverPrevious(bc, active, gen, oldMain, fmt.Errorf("activate generation: %w", err))
+		}
+		return fmt.Errorf("birddeploy: activate generation: %w", err)
 	}
 	if err := bc.Configure(ctx); err != nil {
-		_ = restoreLKG(lkg, active)
+		return recoverPrevious(bc, active, gen, oldMain, err)
-		_ = bc.Configure(ctx)
+	}
-		return fmt.Errorf("birddeploy: birdc configure failed, restored LKG: %w", err)
+	if err := pruneGenerations(active, gen, oldMain); err != nil {
+		fmt.Fprintf(os.Stderr, "birddeploy: configuration applied; generation cleanup failed: %v\n", err)
 	}
 	return nil
 }
 
-// isDeployableBirdFragment skips UI-only preview keys (e.g. flattened full config) not referenced from bird.conf.
+func recoverPrevious(bc *birdfmt.BirdCtl, active, gen string, oldMain []byte, cause error) error {
-func isDeployableBirdFragment(rel string) bool {
+	ctx, cancel := context.WithTimeout(context.Background(), recoveryTimeout)
-	return rel != "" && !strings.HasPrefix(rel, "_")
+	defer cancel()
+	if _, err := replaceMain(active, oldMain); err != nil {
+		return fmt.Errorf("birddeploy: configure failed: %w; restore failed: %v", cause, err)
+	}
+	if err := bc.Configure(ctx); err != nil {
+		return fmt.Errorf("birddeploy: configure failed: %w; LKG restored on disk but recovery configure failed: %v", cause, err)
+	}
+	if err := os.RemoveAll(gen); err != nil {
+		fmt.Fprintf(os.Stderr, "birddeploy: LKG restored; remove failed generation %s: %v\n", gen, err)
+	}
+	return fmt.Errorf("birddeploy: configure failed: %w; LKG restored and recovery configure succeeded", cause)
 }
 
-func snapshotBirdTree(activeRoot, dstRoot string) error {
+func writeGeneration(active, gen string, fragments map[string]string) error {
-	if err := os.MkdirAll(dstRoot, 0o755); err != nil {
+	return writeGenerationWith(active, gen, fragments, writeSynced)
+}
+
+func writeGenerationWith(active, gen string, fragments map[string]string, writeFile func(string, []byte) error) error {
+	main, ok := fragments["bird.conf"]
+	if !ok {
+		return fmt.Errorf("birddeploy: preview missing bird.conf")
+	}
+	oldBirdD := filepath.Join(active, birdfmt.DirBirdD)
+	newBirdD := filepath.Join(gen, birdfmt.DirBirdD)
+	if err := os.MkdirAll(newBirdD, 0o755); err != nil {
+		return fmt.Errorf("birddeploy: create generation bird.d: %w", err)
+	}
+	if err := filepath.WalkDir(oldBirdD, func(path string, d fs.DirEntry, walkErr error) error {
+		if errors.Is(walkErr, fs.ErrNotExist) && path == oldBirdD {
+			return nil
+		}
+		if walkErr != nil {
+			return walkErr
+		}
+		rel, err := filepath.Rel(oldBirdD, path)
+		if err != nil {
 			return err
 		}
-	mc := filepath.Join(activeRoot, "bird.conf")
+		if d.Type()&os.ModeSymlink != 0 {
-	if b, err := os.ReadFile(mc); err == nil {
+			return fmt.Errorf("operator snippet %q is a symlink", rel)
-		_ = os.WriteFile(filepath.Join(dstRoot, "bird.conf"), b, 0o644)
 		}
-	bd := filepath.Join(activeRoot, "bird.d")
+		if rel == "." {
-	if st, err := os.Stat(bd); err == nil && st.IsDir() {
+			if !d.IsDir() {
-		_ = filepath.WalkDir(bd, func(path string, d fs.DirEntry, err error) error {
+				return fmt.Errorf("birddeploy: bird.d is not a directory")
-			if err != nil || d.IsDir() {
+			}
-				return err
+			return nil
+		}
+		if d.IsDir() {
+			return os.MkdirAll(filepath.Join(newBirdD, rel), 0o755)
+		}
+		if !d.Type().IsRegular() {
+			return fmt.Errorf("operator snippet %q is not a regular file", rel)
+		}
+		_, replaced := fragments[filepath.ToSlash(filepath.Join(birdfmt.DirBirdD, rel))]
+		if strings.HasPrefix(filepath.Base(rel), "evobgp_") || replaced {
+			return nil
 		}
-			suffix, _ := filepath.Rel(activeRoot, path)
-			out := filepath.Join(dstRoot, suffix)
-			_ = os.MkdirAll(filepath.Dir(out), 0o755)
 		b, err := os.ReadFile(path)
 		if err != nil {
-				return nil
+			return err
 		}
-			return os.WriteFile(out, b, 0o644)
+		if countIncludeTokens(string(b)) > 0 {
+			return fmt.Errorf("operator snippet %q has a nested include; cannot pin the complete tree", rel)
+		}
+		return writeFile(filepath.Join(newBirdD, rel), b)
+	}); err != nil {
+		return fmt.Errorf("birddeploy: copy operator snippets: %w", err)
+	}
+	for rel, content := range fragments {
+		if rel == "bird.conf" || strings.HasPrefix(rel, "_") {
+			continue
+		}
+		if countIncludeTokens(content) > 0 {
+			return fmt.Errorf("birddeploy: nested include in preview fragment %q is unsupported", rel)
+		}
+		clean := filepath.Clean(filepath.FromSlash(rel))
+		if filepath.IsAbs(clean) || clean == "." || clean == ".." || strings.HasPrefix(clean, ".."+string(os.PathSeparator)) || !strings.HasPrefix(clean, birdfmt.DirBirdD+string(os.PathSeparator)) || filepath.Ext(clean) != ".conf" {
+			return fmt.Errorf("birddeploy: unsafe preview path %q", rel)
+		}
+		dst := filepath.Join(gen, clean)
+		if err := os.MkdirAll(filepath.Dir(dst), 0o755); err != nil {
+			return fmt.Errorf("birddeploy: create fragment directory: %w", err)
+		}
+		if err := writeFile(dst, []byte(content)); err != nil {
+			return fmt.Errorf("birddeploy: write %s: %w", rel, err)
+		}
+	}
+	var rewriteErr error
+	if !standaloneIncludesOnly(main) {
+		return fmt.Errorf("birddeploy: unsupported include syntax in bird.conf")
+	}
+	rewritten := includeLine.ReplaceAllStringFunc(main, func(line string) string {
+		m := includeLine.FindStringSubmatch(line)
+		inc := m[2]
+		if filepath.IsAbs(inc) {
+			rel, err := filepath.Rel(active, inc)
+			if err != nil || rel == ".." || strings.HasPrefix(rel, ".."+string(os.PathSeparator)) {
+				rewriteErr = fmt.Errorf("birddeploy: external absolute include %q is unsupported", inc)
+				return line
+			}
+			inc = rel
+		}
+		clean := filepath.Clean(filepath.FromSlash(inc))
+		if clean == ".." || strings.HasPrefix(clean, ".."+string(os.PathSeparator)) || !strings.HasPrefix(clean, birdfmt.DirBirdD+string(os.PathSeparator)) {
+			rewriteErr = fmt.Errorf("birddeploy: unsupported relative include %q", inc)
+			return line
+		}
+		return m[1] + filepath.ToSlash(filepath.Join(gen, clean)) + m[3]
 	})
+	if rewriteErr != nil {
+		return rewriteErr
+	}
+	if err := writeFile(filepath.Join(gen, ".evobgp-owned"), []byte("evobgp-generation-v1\n")); err != nil {
+		return fmt.Errorf("birddeploy: mark generation: %w", err)
+	}
+	if err := writeFile(filepath.Join(gen, "bird.conf"), []byte(rewritten)); err != nil {
+		return fmt.Errorf("birddeploy: write generated bird.conf: %w", err)
+	}
+	return syncTree(gen)
+}
+
+func standaloneIncludesOnly(config string) bool {
+	tokens := includeTokenOffsets(config)
+	matches := includeLine.FindAllStringIndex(config, -1)
+	if len(tokens) != len(matches) {
+		return false
+	}
+	for i, match := range matches {
+		word := strings.Index(config[match[0]:match[1]], "include")
+		if word < 0 || tokens[i] != match[0]+word {
+			return false
+		}
+	}
+	return true
+}
+
+func countIncludeTokens(config string) int { return len(includeTokenOffsets(config)) }
+
+// includeTokenOffsets skips BIRD comments and strings, retaining exact offsets
+// so a commented standalone include cannot mask a live inline include.
+func includeTokenOffsets(config string) []int {
+	var offsets []int
+	quote := byte(0)
+	lineComment, blockComment := false, false
+	for i := 0; i < len(config); {
+		c := config[i]
+		if lineComment {
+			if c == '\n' {
+				lineComment = false
+			}
+			i++
+			continue
+		}
+		if blockComment {
+			if c == '*' && i+1 < len(config) && config[i+1] == '/' {
+				blockComment = false
+				i += 2
+			} else {
+				i++
+			}
+			continue
+		}
+		if quote != 0 {
+			// BIRD quoted text does not use C-style backslash escapes.
+			// Skipping the following quote could hide a live include.
+			if c == quote {
+				quote = 0
+			}
+			i++
+			continue
+		}
+		if c == '#' {
+			lineComment = true
+			i++
+			continue
+		}
+		if c == '/' && i+1 < len(config) && config[i+1] == '*' {
+			blockComment = true
+			i += 2
+			continue
+		}
+		if c == '"' || c == '\'' {
+			quote = c
+			i++
+			continue
+		}
+		if (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') || c == '_' {
+			start := i
+			for i < len(config) && ((config[i] >= 'a' && config[i] <= 'z') || (config[i] >= 'A' && config[i] <= 'Z') || (config[i] >= '0' && config[i] <= '9') || config[i] == '_') {
+				i++
+			}
+			if config[start:i] == "include" {
+				offsets = append(offsets, start)
+			}
+			continue
+		}
+		i++
+	}
+	return offsets
+}
+
+func syncTree(root string) error {
+	var dirs []string
+	if err := filepath.WalkDir(root, func(path string, d fs.DirEntry, err error) error {
+		if err != nil {
+			return err
+		}
+		if d.IsDir() {
+			dirs = append(dirs, path)
+		}
+		return nil
+	}); err != nil {
+		return err
+	}
+	sort.Slice(dirs, func(i, j int) bool { return len(dirs[i]) > len(dirs[j]) })
+	for _, dir := range dirs {
+		if err := syncDir(dir); err != nil {
+			return fmt.Errorf("sync %s: %w", dir, err)
+		}
 	}
 	return nil
 }
 
-func restoreLKG(lkg, active string) error {
+func pruneGenerations(active, current string, previous []byte) error {
-	if _, err := os.Stat(filepath.Join(lkg, "bird.conf")); err != nil {
+	entries, err := os.ReadDir(active)
-		return err
-	}
-	_ = filepath.WalkDir(lkg, func(path string, d fs.DirEntry, err error) error {
-		if err != nil || d.IsDir() {
-			return err
-		}
-		suffix, _ := filepath.Rel(lkg, path)
-		out := filepath.Join(active, suffix)
-		_ = os.MkdirAll(filepath.Dir(out), 0o755)
-		b, err := os.ReadFile(path)
 	if err != nil {
+		return err
+	}
+	var stale []string
+	for _, entry := range entries {
+		if !entry.IsDir() || !strings.HasPrefix(entry.Name(), ".evobgp-generation-") {
+			continue
+		}
+		path := filepath.Join(active, entry.Name())
+		if path == current || strings.Contains(string(previous), filepath.ToSlash(path)+"/") {
+			continue
+		}
+		marker, err := os.ReadFile(filepath.Join(path, ".evobgp-owned"))
+		if errors.Is(err, fs.ErrNotExist) {
+			continue
+		}
+		if err != nil {
+			return err
+		}
+		if string(marker) != "evobgp-generation-v1\n" {
+			continue
+		}
+		stale = append(stale, path)
+	}
+	sort.Strings(stale)
+	for _, path := range stale {
+		if err := os.RemoveAll(path); err != nil {
+			return fmt.Errorf("remove %s: %w", path, err)
+		}
+	}
 	return nil
 }
-		return os.WriteFile(out, b, 0o644)
+
-	})
+func replaceMain(active string, content []byte) (bool, error) {
-	return nil
+	tmp, err := os.CreateTemp(active, ".evobgp-bird-conf-")
+	if err != nil {
+		return false, err
+	}
+	defer func() { _ = os.Remove(tmp.Name()) }()
+	if err := tmp.Chmod(0o644); err != nil {
+		_ = tmp.Close()
+		return false, err
+	}
+	if _, err := tmp.Write(content); err != nil {
+		_ = tmp.Close()
+		return false, err
+	}
+	if err := tmp.Sync(); err != nil {
+		_ = tmp.Close()
+		return false, err
+	}
+	if err := tmp.Close(); err != nil {
+		return false, err
+	}
+	if err := os.Rename(tmp.Name(), filepath.Join(active, "bird.conf")); err != nil {
+		return false, err
+	}
+	return true, syncDir(active)
+}
+
+func writeSynced(path string, b []byte) error {
+	f, err := os.OpenFile(path, os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0o644)
+	if err != nil {
+		return err
+	}
+	if _, err := f.Write(b); err != nil {
+		_ = f.Close()
+		return err
+	}
+	if err := f.Sync(); err != nil {
+		_ = f.Close()
+		return err
+	}
+	return f.Close()
 }

internal/birddeploy/apply_linux_test.go

@@ -0,0 +1,279 @@
+//go:build linux
+
+package birddeploy
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"sync"
+	"syscall"
+	"testing"
+	"time"
+
+	"evobgp/internal/birdfmt"
+	"evobgp/internal/store"
+)
+
+func fakeCtl(t *testing.T, active string) *birdfmt.BirdCtl {
+	t.Helper()
+	bin := t.TempDir()
+	parse := filepath.Join(bin, "bird")
+	configure := filepath.Join(bin, "birdc")
+	parseScript := `#!/bin/sh
+if grep -q FAIL_PARSE "$2"; then exit 1; fi
+exit 0
+`
+	configureScript := fmt.Sprintf(`#!/bin/sh
+root=%q
+if [ -e "$root/.configure-wait" ]; then
+  rm "$root/.configure-wait"
+  touch "$root/.configure-started"
+  exec sleep 5
+fi
+if [ -e "$root/.configure-fail-once" ]; then
+  rm "$root/.configure-fail-once"
+  exit 1
+fi
+mkdir "$root/.configuring" || exit 1
+sleep 0.05
+rmdir "$root/.configuring"
+`, active)
+	for path, script := range map[string]string{parse: parseScript, configure: configureScript} {
+		if err := os.WriteFile(path, []byte(script), 0o755); err != nil {
+			t.Fatal(err)
+		}
+	}
+	return &birdfmt.BirdCtl{Bird: parse, Birdc: configure}
+}
+
+func revision(label string) *store.Revision {
+	return &store.Revision{PreviewFragments: map[string]string{
+		"bird.conf":              "router id 192.0.2.1;\ninclude \"bird.d/*.conf\";\n# " + label + "\n",
+		"bird.d/evobgp_new.conf": "# " + label + "\n",
+	}}
+}
+
+func setupActive(t *testing.T) (string, *birdfmt.BirdCtl, []byte) {
+	t.Helper()
+	active := t.TempDir()
+	if err := os.Mkdir(filepath.Join(active, "bird.d"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	old := []byte("router id 192.0.2.2;\ninclude \"bird.d/*.conf\";\n# old\n")
+	if err := os.WriteFile(filepath.Join(active, "bird.conf"), old, 0o644); err != nil {
+		t.Fatal(err)
+	}
+	for name, content := range map[string]string{"evobgp_removed.conf": "# removed\n", "operator.conf": "# operator\n"} {
+		if err := os.WriteFile(filepath.Join(active, "bird.d", name), []byte(content), 0o644); err != nil {
+			t.Fatal(err)
+		}
+	}
+	return active, fakeCtl(t, active), old
+}
+
+func TestApplyRevisionGenerationAndPrune(t *testing.T) {
+	active, ctl, _ := setupActive(t)
+	for _, label := range []string{"one", "two", "three"} {
+		if err := ApplyRevision(context.Background(), ctl, revision(label), Config{ActiveDir: active}); err != nil {
+			t.Fatal(err)
+		}
+		main, err := os.ReadFile(filepath.Join(active, "bird.conf"))
+		if err != nil {
+			t.Fatal(err)
+		}
+		if !strings.Contains(string(main), "# "+label) {
+			t.Fatalf("wrong active generation: %s", main)
+		}
+		if strings.Contains(string(main), `include "bird.d/`) {
+			t.Fatalf("include not pinned: %s", main)
+		}
+	}
+	gens, err := filepath.Glob(filepath.Join(active, ".evobgp-generation-*"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(gens) != 2 {
+		t.Fatalf("want active and previous generations, got %d", len(gens))
+	}
+	for _, gen := range gens {
+		if _, err := os.Stat(filepath.Join(gen, "bird.d", "evobgp_removed.conf")); !errors.Is(err, os.ErrNotExist) {
+			t.Fatalf("removed fragment retained in %s: %v", gen, err)
+		}
+		if _, err := os.Stat(filepath.Join(gen, "bird.d", "operator.conf")); err != nil {
+			t.Fatalf("operator fragment missing: %v", err)
+		}
+	}
+}
+
+func TestApplyRevisionFaults(t *testing.T) {
+	for _, mode := range []string{"parse", "write", "disk-full", "configure", "cancel"} {
+		t.Run(mode, func(t *testing.T) {
+			active, ctl, old := setupActive(t)
+			rev := revision(mode)
+			ctx, cancel := context.WithCancel(context.Background())
+			defer cancel()
+			cfg := Config{ActiveDir: active}
+			switch mode {
+			case "parse":
+				rev.PreviewFragments["bird.conf"] += "FAIL_PARSE\n"
+			case "write":
+				rev.PreviewFragments["../escape.conf"] = "bad"
+			case "disk-full":
+				cfg.writeFile = func(path string, b []byte) error {
+					if filepath.Base(path) == "previous-bird.conf" {
+						return syscall.ENOSPC
+					}
+					return writeSynced(path, b)
+				}
+			case "configure":
+				if err := os.WriteFile(filepath.Join(active, ".configure-fail-once"), nil, 0o644); err != nil {
+					t.Fatal(err)
+				}
+			case "cancel":
+				if err := os.WriteFile(filepath.Join(active, ".configure-wait"), nil, 0o644); err != nil {
+					t.Fatal(err)
+				}
+			}
+			result := make(chan error, 1)
+			go func() { result <- ApplyRevision(ctx, ctl, rev, cfg) }()
+			if mode == "cancel" {
+				deadline := time.After(3 * time.Second)
+				for {
+					if _, err := os.Stat(filepath.Join(active, ".configure-started")); err == nil {
+						break
+					}
+					select {
+					case <-deadline:
+						t.Fatal("configure did not start")
+					case <-time.After(10 * time.Millisecond):
+					}
+				}
+				cancel()
+			}
+			err := <-result
+			if err == nil {
+				t.Fatal("expected apply failure")
+			}
+			main, readErr := os.ReadFile(filepath.Join(active, "bird.conf"))
+			if readErr != nil {
+				t.Fatal(readErr)
+			}
+			if string(main) != string(old) {
+				t.Fatalf("old config not restored: %s; error: %v", main, err)
+			}
+			if mode == "configure" || mode == "cancel" {
+				if !strings.Contains(err.Error(), "recovery configure succeeded") {
+					t.Fatalf("unconfirmed recovery: %v", err)
+				}
+			}
+		})
+	}
+}
+
+func TestApplyRevisionSerializes(t *testing.T) {
+	active, ctl, _ := setupActive(t)
+	var wg sync.WaitGroup
+	errs := make(chan error, 2)
+	for _, label := range []string{"first", "second"} {
+		wg.Add(1)
+		go func(label string) {
+			defer wg.Done()
+			errs <- ApplyRevision(context.Background(), ctl, revision(label), Config{ActiveDir: active})
+		}(label)
+	}
+	wg.Wait()
+	close(errs)
+	for err := range errs {
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+}
+
+func TestLockHelperProcess(t *testing.T) {
+	active := os.Getenv("EVOBGP_LOCK_HELPER_ACTIVE")
+	if active == "" {
+		return
+	}
+	lock, err := lockTarget(context.Background(), filepath.Join(active, ".evobgp-apply.lock"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() { _ = lock.Close() }()
+	if err := os.WriteFile(filepath.Join(active, ".helper-ready"), nil, 0o644); err != nil {
+		t.Fatal(err)
+	}
+	for {
+		if _, err := os.Stat(filepath.Join(active, ".helper-done")); err == nil {
+			return
+		}
+		time.Sleep(10 * time.Millisecond)
+	}
+}
+
+func TestApplyRevisionCrossProcessLock(t *testing.T) {
+	active, ctl, old := setupActive(t)
+	cmd := exec.Command(os.Args[0], "-test.run=^TestLockHelperProcess$")
+	cmd.Env = append(os.Environ(), "EVOBGP_LOCK_HELPER_ACTIVE="+active)
+	if err := cmd.Start(); err != nil {
+		t.Fatal(err)
+	}
+	defer func() {
+		_ = os.WriteFile(filepath.Join(active, ".helper-done"), nil, 0o644)
+		if err := cmd.Wait(); err != nil {
+			t.Errorf("lock helper: %v", err)
+		}
+	}()
+	deadline := time.After(3 * time.Second)
+	for {
+		if _, err := os.Stat(filepath.Join(active, ".helper-ready")); err == nil {
+			break
+		}
+		select {
+		case <-deadline:
+			t.Fatal("lock helper did not become ready")
+		case <-time.After(10 * time.Millisecond):
+		}
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 100*time.Millisecond)
+	defer cancel()
+	if err := ApplyRevision(ctx, ctl, revision("blocked"), Config{ActiveDir: active}); !errors.Is(err, context.DeadlineExceeded) {
+		t.Fatalf("want lock timeout, got %v", err)
+	}
+	main, err := os.ReadFile(filepath.Join(active, "bird.conf"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(main) != string(old) {
+		t.Fatalf("apply bypassed cross-process lock: %s", main)
+	}
+}
+
+func TestRealBirdParseGeneration(t *testing.T) {
+	bin := strings.TrimSpace(os.Getenv("EVOBGP_TEST_BIRD_BIN"))
+	if bin == "" {
+		t.Skip("EVOBGP_TEST_BIRD_BIN is not set")
+	}
+	active := t.TempDir()
+	old := []byte("router id 192.0.2.2;\nprotocol device {}\n")
+	if err := os.WriteFile(filepath.Join(active, "bird.conf"), old, 0o644); err != nil {
+		t.Fatal(err)
+	}
+	ctl := fakeCtl(t, active)
+	ctl.Bird = bin
+	rev := &store.Revision{PreviewFragments: map[string]string{
+		"bird.conf":              "router id 192.0.2.1;\ninclude \"bird.d/evobgp_new.conf\";\nprotocol device {}\n",
+		"bird.d/evobgp_new.conf": "# complete generated fragment\n",
+	}}
+	if err := ApplyRevision(context.Background(), ctl, rev, Config{ActiveDir: active}); err != nil {
+		t.Fatal(err)
+	}
+	if err := ctl.ParseCheck(context.Background(), filepath.Join(active, "bird.conf")); err != nil {
+		t.Fatalf("activated config fails real bird -p: %v", err)
+	}
+}

internal/birddeploy/apply_test.go

@@ -0,0 +1,204 @@
+package birddeploy
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"evobgp/internal/store"
+)
+
+func TestWriteGenerationPinsCompleteIncludeTree(t *testing.T) {
+	active := t.TempDir()
+	if err := os.Mkdir(filepath.Join(active, "bird.d"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(active, "bird.d", "operator.conf"), []byte("# operator"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(active, "bird.d", "evobgp_removed.conf"), []byte("# removed"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	gen := filepath.Join(active, ".evobgp-generation-test")
+	if err := os.Mkdir(gen, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	fragments := map[string]string{"bird.conf": "include \"bird.d/*.conf\";\n", "bird.d/evobgp_new.conf": "# new"}
+	if err := writeGeneration(active, gen, fragments); err != nil {
+		t.Fatal(err)
+	}
+	main, err := os.ReadFile(filepath.Join(gen, "bird.conf"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(main), filepath.ToSlash(filepath.Join(gen, "bird.d", "*.conf"))) {
+		t.Fatalf("include is not pinned to generation: %s", main)
+	}
+	for _, name := range []string{"operator.conf", "evobgp_new.conf"} {
+		if _, err := os.Stat(filepath.Join(gen, "bird.d", name)); err != nil {
+			t.Fatalf("%s missing: %v", name, err)
+		}
+	}
+	if _, err := os.Stat(filepath.Join(gen, "bird.d", "evobgp_removed.conf")); !os.IsNotExist(err) {
+		t.Fatalf("removed fragment retained: %v", err)
+	}
+}
+
+func TestWriteGenerationRejectsUnsafePaths(t *testing.T) {
+	for _, bad := range []string{"../escape.conf", "bird.d/../../escape.conf", "other.conf"} {
+		t.Run(bad, func(t *testing.T) {
+			active := t.TempDir()
+			gen := filepath.Join(active, ".evobgp-generation-test")
+			if err := os.Mkdir(gen, 0o755); err != nil {
+				t.Fatal(err)
+			}
+			fragments := map[string]string{"bird.conf": "router id 192.0.2.1;\n", bad: "bad"}
+			if err := writeGeneration(active, gen, fragments); err == nil {
+				t.Fatal("unsafe path accepted")
+			}
+		})
+	}
+}
+
+func TestWriteGenerationRejectsExternalAndNestedIncludes(t *testing.T) {
+	for _, main := range []string{
+		`include "/outside/bird.d/peer.conf";` + "\n",
+		`include "other/peer.conf";` + "\n",
+		`include "bird.d/peer.conf"; include "other.conf";` + "\n",
+		`router id 192.0.2.1; include "/outside/peer.conf";` + "\n",
+		"/*\ninclude \"bird.d/decoy.conf\";\n*/\nrouter id 192.0.2.1; include \"/outside/live.conf\";\n",
+	} {
+		active := t.TempDir()
+		gen := filepath.Join(active, ".evobgp-generation-test")
+		if err := os.Mkdir(gen, 0o755); err != nil {
+			t.Fatal(err)
+		}
+		if err := writeGeneration(active, gen, map[string]string{"bird.conf": main}); err == nil {
+			t.Fatalf("include accepted: %s", main)
+		}
+	}
+	active := t.TempDir()
+	if err := os.Mkdir(filepath.Join(active, "bird.d"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(active, "bird.d", "operator.conf"), []byte(`define X=1; include "/outside/foo.conf";`), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	gen := filepath.Join(active, ".evobgp-generation-test")
+	if err := os.Mkdir(gen, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := writeGeneration(active, gen, map[string]string{"bird.conf": "router id 192.0.2.1;\n"}); err == nil {
+		t.Fatal("nested operator include accepted")
+	}
+}
+
+func TestCountIncludeTokensIgnoresCommentsAndStrings(t *testing.T) {
+	content := "# include \"ignored.conf\";\n" +
+		"/* include \"ignored.conf\"; */\n" +
+		"define note = \"include in string\";\n" +
+		"include \"bird.d/real.conf\";\n"
+	if got := countIncludeTokens(content); got != 1 {
+		t.Fatalf("want one include, got %d", got)
+	}
+}
+
+func TestIncludeAfterBackslashStringCannotHideNestedInclude(t *testing.T) {
+	content := `define note = "\";` + "\n" + `include "/outside/hidden.conf";` + "\n"
+	if got := countIncludeTokens(content); got != 1 {
+		t.Fatalf("live include hidden after quoted backslash: %d", got)
+	}
+	active := t.TempDir()
+	gen := filepath.Join(active, ".evobgp-generation-test")
+	if err := os.Mkdir(gen, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	fragments := map[string]string{"bird.conf": "router id 192.0.2.1;\n", "bird.d/evobgp_test.conf": content}
+	if err := writeGeneration(active, gen, fragments); err == nil {
+		t.Fatal("nested external include accepted after quoted backslash")
+	}
+}
+
+func TestWriteGenerationRewritesAbsoluteActiveInclude(t *testing.T) {
+	active := t.TempDir()
+	gen := filepath.Join(active, ".evobgp-generation-test")
+	if err := os.Mkdir(gen, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	inc := filepath.ToSlash(filepath.Join(active, "bird.d", "peer.conf"))
+	main := "include \"" + inc + "\";\n"
+	if err := writeGeneration(active, gen, map[string]string{"bird.conf": main, "bird.d/peer.conf": "# peer"}); err != nil {
+		t.Fatal(err)
+	}
+	b, err := os.ReadFile(filepath.Join(gen, "bird.conf"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(b), filepath.ToSlash(filepath.Join(gen, "bird.d", "peer.conf"))) {
+		t.Fatalf("absolute include not rewritten: %s", b)
+	}
+}
+
+func TestWriteGenerationRejectsSymlinkBirdDRoot(t *testing.T) {
+	active := t.TempDir()
+	outside := t.TempDir()
+	if err := os.WriteFile(filepath.Join(outside, "operator.conf"), []byte("# operator"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Symlink(outside, filepath.Join(active, "bird.d")); err != nil {
+		t.Skipf("symlinks unavailable: %v", err)
+	}
+	old := []byte("router id 192.0.2.2;\n")
+	if err := os.WriteFile(filepath.Join(active, "bird.conf"), old, 0o644); err != nil {
+		t.Fatal(err)
+	}
+	gen := filepath.Join(active, ".evobgp-generation-test")
+	if err := os.Mkdir(gen, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := writeGeneration(active, gen, map[string]string{"bird.conf": "router id 192.0.2.1;\n"}); err == nil {
+		t.Fatal("bird.d symlink accepted")
+	}
+	current, err := os.ReadFile(filepath.Join(active, "bird.conf"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(current) != string(old) {
+		t.Fatalf("active config changed: %s", current)
+	}
+	if _, err := os.ReadFile(filepath.Join(outside, "operator.conf")); err != nil {
+		t.Fatalf("operator snippet changed: %v", err)
+	}
+}
+
+func TestApplyRevisionRejectsSymlinkMain(t *testing.T) {
+	active := t.TempDir()
+	target := filepath.Join(t.TempDir(), "operator-bird.conf")
+	old := []byte("router id 192.0.2.2;\n")
+	if err := os.WriteFile(target, old, 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Symlink(target, filepath.Join(active, "bird.conf")); err != nil {
+		t.Skipf("symlinks unavailable: %v", err)
+	}
+	err := ApplyRevision(context.Background(), nil, &store.Revision{PreviewFragments: map[string]string{"bird.conf": "router id 192.0.2.1;\n"}}, Config{ActiveDir: active})
+	if err == nil || !strings.Contains(err.Error(), "regular file") {
+		t.Fatalf("want explicit symlink rejection, got %v", err)
+	}
+	info, err := os.Lstat(filepath.Join(active, "bird.conf"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if info.Mode()&os.ModeSymlink == 0 {
+		t.Fatal("active symlink replaced")
+	}
+	current, err := os.ReadFile(target)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(current) != string(old) {
+		t.Fatalf("operator target changed: %s", current)
+	}
+}

internal/birddeploy/lock_linux.go

@@ -0,0 +1,35 @@
+//go:build linux
+
+package birddeploy
+
+import (
+	"context"
+	"errors"
+	"os"
+	"time"
+
+	"golang.org/x/sys/unix"
+)
+
+func lockTarget(ctx context.Context, path string) (*os.File, error) {
+	f, err := os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0o600)
+	if err != nil {
+		return nil, err
+	}
+	for {
+		err = unix.Flock(int(f.Fd()), unix.LOCK_EX|unix.LOCK_NB)
+		if err == nil {
+			return f, nil
+		}
+		if !errors.Is(err, unix.EWOULDBLOCK) && !errors.Is(err, unix.EAGAIN) {
+			_ = f.Close()
+			return nil, err
+		}
+		select {
+		case <-ctx.Done():
+			_ = f.Close()
+			return nil, ctx.Err()
+		case <-time.After(25 * time.Millisecond):
+		}
+	}
+}

internal/birddeploy/lock_other.go

@@ -0,0 +1,13 @@
+//go:build !linux && !windows
+
+package birddeploy
+
+import (
+	"context"
+	"errors"
+	"os"
+)
+
+func lockTarget(context.Context, string) (*os.File, error) {
+	return nil, errors.New("birddeploy: cross-process apply lock is unsupported on this platform")
+}

internal/birddeploy/lock_windows.go

@@ -0,0 +1,36 @@
+//go:build windows
+
+package birddeploy
+
+import (
+	"context"
+	"errors"
+	"os"
+	"time"
+
+	"golang.org/x/sys/windows"
+)
+
+func lockTarget(ctx context.Context, path string) (*os.File, error) {
+	f, err := os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0o600)
+	if err != nil {
+		return nil, err
+	}
+	var overlapped windows.Overlapped
+	for {
+		err = windows.LockFileEx(windows.Handle(f.Fd()), windows.LOCKFILE_EXCLUSIVE_LOCK|windows.LOCKFILE_FAIL_IMMEDIATELY, 0, 1, 0, &overlapped)
+		if err == nil {
+			return f, nil
+		}
+		if !errors.Is(err, windows.ERROR_LOCK_VIOLATION) {
+			_ = f.Close()
+			return nil, err
+		}
+		select {
+		case <-ctx.Done():
+			_ = f.Close()
+			return nil, ctx.Err()
+		case <-time.After(25 * time.Millisecond):
+		}
+	}
+}

internal/birddeploy/sync_unix.go

@@ -0,0 +1,14 @@
+//go:build !windows
+
+package birddeploy
+
+import "os"
+
+func syncDir(path string) error {
+	f, err := os.Open(path)
+	if err != nil {
+		return err
+	}
+	defer f.Close()
+	return f.Sync()
+}

internal/birddeploy/sync_windows.go

@@ -0,0 +1,6 @@
+//go:build windows
+
+package birddeploy
+
+// Windows does not support syncing a directory handle through os.File.Sync.
+func syncDir(string) error { return nil }

internal/bundle/extract.go

@@ -3,34 +3,100 @@ package bundle
 import (
 	"fmt"
 	"os"
-	pathpkg "path"
 	"path/filepath"
 	"strings"
 )
 
 // WriteExtractedFiles writes payload files from a verified bundle under root (directories created as needed).
 func WriteExtractedFiles(root string, v *VerifiedContents) error {
+	return WriteExtractedFilesWithLimits(root, v, Limits{})
+}
+
+// WriteExtractedFilesWithLimits confines all filesystem operations to an open
+// directory handle. Replacing existing files uses unlink + exclusive creation,
+// so existing hard links and symlinks cannot redirect writes outside the root.
+func WriteExtractedFilesWithLimits(root string, v *VerifiedContents, limits Limits) error {
+	l, err := limits.Normalized()
+	if err != nil {
+		return err
+	}
+	if v == nil {
+		return fmt.Errorf("bundle: nil verified contents")
+	}
+	if len(v.Manifest.Files) > l.MaxEntries-2 {
+		return ErrLimit
+	}
+	seen := map[string]bool{}
+	var total int64
+	for _, fe := range v.Manifest.Files {
+		if err := validatePath(fe.Path, l.MaxPathBytes); err != nil {
+			return err
+		}
+		if seen[strings.ToLower(fe.Path)] {
+			return fmt.Errorf("bundle: duplicate path %q", fe.Path)
+		}
+		seen[strings.ToLower(fe.Path)] = true
+		data, ok := v.Files[fe.Path]
+		if !ok {
+			return fmt.Errorf("bundle: missing file %q", fe.Path)
+		}
+		if int64(len(data)) > l.MaxFileBytes || int64(len(data)) > l.MaxExpandedBytes-total {
+			return ErrLimit
+		}
+		total += int64(len(data))
+	}
 	root = filepath.Clean(root)
+	if err := os.MkdirAll(root, 0o755); err != nil {
+		return err
+	}
+	confined, err := os.OpenRoot(root)
+	if err != nil {
+		return err
+	}
+	defer func() { _ = confined.Close() }()
 	for _, fe := range v.Manifest.Files {
 		data, ok := v.Files[fe.Path]
 		if !ok {
 			return fmt.Errorf("bundle: missing file %q", fe.Path)
 		}
-		rel := pathpkg.Clean(strings.TrimPrefix(fe.Path, "/"))
+		parts := strings.Split(fe.Path, "/")
-		if rel == "." || strings.HasPrefix(rel, "..") {
+		for i := 1; i < len(parts); i++ {
-			return fmt.Errorf("bundle: unsafe path %q", fe.Path)
+			dir := filepath.FromSlash(strings.Join(parts[:i], "/"))
-		}
+			if err := confined.Mkdir(dir, 0o755); err != nil && !os.IsExist(err) {
-		dest := filepath.Join(root, filepath.FromSlash(rel))
-		relToRoot, err := filepath.Rel(root, dest)
-		if err != nil || strings.HasPrefix(relToRoot, "..") {
-			return fmt.Errorf("bundle: path escapes root: %q", fe.Path)
-		}
-		if err := os.MkdirAll(filepath.Dir(dest), 0o755); err != nil {
 				return err
 			}
-		if err := os.WriteFile(dest, data, 0o644); err != nil {
+			info, err := confined.Lstat(dir)
+			if err != nil {
 				return err
 			}
+			if !info.IsDir() || info.Mode()&os.ModeSymlink != 0 {
+				return fmt.Errorf("bundle: unsafe directory %q", dir)
+			}
+		}
+		dest := filepath.FromSlash(fe.Path)
+		info, err := confined.Lstat(dest)
+		if err == nil {
+			if !info.Mode().IsRegular() {
+				return fmt.Errorf("bundle: unsafe destination %q", dest)
+			}
+			if err = confined.Remove(dest); err != nil {
+				return err
+			}
+		} else if !os.IsNotExist(err) {
+			return err
+		}
+		f, err := confined.OpenFile(dest, os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0o644)
+		if err != nil {
+			return err
+		}
+		_, writeErr := f.Write(data)
+		closeErr := f.Close()
+		if writeErr != nil {
+			return writeErr
+		}
+		if closeErr != nil {
+			return closeErr
+		}
 	}
 	return nil
 }

internal/bundle/legacy_compatibility_test.go

@@ -0,0 +1,120 @@
+package bundle
+
+import (
+	"archive/tar"
+	"bytes"
+	"compress/gzip"
+	"crypto/ed25519"
+	"crypto/sha256"
+	"encoding/base64"
+	"encoding/hex"
+	"encoding/json"
+	"io"
+	"os"
+	"reflect"
+	"sort"
+	"testing"
+)
+
+// legacyManifestV1 mirrors the decoder fields present in the locally available
+// v1.0.0 source. It deliberately has no future capability or deployment fields.
+type legacyManifestV1 struct {
+	RevisionID string `json:"revision_id"`
+	SpeakerID  string `json:"speaker_id"`
+	CreatedAt  string `json:"created_at"`
+	Files      []struct {
+		Path   string `json:"path"`
+		SHA256 string `json:"sha256"`
+	} `json:"files"`
+	Algorithm       string `json:"signature_algorithm"`
+	PublicKeyBase64 string `json:"public_key_base64"`
+}
+
+func TestLegacyCompatibilityFrozenSignedCurrentBaseline(t *testing.T) {
+	raw, err := os.ReadFile("testdata/current_baseline_signed_bundle.tar.gz")
+	if err != nil {
+		t.Fatal(err)
+	}
+	pub := ed25519.NewKeyFromSeed(bytes.Repeat([]byte{0x42}, ed25519.SeedSize)).Public().(ed25519.PublicKey)
+	verified, err := VerifyGzippedTar(raw, pub)
+	if err != nil {
+		t.Fatalf("frozen signed baseline rejected: %v", err)
+	}
+	if verified.Manifest.RevisionID != "revision-legacy" || verified.Manifest.SpeakerID != "speaker-legacy" || len(verified.Files) != 2 {
+		t.Fatalf("frozen baseline contents: %+v", verified.Manifest)
+	}
+}
+
+func TestLegacyCompatibilityBundleWire(t *testing.T) {
+	seed := bytes.Repeat([]byte{0x42}, ed25519.SeedSize)
+	priv := ed25519.NewKeyFromSeed(seed)
+	pub := priv.Public().(ed25519.PublicKey)
+	fragments := map[string]string{
+		"bird.conf":                      "router id 192.0.2.1;\n",
+		"bird.d/evobgp_prefixes_v4.conf": "# revision\n",
+		"_preview":                       "must not ship",
+	}
+	raw, err := BuildGzippedTar("revision-legacy", "speaker-legacy", fragments, priv)
+	if err != nil {
+		t.Fatal(err)
+	}
+	gr, err := gzip.NewReader(bytes.NewReader(raw))
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() { _ = gr.Close() }()
+	tr := tar.NewReader(gr)
+	entries := map[string][]byte{}
+	var order []string
+	for {
+		h, err := tr.Next()
+		if err == io.EOF {
+			break
+		}
+		if err != nil {
+			t.Fatal(err)
+		}
+		body, err := io.ReadAll(tr)
+		if err != nil {
+			t.Fatal(err)
+		}
+		order = append(order, h.Name)
+		entries[h.Name] = body
+	}
+	wantOrder := []string{"manifest.json", "bundle.sig", "bird.conf", "bird.d/evobgp_prefixes_v4.conf"}
+	if !reflect.DeepEqual(order, wantOrder) {
+		t.Fatalf("legacy tar layout: got %v, want %v", order, wantOrder)
+	}
+	var top map[string]json.RawMessage
+	if err := json.Unmarshal(entries["manifest.json"], &top); err != nil {
+		t.Fatal(err)
+	}
+	wantFields := []string{"created_at", "files", "public_key_base64", "revision_id", "signature_algorithm", "speaker_id"}
+	fields := make([]string, 0, len(top))
+	for k := range top {
+		fields = append(fields, k)
+	}
+	sort.Strings(fields)
+	if !reflect.DeepEqual(fields, wantFields) {
+		t.Fatalf("manifest fields: got %v, want %v", fields, wantFields)
+	}
+	var m legacyManifestV1
+	if err := json.Unmarshal(entries["manifest.json"], &m); err != nil {
+		t.Fatal(err)
+	}
+	if m.RevisionID != "revision-legacy" || m.SpeakerID != "speaker-legacy" || m.CreatedAt == "" || m.Algorithm != "ed25519" || m.PublicKeyBase64 != base64.StdEncoding.EncodeToString(pub) {
+		t.Fatalf("legacy manifest fields: %+v", m)
+	}
+	if !ed25519.Verify(pub, entries["manifest.json"], entries["bundle.sig"]) {
+		t.Fatal("legacy Ed25519 verifier rejected current signed manifest")
+	}
+	if len(m.Files) != 2 {
+		t.Fatalf("legacy manifest file count: %d", len(m.Files))
+	}
+	for _, f := range m.Files {
+		sum := sha256.Sum256(entries[f.Path])
+		if hex.EncodeToString(sum[:]) != f.SHA256 {
+			t.Fatalf("legacy SHA-256 mismatch for %q", f.Path)
+		}
+	}
+}

internal/bundle/limits.go

@@ -0,0 +1,103 @@
+package bundle
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"path"
+	"strings"
+)
+
+// ErrLimit reports that a transport, archive or extraction budget was exceeded.
+var ErrLimit = errors.New("bundle: resource limit exceeded")
+
+// Limits bounds both archive transport and decoded tar data, including metadata.
+// Zero fields select defaults. These ceilings permit large (1M-prefix) configs;
+// they are safety ceilings, not a promise of suitability for a small-RAM node.
+type Limits struct {
+	MaxCompressedBytes int64
+	MaxExpandedBytes   int64
+	MaxFileBytes       int64
+	MaxManifestBytes   int64
+	MaxEntries         int
+	MaxPathBytes       int
+}
+
+// DefaultLimits permits large production configs while bounding untrusted input.
+func DefaultLimits() Limits {
+	return Limits{MaxCompressedBytes: 128 << 20, MaxExpandedBytes: 512 << 20, MaxFileBytes: 256 << 20, MaxManifestBytes: 2 << 20, MaxEntries: 4096, MaxPathBytes: 1024}
+}
+
+// Normalized fills zero fields with defaults and rejects invalid budgets.
+func (l Limits) Normalized() (Limits, error) {
+	d := DefaultLimits()
+	values := []struct {
+		v        *int64
+		fallback int64
+	}{{&l.MaxCompressedBytes, d.MaxCompressedBytes}, {&l.MaxExpandedBytes, d.MaxExpandedBytes}, {&l.MaxFileBytes, d.MaxFileBytes}, {&l.MaxManifestBytes, d.MaxManifestBytes}}
+	for _, x := range values {
+		if *x.v < 0 || *x.v == int64(^uint64(0)>>1) {
+			return l, fmt.Errorf("bundle: invalid byte limit")
+		}
+		if *x.v == 0 {
+			*x.v = x.fallback
+		}
+	}
+	if l.MaxEntries < 0 || l.MaxPathBytes < 0 {
+		return l, fmt.Errorf("bundle: invalid entry/path limit")
+	}
+	if l.MaxEntries == 0 {
+		l.MaxEntries = d.MaxEntries
+	}
+	if l.MaxPathBytes == 0 {
+		l.MaxPathBytes = d.MaxPathBytes
+	}
+	return l, nil
+}
+
+func validatePath(name string, limit int) error {
+	if name == "" || len(name) > limit || path.IsAbs(name) || path.Clean(name) != name || name == "." || strings.ContainsAny(name, "\\:\x00") {
+		return fmt.Errorf("bundle: unsafe path %q", name)
+	}
+	for _, part := range strings.Split(name, "/") {
+		if part == ".." || strings.TrimRight(part, " .") != part {
+			return fmt.Errorf("bundle: unsafe path %q", name)
+		}
+		base := strings.ToUpper(strings.SplitN(part, ".", 2)[0])
+		if base == "CON" || base == "PRN" || base == "AUX" || base == "NUL" || (len(base) == 4 && (strings.HasPrefix(base, "COM") || strings.HasPrefix(base, "LPT")) && base[3] >= '1' && base[3] <= '9') {
+			return fmt.Errorf("bundle: reserved path %q", name)
+		}
+	}
+	return nil
+}
+
+type limitedWriter struct {
+	w         io.Writer
+	remaining int64
+}
+
+func (w *limitedWriter) Write(p []byte) (int, error) {
+	if int64(len(p)) > w.remaining {
+		return 0, ErrLimit
+	}
+	n, err := w.w.Write(p)
+	w.remaining -= int64(n)
+	return n, err
+}
+
+// ReadCompressed rejects oversized input while reading, including chunked HTTP
+// bodies and local files whose size changes after stat.
+func ReadCompressed(r io.Reader, limits Limits) ([]byte, error) {
+	l, err := limits.Normalized()
+	if err != nil {
+		return nil, err
+	}
+	b, err := io.ReadAll(io.LimitReader(r, l.MaxCompressedBytes+1))
+	if err != nil {
+		return nil, err
+	}
+	if int64(len(b)) > l.MaxCompressedBytes {
+		return nil, ErrLimit
+	}
+	return b, nil
+}

internal/bundle/limits_test.go

@@ -0,0 +1,286 @@
+package bundle
+
+import (
+	"archive/tar"
+	"bytes"
+	"compress/gzip"
+	"crypto/ed25519"
+	"errors"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"runtime"
+	"strings"
+	"testing"
+)
+
+type testEntry struct {
+	header tar.Header
+	body   []byte
+}
+
+func archiveEntries(t *testing.T, entries []testEntry) []byte {
+	t.Helper()
+	var b bytes.Buffer
+	gz := gzip.NewWriter(&b)
+	tw := tar.NewWriter(gz)
+	for _, entry := range entries {
+		h := entry.header
+		if h.Typeflag == 0 {
+			h.Typeflag = tar.TypeReg
+		}
+		h.Size = int64(len(entry.body))
+		if err := tw.WriteHeader(&h); err != nil {
+			t.Fatal(err)
+		}
+		if _, err := tw.Write(entry.body); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if err := tw.Close(); err != nil {
+		t.Fatal(err)
+	}
+	if err := gz.Close(); err != nil {
+		t.Fatal(err)
+	}
+	return b.Bytes()
+}
+func testKey() ed25519.PrivateKey {
+	return ed25519.NewKeyFromSeed(bytes.Repeat([]byte{7}, ed25519.SeedSize))
+}
+
+func TestBundleResourceLimits(t *testing.T) {
+	pub := testKey().Public().(ed25519.PublicKey)
+	cases := []struct {
+		name    string
+		entries []testEntry
+		limits  Limits
+	}{
+		{"file", []testEntry{{header: tar.Header{Name: "bird.conf"}, body: bytes.Repeat([]byte{'a'}, 1025)}}, Limits{MaxFileBytes: 1024}},
+		{"expanded", []testEntry{{header: tar.Header{Name: "bird.conf"}, body: bytes.Repeat([]byte{'a'}, 8192)}}, Limits{MaxExpandedBytes: 4096}},
+		{"manifest", []testEntry{{header: tar.Header{Name: manifestName}, body: bytes.Repeat([]byte{'a'}, 1025)}}, Limits{MaxManifestBytes: 1024}},
+		{"entries", []testEntry{{header: tar.Header{Name: "a"}}, {header: tar.Header{Name: "b"}}, {header: tar.Header{Name: "c"}}}, Limits{MaxEntries: 2}},
+		{"pax-metadata", []testEntry{{header: tar.Header{Name: "bird.conf", Format: tar.FormatPAX, PAXRecords: map[string]string{"comment": strings.Repeat("a", 8192)}}}}, Limits{MaxExpandedBytes: 4096}},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			raw := archiveEntries(t, tc.entries)
+			if _, err := VerifyGzippedTarWithLimits(raw, pub, tc.limits); !errors.Is(err, ErrLimit) {
+				t.Fatalf("want limit, got %v", err)
+			}
+		})
+	}
+	if _, err := VerifyGzippedTarWithLimits(make([]byte, 65), pub, Limits{MaxCompressedBytes: 64}); !errors.Is(err, ErrLimit) {
+		t.Fatalf("compressed cap: %v", err)
+	}
+}
+
+func TestBundleRejectsUnsafeEntries(t *testing.T) {
+	pub := testKey().Public().(ed25519.PublicKey)
+	for _, name := range []string{"../escape", "/etc/bird.conf", `bird.d\escape`, "C:/escape", "bird.d/../escape", "a//b", "a/./b", "NUL", "bird.conf:stream", "a. /b"} {
+		t.Run(name, func(t *testing.T) {
+			raw := archiveEntries(t, []testEntry{{header: tar.Header{Name: name}, body: []byte("x")}})
+			if _, err := VerifyGzippedTar(raw, pub); err == nil {
+				t.Fatal("unsafe path accepted")
+			}
+		})
+	}
+	for _, kind := range []byte{tar.TypeSymlink, tar.TypeLink, tar.TypeDir, tar.TypeFifo} {
+		raw := archiveEntries(t, []testEntry{{header: tar.Header{Name: "bird.conf", Typeflag: kind, Linkname: "outside"}}})
+		if _, err := VerifyGzippedTar(raw, pub); err == nil {
+			t.Fatalf("type %d accepted", kind)
+		}
+	}
+	raw := archiveEntries(t, []testEntry{{header: tar.Header{Name: "bird.conf"}}, {header: tar.Header{Name: "bird.conf"}}})
+	if _, err := VerifyGzippedTar(raw, pub); err == nil || !strings.Contains(err.Error(), "duplicate") {
+		t.Fatalf("duplicate accepted: %v", err)
+	}
+}
+
+func TestBundleChecksTrailerAndTrailingExpansion(t *testing.T) {
+	priv := testKey()
+	pub := priv.Public().(ed25519.PublicKey)
+	raw, err := BuildGzippedTar("r", "s", map[string]string{"bird.conf": "router id 192.0.2.1;"}, priv)
+	if err != nil {
+		t.Fatal(err)
+	}
+	corrupt := append([]byte{}, raw...)
+	corrupt[len(corrupt)-8] ^= 0xff
+	if _, err = VerifyGzippedTar(corrupt, pub); err == nil {
+		t.Fatal("gzip checksum ignored after tar EOF")
+	}
+	var extra bytes.Buffer
+	gz := gzip.NewWriter(&extra)
+	if _, err = gz.Write(bytes.Repeat([]byte{'x'}, 32<<10)); err != nil {
+		t.Fatal(err)
+	}
+	if err = gz.Close(); err != nil {
+		t.Fatal(err)
+	}
+	joined := append(append([]byte{}, raw...), extra.Bytes()...)
+	if _, err = VerifyGzippedTarWithLimits(joined, pub, Limits{MaxExpandedBytes: 8 << 10}); !errors.Is(err, ErrLimit) {
+		t.Fatalf("trailing gzip member bypassed budget: %v", err)
+	}
+}
+
+func TestBundleRejectsPAXSparseRegularFile(t *testing.T) {
+	// archive/tar expands PAX sparse holes even though the returned type is a
+	// regular file. Build a physical one-byte entry with a 4096-byte logical size.
+	record := func(key, value string) string {
+		body := key + "=" + value + "\n"
+		n := len(body) + 2
+		for {
+			next := len(fmt.Sprint(n)) + 1 + len(body)
+			if next == n {
+				return fmt.Sprintf("%d %s", n, body)
+			}
+			n = next
+		}
+	}
+	pax := record("GNU.sparse.size", "4096") + record("GNU.sparse.numblocks", "1") + record("GNU.sparse.map", "0,1")
+	var physical bytes.Buffer
+	tw := tar.NewWriter(&physical)
+	for _, entry := range []testEntry{{header: tar.Header{Name: "PaxHeader", Size: int64(len(pax))}, body: []byte(pax)}, {header: tar.Header{Name: "bird.conf", Size: 1}, body: []byte("x")}} {
+		if err := tw.WriteHeader(&entry.header); err != nil {
+			t.Fatal(err)
+		}
+		if _, err := tw.Write(entry.body); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if err := tw.Close(); err != nil {
+		t.Fatal(err)
+	}
+	data := physical.Bytes()
+	data[156] = tar.TypeXHeader
+	for i := 148; i < 156; i++ {
+		data[i] = ' '
+	}
+	sum := 0
+	for _, b := range data[:512] {
+		sum += int(b)
+	}
+	copy(data[148:156], fmt.Sprintf("%06o\x00 ", sum))
+	h, err := tar.NewReader(bytes.NewReader(data)).Next()
+	if err != nil || h.Typeflag != tar.TypeReg || h.Size != 4096 {
+		t.Fatalf("invalid sparse regression fixture %+v %v", h, err)
+	}
+	var compressed bytes.Buffer
+	gz := gzip.NewWriter(&compressed)
+	if _, err = gz.Write(data); err != nil {
+		t.Fatal(err)
+	}
+	if err = gz.Close(); err != nil {
+		t.Fatal(err)
+	}
+	if _, err = VerifyGzippedTar(compressed.Bytes(), testKey().Public().(ed25519.PublicKey)); err == nil || !strings.Contains(err.Error(), "sparse") {
+		t.Fatalf("PAX sparse accepted: %v", err)
+	}
+}
+
+func TestBundleRejectsUnmanifestedPayload(t *testing.T) {
+	priv := testKey()
+	raw, err := BuildGzippedTar("r", "s", map[string]string{"bird.conf": "x"}, priv)
+	if err != nil {
+		t.Fatal(err)
+	}
+	gr, err := gzip.NewReader(bytes.NewReader(raw))
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() { _ = gr.Close() }()
+	tr := tar.NewReader(gr)
+	var entries []testEntry
+	for {
+		h, e := tr.Next()
+		if e == io.EOF {
+			break
+		}
+		if e != nil {
+			t.Fatal(e)
+		}
+		b, e := io.ReadAll(tr)
+		if e != nil {
+			t.Fatal(e)
+		}
+		entries = append(entries, testEntry{header: *h, body: b})
+	}
+	entries = append(entries, testEntry{header: tar.Header{Name: "unsigned.conf"}, body: []byte("unsigned")})
+	if _, err = VerifyGzippedTar(archiveEntries(t, entries), priv.Public().(ed25519.PublicKey)); err == nil {
+		t.Fatal("unmanifested payload accepted")
+	}
+}
+
+func TestExtractionDoesNotFollowLinks(t *testing.T) {
+	root := t.TempDir()
+	outside := filepath.Join(t.TempDir(), "outside")
+	if err := os.WriteFile(outside, []byte("preserve"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	dest := filepath.Join(root, "bird.conf")
+	if err := os.Link(outside, dest); err != nil {
+		if runtime.GOOS == "linux" {
+			t.Fatalf("Linux link regression requires hard links: %v", err)
+		}
+		t.Skipf("hard links unavailable: %v", err)
+	}
+	v := &VerifiedContents{Manifest: Manifest{Files: []FileEntry{{Path: "bird.conf"}}}, Files: map[string][]byte{"bird.conf": []byte("replacement")}}
+	if err := WriteExtractedFiles(root, v); err != nil {
+		t.Fatal(err)
+	}
+	got, err := os.ReadFile(outside)
+	if err != nil || string(got) != "preserve" {
+		t.Fatalf("hard link modified outside: %s %v", got, err)
+	}
+	if err = os.Remove(dest); err != nil {
+		t.Fatal(err)
+	}
+	if err = os.Symlink(outside, dest); err != nil {
+		if runtime.GOOS == "linux" {
+			t.Fatalf("Linux link regression requires symlinks: %v", err)
+		}
+		t.Skipf("symlinks unavailable: %v", err)
+	}
+	if err = WriteExtractedFiles(root, v); err == nil {
+		t.Fatal("symlink destination accepted")
+	}
+	got, err = os.ReadFile(outside)
+	if err != nil || string(got) != "preserve" {
+		t.Fatal("symlink modified outside")
+	}
+}
+
+func TestBundleBuilderBoundsAndPortablePaths(t *testing.T) {
+	for _, name := range []string{"/bird.conf", "a/../bird.conf", manifestName, sigFileName, `bird.d\bad`} {
+		if _, err := BuildGzippedTar("r", "s", map[string]string{name: "x"}, testKey()); err == nil {
+			t.Fatalf("invalid builder path accepted: %q", name)
+		}
+	}
+	if _, err := BuildGzippedTarWithLimits("r", "s", map[string]string{"bird.conf": strings.Repeat("x", 1025)}, testKey(), Limits{MaxFileBytes: 1024}); !errors.Is(err, ErrLimit) {
+		t.Fatalf("builder file cap %v", err)
+	}
+	if _, err := BuildGzippedTarWithLimits("r", "s", map[string]string{"bird.conf": "x"}, testKey(), Limits{MaxExpandedBytes: 1024}); !errors.Is(err, ErrLimit) {
+		t.Fatalf("builder tar overhead cap %v", err)
+	}
+}
+
+// This is a decoded-size regression using repeated representative route lines,
+// not a unique-prefix cardinality test or a BIRD validity/performance benchmark.
+func TestDefaultLimitsAllowMillionRepresentativeIPv6RouteLines(t *testing.T) {
+	if testing.Short() {
+		t.Skip("large representative config")
+	}
+	config := strings.Repeat("route 2001:db8:1234:5678::/64 blackhole;\n", 1_000_000)
+	raw, err := BuildGzippedTar("large", "speaker", map[string]string{"bird.conf": config}, testKey())
+	if err != nil {
+		t.Fatal(err)
+	}
+	v, err := VerifyGzippedTar(raw, testKey().Public().(ed25519.PublicKey))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(v.Files["bird.conf"]) != len(config) {
+		t.Fatal("large config truncated")
+	}
+}

internal/bundle/pack.go

@@ -9,7 +9,6 @@ import (
 	"encoding/base64"
 	"encoding/json"
 	"fmt"
-	"path"
 	"sort"
 	"strings"
 	"time"
@@ -36,22 +35,47 @@ const manifestName = "manifest.json"
 
 // BuildGzippedTar builds a .tar.gz with manifest.json, bundle.sig (Ed25519 over manifest JSON), and payload files.
 func BuildGzippedTar(revisionID, speakerID string, fragments map[string]string, priv ed25519.PrivateKey) ([]byte, error) {
+	return BuildGzippedTarWithLimits(revisionID, speakerID, fragments, priv, Limits{})
+}
+
+// BuildGzippedTarWithLimits preserves the manifest/signature wire format while
+// refusing payloads that exceed the receiving node's configured budgets.
+func BuildGzippedTarWithLimits(revisionID, speakerID string, fragments map[string]string, priv ed25519.PrivateKey, limits Limits) ([]byte, error) {
+	l, err := limits.Normalized()
+	if err != nil {
+		return nil, err
+	}
 	if len(priv) != ed25519.PrivateKeySize {
 		return nil, fmt.Errorf("bundle: invalid ed25519 private key size")
 	}
 	pub := priv.Public().(ed25519.PublicKey)
 
-	norm := make(map[string]string, len(fragments))
+	norm := make(map[string]string)
+	seen := map[string]bool{}
+	var total int64
 	for p, content := range fragments {
-		p = path.Clean(strings.TrimPrefix(p, "/"))
-		if p == "." || strings.HasPrefix(p, "..") {
-			return nil, fmt.Errorf("bundle: invalid path %q", p)
-		}
 		if strings.HasPrefix(p, "_") {
 			continue // UI-only preview aggregate, not shipped to nodes
 		}
+		if len(norm) >= l.MaxEntries-2 {
+			return nil, ErrLimit
+		}
+		if err := validatePath(p, l.MaxPathBytes); err != nil {
+			return nil, err
+		}
+		if p == manifestName || p == sigFileName || seen[strings.ToLower(p)] {
+			return nil, fmt.Errorf("bundle: duplicate or reserved path %q", p)
+		}
+		seen[strings.ToLower(p)] = true
+		if int64(len(content)) > l.MaxFileBytes || int64(len(content)) > l.MaxExpandedBytes-total {
+			return nil, ErrLimit
+		}
+		total += int64(len(content))
 		norm[p] = content
 	}
+	if len(norm) > l.MaxEntries-2 {
+		return nil, ErrLimit
+	}
 
 	var files []FileEntry
 	var payload [][]byte
@@ -75,11 +99,14 @@ func BuildGzippedTar(revisionID, speakerID string, fragments map[string]string,
 	if err != nil {
 		return nil, err
 	}
+	if int64(len(manifestJSON)) > l.MaxManifestBytes {
+		return nil, ErrLimit
+	}
 	sig := ed25519.Sign(priv, manifestJSON)
 
 	var buf bytes.Buffer
-	gw := gzip.NewWriter(&buf)
+	gw := gzip.NewWriter(&limitedWriter{w: &buf, remaining: l.MaxCompressedBytes})
-	tw := tar.NewWriter(gw)
+	tw := tar.NewWriter(&limitedWriter{w: gw, remaining: l.MaxExpandedBytes})
 
 	add := func(name string, body []byte) error {
 		hdr := &tar.Header{

internal/bundle/verify.go

@@ -11,44 +11,95 @@ import (
 	"encoding/json"
 	"fmt"
 	"io"
-	"path"
 	"strings"
 )
 
-// VerifiedContents is the result of verifying a downloaded bundle on evobgp-node.
+// VerifiedContents contains only payloads covered by a verified signed manifest.
 type VerifiedContents struct {
 	Manifest Manifest
 	Files    map[string][]byte
 }
 
-// VerifyGzippedTar checks Ed25519 signature over manifest.json and SHA-256 of each file.
+// VerifyGzippedTar verifies the unchanged signed manifest format with default limits.
-func VerifyGzippedTar(bundle []byte, pub ed25519.PublicKey) (*VerifiedContents, error) {
+func VerifyGzippedTar(raw []byte, pub ed25519.PublicKey) (*VerifiedContents, error) {
-	gr, err := gzip.NewReader(bytes.NewReader(bundle))
+	return VerifyGzippedTarWithLimits(raw, pub, Limits{})
+}
+
+// VerifyGzippedTarWithLimits verifies the signature and hashes under explicit
+// compressed, expanded, per-entry and metadata budgets.
+func VerifyGzippedTarWithLimits(raw []byte, pub ed25519.PublicKey, limits Limits) (*VerifiedContents, error) {
+	l, err := limits.Normalized()
+	if err != nil {
+		return nil, err
+	}
+	if int64(len(raw)) > l.MaxCompressedBytes {
+		return nil, ErrLimit
+	}
+	if len(pub) != ed25519.PublicKeySize {
+		return nil, fmt.Errorf("bundle: invalid public key size")
+	}
+	gr, err := gzip.NewReader(bytes.NewReader(raw))
 	if err != nil {
 		return nil, err
 	}
 	defer func() { _ = gr.Close() }()
-
+	decoded := &io.LimitedReader{R: gr, N: l.MaxExpandedBytes + 1}
-	var manifestRaw []byte
+	tr := tar.NewReader(decoded)
-	var sig []byte
+	var manifestRaw, sig []byte
-	files := make(map[string][]byte)
+	files := map[string][]byte{}
-	tr := tar.NewReader(gr)
+	seen := map[string]bool{}
+	count := 0
+	var logicalBytes int64
 	for {
 		hdr, err := tr.Next()
+		if decoded.N == 0 {
+			return nil, ErrLimit
+		}
 		if err == io.EOF {
 			break
 		}
 		if err != nil {
 			return nil, err
 		}
-		name := path.Clean(hdr.Name)
+		count++
-		if strings.HasPrefix(name, "..") {
+		if count > l.MaxEntries {
-			return nil, fmt.Errorf("bundle: illegal tar entry %q", hdr.Name)
+			return nil, ErrLimit
 		}
-		data, err := io.ReadAll(tr)
+		if err = validatePath(hdr.Name, l.MaxPathBytes); err != nil {
+			return nil, err
+		}
+		name := hdr.Name
+		folded := strings.ToLower(name)
+		if seen[folded] {
+			return nil, fmt.Errorf("bundle: duplicate tar entry %q", name)
+		}
+		seen[folded] = true
+		if hdr.Typeflag != tar.TypeReg && hdr.Typeflag != tar.TypeRegA {
+			return nil, fmt.Errorf("bundle: nonregular tar entry %q", name)
+		}
+		for key := range hdr.PAXRecords {
+			if strings.HasPrefix(key, "GNU.sparse.") {
+				return nil, fmt.Errorf("bundle: sparse entry %q is unsupported", name)
+			}
+		}
+		max := l.MaxFileBytes
+		if name == manifestName {
+			max = l.MaxManifestBytes
+		}
+		if name == sigFileName {
+			max = ed25519.SignatureSize
+		}
+		if hdr.Size < 0 || hdr.Size > max || hdr.Size >= decoded.N || hdr.Size > l.MaxExpandedBytes-logicalBytes {
+			return nil, ErrLimit
+		}
+		logicalBytes += hdr.Size
+		data, err := io.ReadAll(io.LimitReader(tr, max+1))
 		if err != nil {
 			return nil, err
 		}
+		if int64(len(data)) > max || decoded.N == 0 {
+			return nil, ErrLimit
+		}
 		switch name {
 		case manifestName:
 			manifestRaw = data
@@ -58,17 +109,36 @@ func VerifyGzippedTar(bundle []byte, pub ed25519.PublicKey) (*VerifiedContents,
 			files[name] = data
 		}
 	}
-	if len(manifestRaw) == 0 || len(sig) == 0 {
+	// tar EOF precedes the gzip trailer. Drain through the same byte budget so
+	// checksums, extra gzip members and trailing decompression are also bounded.
+	if _, err = io.Copy(io.Discard, decoded); err != nil {
+		return nil, err
+	}
+	if decoded.N == 0 {
+		return nil, ErrLimit
+	}
+	if len(manifestRaw) == 0 || len(sig) != ed25519.SignatureSize {
 		return nil, fmt.Errorf("bundle: missing manifest or signature")
 	}
 	if !ed25519.Verify(pub, manifestRaw, sig) {
 		return nil, fmt.Errorf("bundle: ed25519 signature mismatch")
 	}
 	var m Manifest
-	if err := json.Unmarshal(manifestRaw, &m); err != nil {
+	if err = json.Unmarshal(manifestRaw, &m); err != nil {
 		return nil, err
 	}
+	if len(m.Files) > l.MaxEntries-2 {
+		return nil, ErrLimit
+	}
+	listed := map[string]bool{}
 	for _, fe := range m.Files {
+		if err = validatePath(fe.Path, l.MaxPathBytes); err != nil {
+			return nil, err
+		}
+		if fe.Path == manifestName || fe.Path == sigFileName || listed[strings.ToLower(fe.Path)] {
+			return nil, fmt.Errorf("bundle: duplicate or reserved manifest path %q", fe.Path)
+		}
+		listed[strings.ToLower(fe.Path)] = true
 		body, ok := files[fe.Path]
 		if !ok {
 			return nil, fmt.Errorf("bundle: missing file %q", fe.Path)
@@ -78,10 +148,13 @@ func VerifyGzippedTar(bundle []byte, pub ed25519.PublicKey) (*VerifiedContents,
 			return nil, fmt.Errorf("bundle: checksum mismatch for %q", fe.Path)
 		}
 	}
+	if len(listed) != len(files) {
+		return nil, fmt.Errorf("bundle: payload is not covered by manifest")
+	}
 	return &VerifiedContents{Manifest: m, Files: files}, nil
 }
 
-// ParsePublicKeyHex decodes a 64-char hex Ed25519 public key (32 bytes).
+// ParsePublicKeyHex decodes a 32-byte Ed25519 public key from hexadecimal.
 func ParsePublicKeyHex(s string) (ed25519.PublicKey, error) {
 	s = strings.TrimSpace(strings.TrimPrefix(strings.TrimPrefix(s, "0x"), "0X"))
 	raw, err := hex.DecodeString(s)
@@ -94,7 +167,8 @@ func ParsePublicKeyHex(s string) (ed25519.PublicKey, error) {
 	return ed25519.PublicKey(raw), nil
 }
 
-// PublicKeyFromManifestBase64 uses the key embedded in the signed manifest (cross-check with expected pubkey optional).
+// PublicKeyFromManifestBase64 decodes the manifest key. Callers must still trust
+// an independently provisioned key when authenticating the bundle.
 func PublicKeyFromManifestBase64(m *Manifest) (ed25519.PublicKey, error) {
 	if m.PublicKeyB64 == "" {
 		return nil, fmt.Errorf("bundle: manifest missing public_key_base64")

internal/db/apply.go

@@ -8,48 +8,292 @@ import (
 	"path"
 	"sort"
 	"strings"
+	"time"
 
 	"github.com/jackc/pgx/v5"
 	"github.com/jackc/pgx/v5/pgxpool"
 )
 
-// ApplyPostgresMigrations runs ordered *.up.sql from an embedded subtree (idempotent via schema_migrations).
+const (
-func ApplyPostgresMigrations(ctx context.Context, pool *pgxpool.Pool, fsys fs.FS, subdir string) error {
+	postgresMigrationLockKey int64 = 0x45766f4247504d47 // "EvoBGPMG", scoped to the database.
-	if _, err := pool.Exec(ctx, `CREATE TABLE IF NOT EXISTS schema_migrations (version TEXT NOT NULL PRIMARY KEY)`); err != nil {
+	migrationLockTimeout           = 30 * time.Second
-		return fmt.Errorf("db: schema_migrations: %w", err)
+	migrationCleanupTimeout        = 5 * time.Second
-	}
+)
+
+// PostgreSQL migrations are transaction-only. Non-transactional statements
+// need an explicit, separately reviewed execution path.
+
+// ApplyPostgresMigrations runs ordered *.up.sql from an embedded subtree.
+// Each migration and its version row commit atomically under a session advisory lock.
+func ApplyPostgresMigrations(ctx context.Context, pool *pgxpool.Pool, fsys fs.FS, subdir string) (retErr error) {
 	entries, err := fs.ReadDir(fsys, subdir)
 	if err != nil {
-		return err
+		return fmt.Errorf("db: read migrations: %w", err)
 	}
 	var ups []string
 	for _, e := range entries {
-		if e.IsDir() || !strings.HasSuffix(e.Name(), ".up.sql") {
+		if !e.IsDir() && strings.HasSuffix(e.Name(), ".up.sql") {
-			continue
-		}
 			ups = append(ups, e.Name())
 		}
+	}
 	sort.Strings(ups)
-	for _, name := range ups {
+	// Include pool acquisition in the lock timeout. The lock precedes even
-		ver := strings.TrimSuffix(name, ".up.sql")
+	// schema_migrations creation, because another process may start from an empty DB.
-		var dummy int
+	lockCtx, cancelLock := context.WithTimeout(ctx, migrationLockTimeout)
-		err := pool.QueryRow(ctx, `SELECT 1 FROM schema_migrations WHERE version = $1`, ver).Scan(&dummy)
+	conn, err := pool.Acquire(lockCtx)
+	if err != nil {
+		cancelLock()
+		return fmt.Errorf("db: acquire migration connection: %w", err)
+	}
+	defer conn.Release()
+	if _, err := conn.Exec(lockCtx, `SELECT pg_advisory_lock($1)`, postgresMigrationLockKey); err != nil {
+		cancelLock()
+		// Cancellation may race with a successful server-side lock acquisition.
+		closeCtx, closeCancel := context.WithTimeout(context.Background(), migrationCleanupTimeout)
+		defer closeCancel()
+		if closeErr := conn.Conn().Close(closeCtx); closeErr != nil {
+			return errors.Join(fmt.Errorf("db: acquire migration lock: %w", err), fmt.Errorf("db: close migration connection: %w", closeErr))
+		}
+		return fmt.Errorf("db: acquire migration lock: %w", err)
+	}
+	cancelLock()
+	defer func() {
+		cleanupCtx, cancel := context.WithTimeout(context.Background(), migrationCleanupTimeout)
+		defer cancel()
+		var unlocked bool
+		if err := conn.QueryRow(cleanupCtx, `SELECT pg_advisory_unlock($1)`, postgresMigrationLockKey).Scan(&unlocked); err != nil || !unlocked {
 			if err == nil {
-			continue
+				err = errors.New("lock was not held")
 			}
-		if !errors.Is(err, pgx.ErrNoRows) {
+			retErr = errors.Join(retErr, fmt.Errorf("db: release migration lock: %w", err))
-			return err
+			// Never return a possibly locked session to the pool.
+			closeCtx, closeCancel := context.WithTimeout(context.Background(), migrationCleanupTimeout)
+			defer closeCancel()
+			if closeErr := conn.Conn().Close(closeCtx); closeErr != nil {
+				retErr = errors.Join(retErr, fmt.Errorf("db: close migration connection: %w", closeErr))
 			}
+		}
+	}()
+	if _, err := conn.Exec(ctx, `CREATE TABLE IF NOT EXISTS schema_migrations (version TEXT NOT NULL PRIMARY KEY)`); err != nil {
+		return fmt.Errorf("db: schema_migrations: %w", err)
+	}
+	for _, name := range ups {
 		body, err := fs.ReadFile(fsys, path.Join(subdir, name))
 		if err != nil {
-			return err
+			return fmt.Errorf("db: read migration %s: %w", name, err)
 		}
-		if _, err := pool.Exec(ctx, string(body)); err != nil {
+		if bad := unsupportedMigrationStatement(string(body)); bad != "" {
+			return fmt.Errorf("db: migration %s uses unsupported non-transactional SQL %q", name, bad)
+		}
+		if err := applyPostgresMigration(ctx, conn, strings.TrimSuffix(name, ".up.sql"), body); err != nil {
 			return fmt.Errorf("db: migrate %s: %w", name, err)
 		}
-		if _, err := pool.Exec(ctx, `INSERT INTO schema_migrations (version) VALUES ($1)`, ver); err != nil {
-			return err
-		}
 	}
 	return nil
 }
+
+// unsupportedMigrationStatement inspects statement starts after masking SQL
+// comments and quoted strings. In particular, a comment after a semicolon
+// must not conceal COMMIT or another command that would escape our transaction.
+func unsupportedMigrationStatement(sql string) string {
+	const (
+		normal = iota
+		singleQuote
+		doubleQuote
+		lineComment
+		blockComment
+		dollarQuote
+	)
+	state := normal
+	blockDepth := 0
+	dollarTag := ""
+	escapeString := false
+	var clean strings.Builder
+	for i := 0; i < len(sql); {
+		switch state {
+		case normal:
+			switch {
+			case sql[i] == '\'':
+				escapeString = i > 0 && (sql[i-1] == 'E' || sql[i-1] == 'e') && (i == 1 || !isSQLIdentifierByte(sql[i-2]))
+				state = singleQuote
+				clean.WriteByte(' ')
+				i++
+			case sql[i] == '"':
+				state = doubleQuote
+				clean.WriteByte(' ')
+				i++
+			case strings.HasPrefix(sql[i:], "--"):
+				state = lineComment
+				clean.WriteByte(' ')
+				i += 2
+			case strings.HasPrefix(sql[i:], "/*"):
+				state = blockComment
+				blockDepth = 1
+				clean.WriteByte(' ')
+				i += 2
+			case sql[i] == '$' && (i == 0 || !isSQLIdentifierByte(sql[i-1])):
+				end := i + 1
+				if end < len(sql) && isDollarTagStartByte(sql[end]) {
+					end++
+					for end < len(sql) && isDollarTagByte(sql[end]) {
+						end++
+					}
+				}
+				if end < len(sql) && sql[end] == '$' {
+					dollarTag = sql[i : end+1]
+					state = dollarQuote
+					clean.WriteByte(' ')
+					i = end + 1
+				} else {
+					clean.WriteByte(sql[i])
+					i++
+				}
+			default:
+				clean.WriteByte(sql[i])
+				i++
+			}
+		case singleQuote:
+			switch {
+			case escapeString && sql[i] == '\\' && i+1 < len(sql):
+				i += 2
+			case sql[i] == '\'' && i+1 < len(sql) && sql[i+1] == '\'':
+				i += 2
+			case sql[i] == '\'':
+				state = normal
+				i++
+			default:
+				i++
+			}
+		case doubleQuote:
+			if sql[i] == '"' {
+				if i+1 < len(sql) && sql[i+1] == '"' {
+					i += 2
+				} else {
+					state = normal
+					i++
+				}
+			} else {
+				i++
+			}
+		case lineComment:
+			if sql[i] == '\n' || sql[i] == '\r' {
+				clean.WriteByte('\n')
+				state = normal
+			}
+			i++
+		case blockComment:
+			switch {
+			case strings.HasPrefix(sql[i:], "/*"):
+				blockDepth++
+				i += 2
+			case strings.HasPrefix(sql[i:], "*/"):
+				blockDepth--
+				i += 2
+				if blockDepth == 0 {
+					state = normal
+				}
+			default:
+				i++
+			}
+		case dollarQuote:
+			if strings.HasPrefix(sql[i:], dollarTag) {
+				i += len(dollarTag)
+				state = normal
+			} else {
+				i++
+			}
+		}
+	}
+	for _, statement := range strings.Split(clean.String(), ";") {
+		words := strings.Fields(strings.ToUpper(statement))
+		if isUnsupportedMigrationCommand(words) {
+			return strings.Join(words[:min(len(words), 4)], " ")
+		}
+	}
+	return ""
+}
+
+func isSQLIdentifierByte(b byte) bool {
+	return isDollarTagByte(b) || b == '$'
+}
+
+func isDollarTagStartByte(b byte) bool {
+	return b >= 'a' && b <= 'z' || b >= 'A' && b <= 'Z' || b == '_' || b >= 0x80
+}
+
+func isDollarTagByte(b byte) bool {
+	return isDollarTagStartByte(b) || b >= '0' && b <= '9'
+}
+
+func isUnsupportedMigrationCommand(words []string) bool {
+	if len(words) == 0 {
+		return false
+	}
+	second := ""
+	if len(words) > 1 {
+		second = words[1]
+	}
+	switch words[0] {
+	case "BEGIN", "COMMIT", "END", "ROLLBACK", "ABORT", "SAVEPOINT", "REINDEX", "VACUUM", "CLUSTER":
+		return true
+	case "START", "PREPARE":
+		return second == "TRANSACTION"
+	case "RELEASE":
+		return second == "SAVEPOINT" || second != ""
+	case "ALTER":
+		return second == "SYSTEM"
+	case "CREATE":
+		if second == "DATABASE" || second == "TABLESPACE" {
+			return true
+		}
+		idx := 2
+		if second == "UNIQUE" {
+			idx = 3
+		}
+		return len(words) > idx && words[idx-1] == "INDEX" && words[idx] == "CONCURRENTLY"
+	case "DROP":
+		return second == "DATABASE" || second == "TABLESPACE" || second == "INDEX" && len(words) > 2 && words[2] == "CONCURRENTLY"
+	case "REFRESH":
+		return len(words) > 3 && second == "MATERIALIZED" && words[2] == "VIEW" && words[3] == "CONCURRENTLY"
+	case "DISCARD":
+		return second == "ALL"
+	}
+	return false
+}
+
+func applyPostgresMigration(ctx context.Context, conn *pgxpool.Conn, version string, body []byte) (retErr error) {
+	tx, err := conn.Begin(ctx)
+	if err != nil {
+		return err
+	}
+	committed := false
+	defer func() {
+		if committed {
+			return
+		}
+		cleanupCtx, cancel := context.WithTimeout(context.Background(), migrationCleanupTimeout)
+		defer cancel()
+		if err := tx.Rollback(cleanupCtx); err != nil && !errors.Is(err, pgx.ErrTxClosed) {
+			retErr = errors.Join(retErr, fmt.Errorf("db: rollback migration %s: %w", version, err))
+		}
+	}()
+	var present int
+	err = tx.QueryRow(ctx, `SELECT 1 FROM schema_migrations WHERE version = $1`, version).Scan(&present)
+	if err == nil {
+		return nil
+	}
+	if !errors.Is(err, pgx.ErrNoRows) {
+		return err
+	}
+	if _, err := tx.Exec(ctx, string(body)); err != nil {
+		return err
+	}
+	if _, err := tx.Exec(ctx, `INSERT INTO schema_migrations (version) VALUES ($1)`, version); err != nil {
+		return err
+	}
+	if err := tx.Commit(ctx); err != nil {
+		return err
+	}
+	committed = true
+	return nil
+}

internal/db/apply_test.go

@@ -0,0 +1,63 @@
+package db
+
+import (
+	"io/fs"
+	"path"
+	"strings"
+	"testing"
+
+	"evobgp/migrations"
+)
+
+func TestPostgresEmbeddedMigrationsTransactional(t *testing.T) {
+	entries, err := fs.ReadDir(migrations.Postgres, "postgres")
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, entry := range entries {
+		if !strings.HasSuffix(entry.Name(), ".up.sql") {
+			continue
+		}
+		body, err := fs.ReadFile(migrations.Postgres, path.Join("postgres", entry.Name()))
+		if err != nil {
+			t.Fatal(err)
+		}
+		if bad := unsupportedMigrationStatement(string(body)); bad != "" {
+			t.Errorf("%s contains unsupported transaction control or DDL: %q", entry.Name(), bad)
+		}
+	}
+}
+
+func TestPostgresMigrationRejectsTransactionControl(t *testing.T) {
+	for _, statement := range []string{
+		"COMMIT;", "END;", "ROLLBACK;", "ABORT;", "BEGIN;",
+		"CREATE INDEX CONCURRENTLY idx ON t (id);", "VACUUM t;",
+		"CREATE TABLE t(id int); /* boundary */ COMMIT; SELECT 1/0;",
+		"/* comment */COMMIT;",
+		"-- comment\nCOMMIT;",
+		"/* nested /* comment */ still comment */ END;",
+		"CREATE TABLE foo$tag$ (id int); COMMIT; -- $tag$",
+		"-- comment\rCOMMIT;",
+		`foo$E'abc\'; COMMIT;`,
+		`éE'abc\'; COMMIT;`,
+	} {
+		if unsupportedMigrationStatement(statement) == "" {
+			t.Errorf("did not reject %q", statement)
+		}
+	}
+}
+
+func TestPostgresMigrationIgnoresQuotedControlWords(t *testing.T) {
+	for _, statement := range []string{
+		"SELECT 'COMMIT;'::text;",
+		"SELECT $$COMMIT;$$;",
+		"SELECT $tag$END;$tag$;",
+		"SELECT E'ABORT;\\'';",
+		"SELECT \"ROLLBACK\" FROM t;",
+		"-- COMMIT;\nSELECT 1;",
+	} {
+		if bad := unsupportedMigrationStatement(statement); bad != "" {
+			t.Errorf("rejected quoted or commented SQL %q: %s", statement, bad)
+		}
+	}
+}

internal/db/open.go

@@ -25,17 +25,7 @@ func OpenPostgresPool(ctx context.Context, dsn string) (*pgxpool.Pool, error) {
 	if err != nil {
 		return nil, err
 	}
-	if max := os.Getenv("EVOBGP_DB_MAX_CONNS"); max != "" {
+	applyPostgresPoolConfig(cfg)
-		if n, err := strconv.Atoi(strings.TrimSpace(max)); err == nil && n > 0 {
-			cfg.MaxConns = int32(n)
-		}
-	}
-	if min := os.Getenv("EVOBGP_DB_MIN_CONNS"); min != "" {
-		if n, err := strconv.Atoi(strings.TrimSpace(min)); err == nil && n >= 0 {
-			cfg.MinConns = int32(n)
-		}
-	}
-	cfg.MaxConnLifetime = 30 * time.Minute
 	pool, err := pgxpool.NewWithConfig(ctx, cfg)
 	if err != nil {
 		return nil, err
@@ -47,6 +37,29 @@ func OpenPostgresPool(ctx context.Context, dsn string) (*pgxpool.Pool, error) {
 	return pool, nil
 }
 
+// applyPostgresPoolConfig sets pool limits. An explicit EVOBGP_DB_MAX_CONNS is kept.
+// When it is unset, the pool defaults to 8 connections so it can cover the default
+// worker count instead of pgx's default of 4.
+func applyPostgresPoolConfig(cfg *pgxpool.Config) {
+	if cfg == nil {
+		return
+	}
+	if max := strings.TrimSpace(os.Getenv("EVOBGP_DB_MAX_CONNS")); max != "" {
+		if n, err := strconv.Atoi(max); err == nil && n > 0 {
+			cfg.MaxConns = int32(n)
+		}
+	} else {
+		cfg.MaxConns = 8
+	}
+	if min := strings.TrimSpace(os.Getenv("EVOBGP_DB_MIN_CONNS")); min != "" {
+		if n, err := strconv.Atoi(min); err == nil && n >= 0 {
+			cfg.MinConns = int32(n)
+		}
+	}
+	cfg.MaxConnLifetime = 30 * time.Minute
+	cfg.MaxConnIdleTime = 5 * time.Minute
+}
+
 // OpenSQLite applies sqlite/*.up.sql then opens the DB (foreign keys on).
 func OpenSQLite(ctx context.Context, filePath string) (*sql.DB, error) {
 	dsn := fmt.Sprintf("file:%s?_pragma=foreign_keys(1)", strings.TrimPrefix(filePath, "file:"))

internal/db/pool_config_test.go

@@ -0,0 +1,34 @@
+package db
+
+import (
+	"testing"
+	"time"
+
+	"github.com/jackc/pgx/v5/pgxpool"
+)
+
+func TestApplyPostgresPoolConfigDefaultAndExplicit(t *testing.T) {
+	t.Setenv("EVOBGP_DB_MAX_CONNS", "")
+	t.Setenv("EVOBGP_DB_MIN_CONNS", "")
+	cfg, err := pgxpool.ParseConfig("postgres://localhost/evobgp?sslmode=disable")
+	if err != nil {
+		t.Fatal(err)
+	}
+	applyPostgresPoolConfig(cfg)
+	if cfg.MaxConns != 8 {
+		t.Fatalf("default max conns: %d", cfg.MaxConns)
+	}
+	if cfg.MaxConnIdleTime != 5*time.Minute {
+		t.Fatalf("idle time: %s", cfg.MaxConnIdleTime)
+	}
+
+	t.Setenv("EVOBGP_DB_MAX_CONNS", "25")
+	cfg, err = pgxpool.ParseConfig("postgres://localhost/evobgp?sslmode=disable")
+	if err != nil {
+		t.Fatal(err)
+	}
+	applyPostgresPoolConfig(cfg)
+	if cfg.MaxConns != 25 {
+		t.Fatalf("explicit max conns overridden: %d", cfg.MaxConns)
+	}
+}

internal/db/postgres_integration_test.go

@@ -0,0 +1,227 @@
+package db
+
+import (
+	"context"
+	"crypto/rand"
+	"encoding/hex"
+	"errors"
+	"io/fs"
+	"os"
+	"strings"
+	"sync"
+	"testing"
+	"testing/fstest"
+	"time"
+
+	"evobgp/migrations"
+	"github.com/jackc/pgx/v5/pgxpool"
+)
+
+func TestPostgresMigrationsIntegration(t *testing.T) {
+	pool := isolatedMigrationPool(t)
+	ctx := context.Background()
+	if err := ApplyPostgresMigrations(ctx, pool, migrations.Postgres, "postgres"); err != nil {
+		t.Fatalf("first migration pass: %v", err)
+	}
+	if err := ApplyPostgresMigrations(ctx, pool, migrations.Postgres, "postgres"); err != nil {
+		t.Fatalf("second migration pass: %v", err)
+	}
+	files, err := fs.ReadDir(migrations.Postgres, "postgres")
+	if err != nil {
+		t.Fatal(err)
+	}
+	want := 0
+	for _, file := range files {
+		if strings.HasSuffix(file.Name(), ".up.sql") {
+			want++
+		}
+	}
+	var got int
+	if err := pool.QueryRow(ctx, `SELECT count(*)::int FROM schema_migrations`).Scan(&got); err != nil {
+		t.Fatal(err)
+	}
+	if got != want {
+		t.Fatalf("applied migrations = %d, want %d", got, want)
+	}
+	var extension string
+	if err := pool.QueryRow(ctx, `SELECT extname FROM pg_extension WHERE extname = 'pg_stat_statements'`).Scan(&extension); err != nil {
+		t.Fatalf("pg_stat_statements extension missing: %v", err)
+	}
+}
+
+func TestPostgresMigrationConcurrentApplyIntegration(t *testing.T) {
+	pool := isolatedMigrationPool(t)
+	files := fstest.MapFS{"sql/000001_probe.up.sql": {Data: []byte(`CREATE TABLE migration_probe (id integer PRIMARY KEY); SELECT pg_sleep(0.1);`)}}
+	var wg sync.WaitGroup
+	errors := make(chan error, 2)
+	for i := 0; i < 2; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			errors <- ApplyPostgresMigrations(context.Background(), pool, files, "sql")
+		}()
+	}
+	wg.Wait()
+	close(errors)
+	for err := range errors {
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+	assertMigrationVersionCount(t, pool, 1)
+}
+
+func TestPostgresMigrationRollbackIntegration(t *testing.T) {
+	pool := isolatedMigrationPool(t)
+	files := fstest.MapFS{"sql/000001_broken.up.sql": {Data: []byte(`CREATE TABLE migration_partial (id integer); SELECT 1 / 0;`)}}
+	if err := ApplyPostgresMigrations(context.Background(), pool, files, "sql"); err == nil {
+		t.Fatal("expected migration error")
+	}
+	assertMigrationVersionCount(t, pool, 0)
+	var exists bool
+	if err := pool.QueryRow(context.Background(), `SELECT to_regclass('migration_partial') IS NOT NULL`).Scan(&exists); err != nil {
+		t.Fatal(err)
+	}
+	if exists {
+		t.Fatal("failed migration left a table behind")
+	}
+	// A transaction terminator hidden behind a comment must be rejected before
+	// the first DDL statement can run.
+	files = fstest.MapFS{"sql/000001_broken.up.sql": {Data: []byte(`CREATE TABLE migration_partial (id integer); /* boundary */ COMMIT; SELECT 1 / 0;`)}}
+	if err := ApplyPostgresMigrations(context.Background(), pool, files, "sql"); err == nil || !strings.Contains(err.Error(), "unsupported non-transactional SQL") {
+		t.Fatalf("comment-hidden COMMIT error = %v", err)
+	}
+	assertMigrationVersionCount(t, pool, 0)
+	if err := pool.QueryRow(context.Background(), `SELECT to_regclass('migration_partial') IS NOT NULL`).Scan(&exists); err != nil {
+		t.Fatal(err)
+	}
+	if exists {
+		t.Fatal("rejected migration left a table behind")
+	}
+	files = fstest.MapFS{"sql/000001_broken.up.sql": {Data: []byte(`CREATE TABLE foo$tag$ (id int); COMMIT; -- $tag$`)}}
+	if err := ApplyPostgresMigrations(context.Background(), pool, files, "sql"); err == nil || !strings.Contains(err.Error(), "unsupported non-transactional SQL") {
+		t.Fatalf("identifier-embedded dollar tag error = %v", err)
+	}
+	assertMigrationVersionCount(t, pool, 0)
+	if err := pool.QueryRow(context.Background(), `SELECT to_regclass('foo$tag$') IS NOT NULL`).Scan(&exists); err != nil {
+		t.Fatal(err)
+	}
+	if exists {
+		t.Fatal("identifier-embedded dollar tag left a table behind")
+	}
+}
+
+func TestPostgresMigrationIdempotentIntegration(t *testing.T) {
+	pool := isolatedMigrationPool(t)
+	files := fstest.MapFS{"sql/000001_once.up.sql": {Data: []byte(`CREATE TABLE migration_once (id integer); INSERT INTO migration_once VALUES (1);`)}}
+	for i := 0; i < 2; i++ {
+		if err := ApplyPostgresMigrations(context.Background(), pool, files, "sql"); err != nil {
+			t.Fatal(err)
+		}
+	}
+	assertMigrationVersionCount(t, pool, 1)
+	var rows int
+	if err := pool.QueryRow(context.Background(), `SELECT count(*) FROM migration_once`).Scan(&rows); err != nil {
+		t.Fatal(err)
+	}
+	if rows != 1 {
+		t.Fatalf("migration rows = %d, want 1", rows)
+	}
+}
+
+func TestPostgresMigrationLockCancellationIntegration(t *testing.T) {
+	pool := isolatedMigrationPool(t)
+	locker, err := pool.Acquire(context.Background())
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer locker.Release()
+	if _, err := locker.Exec(context.Background(), `SELECT pg_advisory_lock($1)`, postgresMigrationLockKey); err != nil {
+		t.Fatal(err)
+	}
+	defer func() {
+		_, _ = locker.Exec(context.Background(), `SELECT pg_advisory_unlock($1)`, postgresMigrationLockKey)
+	}()
+	files := fstest.MapFS{"sql/000001_probe.up.sql": {Data: []byte(`CREATE TABLE lock_probe (id integer);`)}}
+	ctx, cancel := context.WithTimeout(context.Background(), 250*time.Millisecond)
+	defer cancel()
+	err = ApplyPostgresMigrations(ctx, pool, files, "sql")
+	if !errors.Is(err, context.DeadlineExceeded) {
+		t.Fatalf("lock wait error = %v, want context deadline exceeded", err)
+	}
+	var exists bool
+	if err := pool.QueryRow(context.Background(), `SELECT to_regclass('schema_migrations') IS NOT NULL`).Scan(&exists); err != nil {
+		t.Fatal(err)
+	}
+	if exists {
+		t.Fatal("schema_migrations was created before migration lock acquisition")
+	}
+	var unlocked bool
+	if err := locker.QueryRow(context.Background(), `SELECT pg_advisory_unlock($1)`, postgresMigrationLockKey).Scan(&unlocked); err != nil || !unlocked {
+		t.Fatalf("release held lock: unlocked=%v err=%v", unlocked, err)
+	}
+	if err := ApplyPostgresMigrations(context.Background(), pool, files, "sql"); err != nil {
+		t.Fatalf("apply after lock release: %v", err)
+	}
+}
+
+func isolatedMigrationPool(t *testing.T) *pgxpool.Pool {
+	t.Helper()
+	dsn := os.Getenv("EVOBGP_TEST_DATABASE_URL")
+	if dsn == "" {
+		t.Skip("EVOBGP_TEST_DATABASE_URL not set")
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+	defer cancel()
+	admin, err := pgxpool.New(ctx, dsn)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var nonce [8]byte
+	if _, err := rand.Read(nonce[:]); err != nil {
+		admin.Close()
+		t.Fatal(err)
+	}
+	schema := "evobgp_mig_test_" + hex.EncodeToString(nonce[:])
+	if _, err := admin.Exec(ctx, `CREATE SCHEMA `+schema); err != nil {
+		admin.Close()
+		t.Fatal(err)
+	}
+	var pool *pgxpool.Pool
+	t.Cleanup(func() {
+		if pool != nil {
+			pool.Close()
+		}
+		cleanupCtx, cleanupCancel := context.WithTimeout(context.Background(), 30*time.Second)
+		defer cleanupCancel()
+		if _, err := admin.Exec(cleanupCtx, `DROP SCHEMA `+schema+` CASCADE`); err != nil {
+			t.Errorf("drop isolated schema %s: %v", schema, err)
+		}
+		admin.Close()
+	})
+	cfg, err := pgxpool.ParseConfig(dsn)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if cfg.ConnConfig.RuntimeParams == nil {
+		cfg.ConnConfig.RuntimeParams = make(map[string]string)
+	}
+	cfg.ConnConfig.RuntimeParams["search_path"] = schema
+	cfg.MaxConns = 3
+	pool, err = pgxpool.NewWithConfig(ctx, cfg)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return pool
+}
+
+func assertMigrationVersionCount(t *testing.T, pool *pgxpool.Pool, want int) {
+	t.Helper()
+	var got int
+	if err := pool.QueryRow(context.Background(), `SELECT count(*) FROM schema_migrations`).Scan(&got); err != nil {
+		t.Fatal(err)
+	}
+	if got != want {
+		t.Fatalf("migration versions = %d, want %d", got, want)
+	}
+}

internal/httpapi/backend_stage0_repro_test.go

@@ -0,0 +1,39 @@
+package httpapi
+
+import (
+	"testing"
+
+	"evobgp/internal/store"
+)
+
+// These regressions ensure unobserved live state cannot inherit persisted health.
+func TestStage0ReproPeerWithoutObservationIsNotEstablished(t *testing.T) {
+	peer := &store.BGPPeer{ID: "aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee", Neighbor: "198.51.100.2"}
+	cases := []struct {
+		name  string
+		views []speakerBGPLive
+	}{
+		{name: "no_collector"},
+		{name: "all_timeout", views: []speakerBGPLive{{SpeakerID: "replica", Error: "timeout"}}},
+		{name: "session_absent", views: []speakerBGPLive{{SpeakerID: "replica", Label: "replica"}}},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			row := map[string]any{"session_state": "Established"}
+			applyPeerLiveFields(row, peer, tc.views)
+			if row["session_state"] == "Established" {
+				t.Fatal("S01: persisted Established is exposed without a current session observation")
+			}
+		})
+	}
+}
+
+func TestStage0ReproUnpolledMasterIsNotHealthy(t *testing.T) {
+	sp := &store.Speaker{ID: "master", Role: "master"}
+	// collectSpeakerLiveStatus constructs this view when no poll result exists.
+	view := speakerBGPLive{SpeakerID: sp.ID, Label: "master"}
+	row := speakerLiveStatusJSON(sp, view, nil)
+	if row["bgp_poll_ok"] == true || row["agent_ok"] == true {
+		t.Fatal("S01: absent poll is exposed as successful BGP/agent health")
+	}
+}

internal/httpapi/bootstrap.go

@@ -7,9 +7,9 @@ import (
 	"time"
 
 	"evobgp/internal/db"
-	"evobgp/internal/httpclient"
 	"evobgp/internal/jobs"
 	"evobgp/internal/observability"
+	"evobgp/internal/pipeline"
 	"evobgp/internal/repository"
 	"evobgp/internal/store"
 
@@ -18,7 +18,7 @@ import (
 
 // NewCDNHTTPClient returns the shared HTTP client for CDN and preview fetches (PERF-02 / ERR-03).
 func NewCDNHTTPClient() *http.Client {
-	return httpclient.New(httpclient.DefaultTimeout)
+	return pipeline.NewCDNHTTPClient()
 }
 
 // BootstrapWorkers opens the same store.Backend and jobs.Registry as New (without HTTP or bundle keys).
@@ -50,46 +50,15 @@ func BootstrapWorkers(ctx context.Context, opts Options) (store.Backend, *jobs.R
 		backend = mem
 	}
 
-	cdnHTTP := NewCDNHTTPClient()
+	// Configured DoH profiles and RIPE service overrides retain the previous
-	wk := &jobs.Worker{Store: backend, PgPool: pool, HTTPClient: cdnHTTP}
+	// general HTTP/HTTPS transport, including explicitly configured private
+	// endpoints. CDN calls independently secure this client in UpstreamHTTPDo.
+	workerHTTP := pipeline.NewPipelineHTTPClient()
+	wk := &jobs.Worker{Store: backend, PgPool: pool, HTTPClient: workerHTTP}
 	reg := jobs.NewRegistry(wk.Process)
 	wk.Registry = reg
 	if pool != nil {
-		audit := repository.NewJobAuditWriter(pool)
+		reg.UseDurableQueue(backend)
-		jobMeta := func(j *jobs.Job) map[string]any {
-			if j == nil {
-				return nil
-			}
-			st := j.Snapshot()
-			meta, _ := st["meta"].(map[string]any)
-			return meta
-		}
-		reg.SetPersistHooks(
-			func(j *jobs.Job) {
-				if j == nil {
-					return
-				}
-				audit.UpsertQueued(context.Background(), j.TenantID, j.ID, j.Kind, j.IdempotencyKey, j.ModuleID, jobMeta(j))
-			},
-			func(j *jobs.Job) {
-				if j == nil {
-					return
-				}
-				audit.UpsertRunning(context.Background(), j.TenantID, j.ID, j.Kind, j.IdempotencyKey, jobMeta(j))
-			},
-			func(j *jobs.Job) {
-				if j == nil {
-					return
-				}
-				st := j.Snapshot()
-				status, _ := st["status"].(string)
-				var errMsg *string
-				if e, ok := st["error"].(string); ok && e != "" {
-					errMsg = &e
-				}
-				audit.MarkTerminal(context.Background(), j.TenantID, j.ID, status, errMsg, time.Now().UTC())
-			},
-		)
 	}
 	observability.RegisterStoreBackend(backend)
 	return backend, reg, pool, nil

internal/httpapi/bootstrap_outbound_test.go

@@ -0,0 +1,110 @@
+package httpapi
+
+import (
+	"context"
+	"encoding/base64"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"evobgp/internal/httpclient"
+	"evobgp/internal/jobs"
+	"evobgp/internal/pipeline"
+	"evobgp/internal/store"
+	"github.com/miekg/dns"
+)
+
+func TestBootstrapWorkerPreservesExplicitPrivateDOH(t *testing.T) {
+	t.Setenv("EVOBGP_NODE_DISPATCH_ENABLED", "0")
+	t.Setenv("EVOBGP_BIRDC_SOCKET", "")
+	t.Setenv("EVOBGP_CDN_ALLOW_PRIVATE", "0")
+	var calls atomic.Int32
+	resolver := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		calls.Add(1)
+		raw, err := base64.RawURLEncoding.DecodeString(r.URL.Query().Get("dns"))
+		if err != nil {
+			w.WriteHeader(400)
+			return
+		}
+		query := new(dns.Msg)
+		if err = query.Unpack(raw); err != nil || len(query.Question) == 0 {
+			w.WriteHeader(400)
+			return
+		}
+		answer := new(dns.Msg)
+		answer.SetReply(query)
+		if query.Question[0].Qtype == dns.TypeA {
+			answer.Answer = []dns.RR{&dns.A{Hdr: dns.RR_Header{Name: query.Question[0].Name, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: 60}, A: net.IPv4(192, 0, 2, 20)}}
+		}
+		body, err := answer.Pack()
+		if err != nil {
+			t.Error(err)
+			return
+		}
+		w.Header().Set("Content-Type", "application/dns-message")
+		_, _ = w.Write(body)
+	}))
+	defer resolver.Close()
+	st, reg, _, err := BootstrapWorkers(context.Background(), Options{SeedDemo: true})
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() {
+		reg.RequestCancelAll()
+		ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+		defer cancel()
+		if e := reg.Drain(ctx); e != nil {
+			t.Error(e)
+		}
+	}()
+	tenant, _, _, _, _ := st.DemoIDs()
+	profile, err := st.CreateDohProfile(tenant, &store.DohProfile{Name: "explicit private resolver", URL: resolver.URL})
+	if err != nil {
+		t.Fatal(err)
+	}
+	mod, err := st.CreateModule(tenant, &store.Module{Name: "private-doh", Type: "DOMAINS", Enabled: true, DohProfileIDs: []string{profile.ID}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err = st.CreateDomainEntry(tenant, mod.ID, &store.DomainEntry{FQDN: "private-resolver.example.test"}); err != nil {
+		t.Fatal(err)
+	}
+	j, _, err := reg.Enqueue(tenant, jobs.KindModuleRefresh, nil, &mod.ID, map[string]any{"module_id": mod.ID})
+	if err != nil {
+		t.Fatal(err)
+	}
+	deadline := time.After(5 * time.Second)
+	ticker := time.NewTicker(5 * time.Millisecond)
+	defer ticker.Stop()
+	for {
+		snap := j.Snapshot()
+		status := snap["status"]
+		if status == jobs.StatusSucceeded {
+			break
+		}
+		if status == jobs.StatusFailed || status == jobs.StatusCancelled {
+			t.Fatalf("worker refresh failed: %+v", snap)
+		}
+		select {
+		case <-deadline:
+			t.Fatal("worker did not finish")
+		case <-ticker.C:
+		}
+	}
+	if calls.Load() < 2 {
+		t.Fatalf("bootstrap worker blocked private DoH: %d requests", calls.Load())
+	}
+	// The same general-client choice must not remove CDN's independent ACL.
+	cdn := httptest.NewTLSServer(http.HandlerFunc(func(http.ResponseWriter, *http.Request) { t.Error("private CDN reached") }))
+	defer cdn.Close()
+	req, err := http.NewRequest(http.MethodGet, cdn.URL, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err = pipeline.UpstreamHTTPDo(context.Background(), httpclient.New(httpclient.DefaultTimeout), req); err == nil {
+		t.Fatal("private CDN accepted")
+	}
+}

internal/httpapi/cdn_preview_body_test.go

@@ -0,0 +1,81 @@
+package httpapi
+
+import (
+	"context"
+	"encoding/json"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"evobgp/internal/store"
+)
+
+func TestPreviewCDNSourceRejectsOversizedBody(t *testing.T) {
+	t.Setenv("EVOBGP_CDN_ALLOW_PRIVATE", "1")
+	srv := httptest.NewTLSServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		_, _ = io.Copy(w, strings.NewReader(strings.Repeat("x", (8<<20)+1)))
+	}))
+	defer srv.Close()
+
+	st := store.NewMemory()
+	st.SeedDemo()
+	tenant, _, _, _, _ := st.DemoIDs()
+	mod, err := st.CreateModule(tenant, &store.Module{Type: "CDN_CIDRS", Name: "oversized-preview", Enabled: true})
+	if err != nil {
+		t.Fatal(err)
+	}
+	s := &Server{store: st, cdnHTTP: srv.Client()}
+	body, err := json.Marshal(map[string]string{"url": srv.URL, "source_kind": "plain"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	req := httptest.NewRequest(http.MethodPost, "/v1/modules/"+mod.ID+"/cdn-sources/preview", strings.NewReader(string(body)))
+	req.SetPathValue("module_id", mod.ID)
+	req = req.WithContext(context.WithValue(req.Context(), authCtxKey, Auth{TenantID: tenant, Role: "editor"}))
+	recorder := httptest.NewRecorder()
+	s.handlePreviewCDNSource(recorder, req)
+	if recorder.Code != http.StatusBadGateway {
+		t.Fatalf("oversized preview status=%d body=%s", recorder.Code, recorder.Body.String())
+	}
+}
+
+func TestPreviewCDNErrorBodyBoundAndQuotaBeforeFetch(t *testing.T) {
+	t.Setenv("EVOBGP_CDN_ALLOW_PRIVATE", "1")
+	t.Setenv("EVOBGP_API_CDN_PREVIEW_PER_MINUTE", "1")
+	m := store.NewMemory()
+	m.SeedDemo()
+	tenant, module, _, _, _ := m.DemoIDs()
+	var calls atomic.Int32
+	upstream := httptest.NewTLSServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		calls.Add(1)
+		w.WriteHeader(http.StatusNotFound)
+		_, _ = io.WriteString(w, strings.Repeat("x", 64<<10))
+		w.(http.Flusher).Flush()
+		// No EOF: an unbounded drain would wait until the caller deadline.
+		<-r.Context().Done()
+	}))
+	defer upstream.Close()
+	s := &Server{store: m, cdnHTTP: upstream.Client()}
+	request := func() *http.Request {
+		r := httptest.NewRequest(http.MethodPost, "/v1/modules/"+module+"/cdn-sources/preview", strings.NewReader(`{"url":"`+upstream.URL+`","source_kind":"plain"}`))
+		r.SetPathValue("module_id", module)
+		return r.WithContext(context.WithValue(r.Context(), authCtxKey, Auth{TenantID: tenant, Role: "editor"}))
+	}
+	first := httptest.NewRecorder()
+	firstRequest := request()
+	ctx, cancel := context.WithTimeout(firstRequest.Context(), 5*time.Second)
+	defer cancel()
+	s.handlePreviewCDNSource(first, firstRequest.WithContext(ctx))
+	if first.Code != http.StatusBadGateway || calls.Load() != 1 || ctx.Err() != nil {
+		t.Fatalf("unbounded drain status=%d calls=%d ctx=%v", first.Code, calls.Load(), ctx.Err())
+	}
+	second := httptest.NewRecorder()
+	s.handlePreviewCDNSource(second, request())
+	if second.Code != http.StatusTooManyRequests || calls.Load() != 1 || second.Header().Get("Retry-After") == "" {
+		t.Fatalf("quota fetch bypass status=%d calls=%d", second.Code, calls.Load())
+	}
+}

internal/httpapi/cors.go

@@ -39,6 +39,7 @@ func (s *Server) withCORS(h http.Handler) http.Handler {
 		}
 		w.Header().Set("Access-Control-Allow-Headers", "Authorization, Content-Type, Idempotency-Key, Accept, X-Tenant-Id")
 		w.Header().Set("Access-Control-Allow-Methods", "GET, POST, PATCH, DELETE, OPTIONS")
+		w.Header().Set("Access-Control-Expose-Headers", "Location, X-EvoBGP-Job-Status, Retry-After")
 		if r.Method == http.MethodOptions {
 			w.WriteHeader(http.StatusNoContent)
 			return

internal/httpapi/job_errors.go

@@ -0,0 +1,56 @@
+package httpapi
+
+import (
+	"errors"
+	"net/http"
+
+	"evobgp/internal/jobs"
+	"evobgp/internal/store"
+)
+
+func (s *Server) replayJobResponse(w http.ResponseWriter, r *http.Request, in store.JobEnqueue) bool {
+	j, found, err := s.jobs.ReplayContext(r.Context(), in)
+	if err != nil {
+		writeJobEnqueueError(w, err)
+		return true
+	}
+	if !found {
+		return false
+	}
+	w.Header().Set("Location", "/v1/jobs/"+j.ID)
+	snap := j.Snapshot()
+	body := map[string]any{"job_id": j.ID, "status": snap["status"]}
+	if j.Kind == jobs.KindTenantRefresh {
+		ids := snap["meta"].(map[string]any)["module_ids"]
+		n := 0
+		switch v := ids.(type) {
+		case []any:
+			n = len(v)
+		case []string:
+			n = len(v)
+		}
+		body["module_ids"] = ids
+		body["modules_due"] = n
+	}
+	writeJSON(w, http.StatusAccepted, body)
+	return true
+}
+
+func writeJobEnqueueError(w http.ResponseWriter, err error) {
+	switch {
+	case errors.Is(err, store.ErrJobConflict):
+		writeProblem(w, http.StatusConflict, "Conflict", "idempotency key belongs to a different job request")
+	case errors.Is(err, store.ErrJobQueueFull):
+		writeProblem(w, http.StatusTooManyRequests, "Too Many Requests", "job queue is full")
+	default:
+		writeInternalError(w, "job_enqueue", err)
+	}
+}
+
+func writeJobReadError(w http.ResponseWriter, err error) {
+	if errors.Is(err, jobs.ErrNotFound) || errors.Is(err, store.ErrJobNotFound) {
+		writeProblem(w, http.StatusNotFound, "Not Found", "job not found")
+		return
+	}
+	writeInternalError(w, "job_read", err)
+}

internal/httpapi/job_queue_test.go

@@ -0,0 +1,99 @@
+package httpapi
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"evobgp/internal/jobs"
+	"evobgp/internal/store"
+)
+
+type unavailableJobQueue struct{ store.JobQueue }
+
+func (q unavailableJobQueue) EnqueueJob(context.Context, store.JobEnqueue) (store.JobRecord, bool, error) {
+	return store.JobRecord{}, false, errors.New("database unavailable")
+}
+
+func TestJobEnqueueFailureNeverAccepts(t *testing.T) {
+	r := jobs.NewRegistry(nil)
+	r.UseDurableQueue(unavailableJobQueue{})
+	s := &Server{jobs: r}
+	req := httptest.NewRequest(http.MethodPost, "/v1/bird/reload", nil)
+	req = req.WithContext(context.WithValue(req.Context(), authCtxKey, Auth{TenantID: "tenant", Role: "operator"}))
+	w := httptest.NewRecorder()
+	s.handleBirdReload(w, req)
+	if w.Code == http.StatusAccepted || w.Code < 500 {
+		t.Fatalf("DB failure returned %d: %s", w.Code, w.Body.String())
+	}
+}
+
+func TestAutomaticEnqueueFailureIsVisible(t *testing.T) {
+	r := jobs.NewRegistry(nil)
+	r.UseDurableQueue(unavailableJobQueue{})
+	s := &Server{jobs: r}
+	w := httptest.NewRecorder()
+	req := httptest.NewRequest(http.MethodPost, "/v1/peers", nil)
+	s.enqueuePeerReconcile(w, req, "tenant", "peer_create")
+	if got := w.Header().Get("X-EvoBGP-Job-Status"); got != "enqueue-failed" {
+		t.Fatalf("scheduling failure hidden: %q", got)
+	}
+}
+
+type selectorBackend struct {
+	store.Backend
+	modules []*store.Module
+	reads   int
+}
+
+func (b *selectorBackend) ListModules(string) []*store.Module { b.reads++; return b.modules }
+
+func TestTenantRefreshReplayBeforeDynamicSelection(t *testing.T) {
+	m := store.NewMemory()
+	b := &selectorBackend{Backend: m, modules: []*store.Module{{ID: "module-a", Enabled: true, RefreshIntervalSec: 1}}}
+	r := jobs.NewRegistry(nil)
+	r.UseDurableQueue(m)
+	s := &Server{store: b, jobs: r}
+	request := func() *httptest.ResponseRecorder {
+		req := httptest.NewRequest(http.MethodPost, "/v1/tenant/refresh", strings.NewReader(`{}`))
+		req.Header.Set("Idempotency-Key", "same-request")
+		req = req.WithContext(context.WithValue(req.Context(), authCtxKey, Auth{TenantID: "tenant", Role: "editor"}))
+		w := httptest.NewRecorder()
+		s.handleTenantRefresh(w, req)
+		return w
+	}
+	first := request()
+	if first.Code != http.StatusAccepted {
+		t.Fatalf("first=%d %s", first.Code, first.Body.String())
+	}
+	var accepted map[string]any
+	if err := json.Unmarshal(first.Body.Bytes(), &accepted); err != nil {
+		t.Fatal(err)
+	}
+	claimed, err := m.ClaimJobs(context.Background(), "worker", 1, time.Minute)
+	if err != nil || len(claimed) != 1 {
+		t.Fatalf("claim %+v %v", claimed, err)
+	}
+	claimed[0].Status = "succeeded"
+	if err = m.FinishJob(context.Background(), claimed[0]); err != nil {
+		t.Fatal(err)
+	}
+	b.modules = nil
+	reads := b.reads
+	replay := request()
+	if replay.Code != http.StatusAccepted {
+		t.Fatalf("replay=%d %s", replay.Code, replay.Body.String())
+	}
+	var got map[string]any
+	if err = json.Unmarshal(replay.Body.Bytes(), &got); err != nil {
+		t.Fatal(err)
+	}
+	if got["job_id"] != accepted["job_id"] || got["status"] != "succeeded" || b.reads != reads {
+		t.Fatalf("replay re-resolved selector: %+v reads %d->%d", got, reads, b.reads)
+	}
+}

internal/httpapi/job_report_test.go

@@ -0,0 +1,114 @@
+package httpapi
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	"evobgp/internal/jobs"
+	"evobgp/internal/store"
+)
+
+func TestJobReportDurableProgressAndTenantScope(t *testing.T) {
+	ctx := context.Background()
+	m := store.NewMemory()
+	m.SeedDemo()
+	tenant, _, _, revision, speaker := m.DemoIDs()
+	j, _, err := m.EnqueueJob(ctx, store.JobEnqueue{TenantID: tenant, Kind: jobs.KindDeployApply})
+	if err != nil {
+		t.Fatal(err)
+	}
+	claimed, err := m.ClaimJobs(ctx, "worker", 1, time.Minute)
+	if err != nil || len(claimed) != 1 {
+		t.Fatalf("claim: %v %v", claimed, err)
+	}
+	f := store.JobFence{JobID: j.ID, TenantID: tenant, OwnerID: "worker", Token: claimed[0].FencingToken}
+	d, err := m.BeginDeploymentSpeaker(ctx, f, speaker, revision)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := m.RecordDeploymentSpeaker(ctx, f, d, "unknown", "unconfirmed", "response lost"); err != nil {
+		t.Fatal(err)
+	}
+	// A fresh registry has no in-process job or progress metadata.
+	r := jobs.NewRegistry(nil)
+	r.UseDurableQueue(m)
+	s := &Server{store: m, jobs: r}
+	request := func(tenantID, jobID string) *httptest.ResponseRecorder {
+		t.Helper()
+		req := httptest.NewRequest(http.MethodGet, "/v1/jobs/"+jobID+"/report", nil)
+		req.SetPathValue("job_id", jobID)
+		req = req.WithContext(context.WithValue(req.Context(), authCtxKey, Auth{TenantID: tenantID, Role: "operator"}))
+		w := httptest.NewRecorder()
+		s.handleGetJobReport(w, req)
+		return w
+	}
+	check := func(id string) {
+		t.Helper()
+		w := request(tenant, id)
+		if w.Code != http.StatusOK {
+			t.Fatalf("report: %d %s", w.Code, w.Body.String())
+		}
+		var body struct {
+			Progress []struct {
+				State, Confirmation string
+				JobID               string `json:"job_id"`
+			} `json:"deployment_progress"`
+		}
+		if err := json.Unmarshal(w.Body.Bytes(), &body); err != nil {
+			t.Fatal(err)
+		}
+		if len(body.Progress) != 1 || body.Progress[0].JobID != j.ID || body.Progress[0].State != "unknown" || body.Progress[0].Confirmation != "unconfirmed" {
+			t.Fatalf("lost persisted progress: %s", w.Body.String())
+		}
+	}
+	check(j.ID)
+	recovery, _, err := r.EnqueueContext(ctx, tenant, jobs.KindDeployApply, nil, nil, map[string]any{"source_deployment_job_id": j.ID})
+	if err != nil {
+		t.Fatal(err)
+	}
+	check(recovery.ID)
+	// A later explicit completion releases the delivery reservation. Once recovery
+	// has its own persisted row, its report must not show the parent's old state.
+	if err := m.RecordDeploymentSpeaker(ctx, f, d, "unconfirmed", "unconfirmed", ""); err != nil {
+		t.Fatal(err)
+	}
+	recoveryRows, err := m.ClaimJobs(ctx, "recovery-worker", 1, time.Minute)
+	if err != nil || len(recoveryRows) != 1 || recoveryRows[0].ID != recovery.ID {
+		t.Fatalf("recovery claim: %v %v", recoveryRows, err)
+	}
+	rf := store.JobFence{JobID: recovery.ID, TenantID: tenant, OwnerID: "recovery-worker", Token: recoveryRows[0].FencingToken}
+	rd, err := m.BeginDeploymentSpeaker(ctx, rf, speaker, revision)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := m.RecordDeploymentSpeaker(ctx, rf, rd, "applied", "agent_reported", ""); err != nil {
+		t.Fatal(err)
+	}
+	w := request(tenant, recovery.ID)
+	var report struct {
+		Progress []struct {
+			State string
+			JobID string `json:"job_id"`
+		} `json:"deployment_progress"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &report); err != nil {
+		t.Fatal(err)
+	}
+	if w.Code != http.StatusOK || len(report.Progress) != 1 || report.Progress[0].JobID != recovery.ID || report.Progress[0].State != "applied" {
+		t.Fatalf("recovery own progress hidden: %d %s", w.Code, w.Body.String())
+	}
+	if w := request("other-tenant", j.ID); w.Code != http.StatusNotFound {
+		t.Fatalf("cross-tenant report: %d", w.Code)
+	}
+	foreign, _, err := r.EnqueueContext(ctx, "other-tenant", jobs.KindDeployApply, nil, nil, map[string]any{"source_deployment_job_id": j.ID})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if w := request("other-tenant", foreign.ID); w.Code != http.StatusNotFound {
+		t.Fatalf("cross-tenant source: %d %s", w.Code, w.Body.String())
+	}
+}

internal/httpapi/legacy_node_binary_test.go

@@ -0,0 +1,89 @@
+package httpapi
+
+import (
+	"context"
+	"crypto/ed25519"
+	"encoding/base64"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"net/http/httptest"
+
+	"evobgp/internal/bundle"
+)
+
+// TestLegacyNodeBinaryCurrentControlPlane exercises an independently built old
+// CLI against today's HTTP handlers. It never calls apply or touches host BIRD.
+func TestLegacyNodeBinaryCurrentControlPlane(t *testing.T) {
+	binary := os.Getenv("EVOBGP_TEST_LEGACY_NODE_BIN")
+	if binary == "" {
+		t.Skip("EVOBGP_TEST_LEGACY_NODE_BIN not set")
+	}
+	binary, err := filepath.Abs(binary)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := os.Stat(binary); err != nil {
+		t.Fatal(err)
+	}
+	srv, err := New(Options{SeedDemo: true, BundleSeedHex: testBundleSeed})
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer srv.Close()
+	tenant, _, _, revision, speaker := srv.Store().DemoIDs()
+	mustSetTestAPIKeys(t, srv, "legacy-node|"+tenant+"|node,other-node|other-tenant|node")
+	if err := srv.Store().PublishRevisionForSpeaker(speaker, revision); err != nil {
+		t.Fatal(err)
+	}
+	server := httptest.NewServer(srv.Handler())
+	defer server.Close()
+	run := func(wantSuccess bool, args ...string) {
+		t.Helper()
+		ctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)
+		defer cancel()
+		output, err := exec.CommandContext(ctx, binary, args...).CombinedOutput()
+		if ctx.Err() != nil {
+			t.Fatalf("legacy CLI timed out: %v", ctx.Err())
+		}
+		if wantSuccess && err != nil {
+			t.Fatalf("legacy CLI failed: %v: %s", err, output)
+		}
+		if !wantSuccess && err == nil {
+			t.Fatalf("legacy CLI unexpectedly accepted invalid request: %s", output)
+		}
+	}
+	path := filepath.Join(t.TempDir(), "current-bundle.tar.gz")
+	run(true, "pull-bundle", "-base-url", server.URL, "-token", "legacy-node", "-speaker-id", speaker, "-o", path)
+	run(true, "verify-bundle", "-f", path, "-pubkey-base64", srv.BundlePublicKeyBase64())
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	pub, err := base64.StdEncoding.DecodeString(srv.BundlePublicKeyBase64())
+	if err != nil {
+		t.Fatal(err)
+	}
+	verified, err := bundle.VerifyGzippedTar(raw, ed25519.PublicKey(pub))
+	if err != nil || verified.Manifest.RevisionID != revision || verified.Manifest.SpeakerID != speaker {
+		t.Fatalf("legacy CLI fetched the wrong revision/speaker or an invalid bundle: %v", err)
+	}
+	// The archive is valid, but verification with an unrelated key must fail.
+	run(false, "verify-bundle", "-f", path, "-pubkey-hex", strings.Repeat("00", ed25519.PublicKeySize))
+	for _, token := range []string{"invalid-node", "other-node"} {
+		denied := filepath.Join(t.TempDir(), "denied.tar.gz")
+		run(false, "pull-bundle", "-base-url", server.URL, "-token", token, "-speaker-id", speaker, "-o", denied)
+		if _, err := os.Stat(denied); !os.IsNotExist(err) {
+			t.Fatalf("unauthorized CLI wrote a bundle: %v", err)
+		}
+	}
+	// A damaged payload must not pass the old verifier with the correct key.
+	if err := os.WriteFile(path, []byte("not a signed bundle"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	run(false, "verify-bundle", "-f", path, "-pubkey-base64", srv.BundlePublicKeyBase64())
+}

internal/httpapi/node_bundle_cache.go

@@ -0,0 +1,310 @@
+package httpapi
+
+import (
+	"context"
+	"crypto/ed25519"
+	"crypto/sha256"
+	"encoding/binary"
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"sort"
+	"strings"
+	"sync"
+
+	"evobgp/internal/bundle"
+	"evobgp/internal/pipeline"
+	"evobgp/internal/store"
+)
+
+const (
+	// nodeBundleCacheMaxEntries is the maximum number of finished node bundles retained.
+	nodeBundleCacheMaxEntries = 32
+	// nodeBundleCacheMaxBytes is the maximum total size of cached bundle bodies (64 MiB).
+	nodeBundleCacheMaxBytes = 64 << 20
+)
+
+// errNodeBundleRevisionMissing is returned when the revision cannot be loaded.
+// It is not cached: the next pull tries again.
+var errNodeBundleRevisionMissing = errors.New("httpapi: revision not found")
+
+// nodeBundleOpError carries the writeInternalError operation name for a failed build.
+type nodeBundleOpError struct {
+	op  string
+	err error
+}
+
+func (e *nodeBundleOpError) Error() string {
+	if e == nil || e.err == nil {
+		return "httpapi: node bundle"
+	}
+	return fmt.Sprintf("httpapi: %s: %v", e.op, e.err)
+}
+
+func (e *nodeBundleOpError) Unwrap() error { return e.err }
+
+// nodeBundleBuilder produces the gzip body served to evobgp-node.
+// The default implementation is bundle.BuildGzippedTar, whose archive contains
+// manifest.json and bundle.sig. Tests replace it to count rebuilds.
+type nodeBundleBuilder func(revisionID, speakerID string, fragments map[string]string, priv ed25519.PrivateKey) ([]byte, error)
+
+// nodeBundleCache stores finished gzip bodies. A hit returns the bytes of the
+// build that filled the entry, signature included. Errors are not stored.
+// Concurrent loads of one key share a single build.
+type nodeBundleCache struct {
+	mu       sync.Mutex
+	maxN     int
+	maxBytes int
+	total    int
+	items    map[string]*nodeBundleEntry
+	lru      []string // index 0 is the oldest idle entry
+	flights  map[string]*nodeBundleFlight
+}
+
+type nodeBundleEntry struct {
+	body []byte
+}
+
+type nodeBundleFlight struct {
+	done   chan struct{}
+	body   []byte
+	err    error
+	joined int
+}
+
+func newNodeBundleCache(maxEntries, maxBytes int) *nodeBundleCache {
+	return &nodeBundleCache{
+		maxN:     maxEntries,
+		maxBytes: maxBytes,
+		items:    make(map[string]*nodeBundleEntry),
+		flights:  make(map[string]*nodeBundleFlight),
+	}
+}
+
+// nodeBundleCacheKey identifies one servable bundle.
+// overlayFP is a digest of the speaker overlay output (router id / source addresses,
+// node IPv4 fallback, peer and discovery render). Revision content itself is immutable,
+// so the revision id stands in for the shared prefix files.
+func nodeBundleCacheKey(tenantID, revisionID, speakerID, overlayFP string) string {
+	var b strings.Builder
+	b.Grow(len(tenantID) + len(revisionID) + len(speakerID) + len(overlayFP) + 3)
+	b.WriteString(tenantID)
+	b.WriteByte(0)
+	b.WriteString(revisionID)
+	b.WriteByte(0)
+	b.WriteString(speakerID)
+	b.WriteByte(0)
+	b.WriteString(overlayFP)
+	return b.String()
+}
+
+// nodeBundleOverlayFingerprint hashes the speaker overlay that handleNodeBundle
+// would apply on top of an immutable revision. Empty base fragments are enough:
+// bird.conf and the peers fragment are fully re-rendered from tenant settings,
+// speaker source addresses, and the live peer set. Revision file bytes are not
+// part of this digest.
+func nodeBundleOverlayFingerprint(st store.Backend, tenantID, speakerID, revisionID string) (string, error) {
+	frags, err := pipeline.OverlayFragmentsForSpeaker(st, tenantID, speakerID, revisionID, nil)
+	if err != nil {
+		return "", fmt.Errorf("httpapi: bundle overlay: %w", err)
+	}
+	keys := make([]string, 0, len(frags))
+	for k := range frags {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	h := sha256.New()
+	for _, k := range keys {
+		if err := hashWriteString(h, k); err != nil {
+			return "", fmt.Errorf("httpapi: bundle overlay: %w", err)
+		}
+		if err := hashWriteString(h, frags[k]); err != nil {
+			return "", fmt.Errorf("httpapi: bundle overlay: %w", err)
+		}
+	}
+	return hex.EncodeToString(h.Sum(nil)), nil
+}
+
+func hashWriteString(h interface{ Write([]byte) (int, error) }, s string) error {
+	var buf [8]byte
+	binary.BigEndian.PutUint64(buf[:], uint64(len(s)))
+	if _, err := h.Write(buf[:]); err != nil {
+		return err
+	}
+	_, err := h.Write([]byte(s))
+	return err
+}
+
+// do returns the cached body or runs build once for key.
+// build's bool is false when the result must not be stored (failure is also not stored).
+// Callers must not mutate the returned slice.
+func (c *nodeBundleCache) do(ctx context.Context, key string, build func(context.Context) ([]byte, bool, error)) (body []byte, err error) {
+	if c == nil {
+		return nil, errors.New("httpapi: nil node bundle cache")
+	}
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	c.mu.Lock()
+	if e, ok := c.items[key]; ok {
+		c.touchLocked(key)
+		body = e.body
+		c.mu.Unlock()
+		return body, nil
+	}
+	if f, ok := c.flights[key]; ok {
+		f.joined++
+		c.mu.Unlock()
+		return waitNodeBundleFlight(ctx, f)
+	}
+	if err = ctx.Err(); err != nil {
+		c.mu.Unlock()
+		return nil, err
+	}
+	f := &nodeBundleFlight{done: make(chan struct{})}
+	c.flights[key] = f
+	c.mu.Unlock()
+
+	var cacheOK bool
+	var finished bool
+	finish := func() {
+		if finished {
+			return
+		}
+		finished = true
+		out := body
+		c.mu.Lock()
+		delete(c.flights, key)
+		if err == nil && cacheOK {
+			if stored, ok := c.storeLocked(key, out); ok {
+				out = stored
+			}
+		}
+		f.body = out
+		f.err = err
+		close(f.done)
+		c.mu.Unlock()
+		body = out
+	}
+	defer func() {
+		if rec := recover(); rec != nil {
+			body = nil
+			cacheOK = false
+			err = fmt.Errorf("httpapi: node bundle build: %v", rec)
+		}
+		finish()
+	}()
+
+	// Detach from the leader's cancel so one disconnected agent does not fail the shared build.
+	body, cacheOK, err = build(context.WithoutCancel(ctx))
+	return
+}
+
+func waitNodeBundleFlight(ctx context.Context, f *nodeBundleFlight) ([]byte, error) {
+	select {
+	case <-f.done:
+		return f.body, f.err
+	case <-ctx.Done():
+		select {
+		case <-f.done:
+			return f.body, f.err
+		default:
+			return nil, ctx.Err()
+		}
+	}
+}
+
+// storeLocked retains body when it fits under both caps, evicting the oldest idle
+// entries first. body is copied. The stored slice is returned.
+func (c *nodeBundleCache) storeLocked(key string, body []byte) ([]byte, bool) {
+	n := len(body)
+	if n == 0 || c.maxN <= 0 || c.maxBytes <= 0 || n > c.maxBytes {
+		return nil, false
+	}
+	if e, ok := c.items[key]; ok {
+		c.total -= len(e.body)
+		delete(c.items, key)
+		c.removeLRULocked(key)
+	}
+	for len(c.items) > 0 && (len(c.items) >= c.maxN || c.total+n > c.maxBytes) {
+		if len(c.lru) == 0 {
+			return nil, false
+		}
+		c.deleteOldestLocked()
+	}
+	if len(c.items) >= c.maxN || c.total+n > c.maxBytes {
+		return nil, false
+	}
+	stored := make([]byte, n)
+	copy(stored, body)
+	c.items[key] = &nodeBundleEntry{body: stored}
+	c.total += n
+	c.lru = append(c.lru, key)
+	return stored, true
+}
+
+func (c *nodeBundleCache) touchLocked(key string) {
+	c.removeLRULocked(key)
+	c.lru = append(c.lru, key)
+}
+
+func (c *nodeBundleCache) deleteOldestLocked() {
+	if len(c.lru) == 0 {
+		return
+	}
+	key := c.lru[0]
+	c.lru = c.lru[1:]
+	if e, ok := c.items[key]; ok {
+		c.total -= len(e.body)
+		delete(c.items, key)
+	}
+}
+
+func (c *nodeBundleCache) removeLRULocked(key string) {
+	for i, k := range c.lru {
+		if k != key {
+			continue
+		}
+		copy(c.lru[i:], c.lru[i+1:])
+		c.lru = c.lru[:len(c.lru)-1]
+		return
+	}
+}
+
+func (s *Server) buildNodeBundle(revisionID, speakerID string, fragments map[string]string) ([]byte, error) {
+	if s != nil && s.nodeBundleBuild != nil {
+		return s.nodeBundleBuild(revisionID, speakerID, fragments, s.bundlePriv)
+	}
+	return bundle.BuildGzippedTar(revisionID, speakerID, fragments, s.bundlePriv)
+}
+
+// loadNodeBundle reads the revision, applies the speaker overlay, and signs the gzip.
+// The bool is false when overlay inputs moved during the load, so the caller must not
+// cache the body under the key captured before the load. Failed loads are not cacheable.
+func (s *Server) loadNodeBundle(ctx context.Context, tenantID, speakerID, revisionID, overlayFP string) ([]byte, bool, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, false, err
+	}
+	fpNow, err := nodeBundleOverlayFingerprint(s.store, tenantID, speakerID, revisionID)
+	if err != nil {
+		return nil, false, &nodeBundleOpError{op: "bundle overlay", err: err}
+	}
+	rev, err := s.store.GetRevision(tenantID, revisionID)
+	if err != nil {
+		return nil, false, errNodeBundleRevisionMissing
+	}
+	overlaid, err := pipeline.OverlayFragmentsForSpeaker(s.store, tenantID, speakerID, revisionID, rev.PreviewFragments)
+	if err != nil {
+		return nil, false, &nodeBundleOpError{op: "bundle overlay", err: err}
+	}
+	fpAfter, err := nodeBundleOverlayFingerprint(s.store, tenantID, speakerID, revisionID)
+	if err != nil {
+		return nil, false, &nodeBundleOpError{op: "bundle overlay", err: err}
+	}
+	body, err := s.buildNodeBundle(revisionID, speakerID, overlaid)
+	if err != nil {
+		return nil, false, &nodeBundleOpError{op: "internal", err: err}
+	}
+	cacheable := fpNow == overlayFP && fpAfter == fpNow && len(body) > 0
+	return body, cacheable, nil
+}

internal/httpapi/node_bundle_cache_test.go

@@ -0,0 +1,791 @@
+package httpapi
+
+import (
+	"archive/tar"
+	"bytes"
+	"compress/gzip"
+	"context"
+	"crypto/ed25519"
+	"encoding/base64"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"evobgp/internal/bundle"
+	"evobgp/internal/signing"
+	"evobgp/internal/store"
+)
+
+func TestNodeBundleCacheKeySeparatesTenantsRevisionsSpeakers(t *testing.T) {
+	a := nodeBundleCacheKey("tenant-a", "rev", "spk", "overlay")
+	if a == nodeBundleCacheKey("tenant-b", "rev", "spk", "overlay") {
+		t.Fatal("tenants collided")
+	}
+	if a == nodeBundleCacheKey("tenant-a", "rev-2", "spk", "overlay") {
+		t.Fatal("revisions collided")
+	}
+	if a == nodeBundleCacheKey("tenant-a", "rev", "spk-2", "overlay") {
+		t.Fatal("speakers collided")
+	}
+	if a == nodeBundleCacheKey("tenant-a", "rev", "spk", "overlay-2") {
+		t.Fatal("overlay inputs collided")
+	}
+}
+
+func TestNodeBundleCacheHitSingleflightAndErrors(t *testing.T) {
+	c := newNodeBundleCache(8, 1<<20)
+	var builds atomic.Int32
+	_, err := c.do(context.Background(), "k", func(context.Context) ([]byte, bool, error) {
+		builds.Add(1)
+		return []byte("signed-body"), true, nil
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	got, err := c.do(context.Background(), "k", func(context.Context) ([]byte, bool, error) {
+		builds.Add(1)
+		return []byte("other"), true, nil
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if builds.Load() != 1 {
+		t.Fatalf("second pull rebuilt: builds=%d", builds.Load())
+	}
+	if !bytes.Equal(got, []byte("signed-body")) {
+		t.Fatalf("hit body = %q", got)
+	}
+
+	t.Run("singleflight", func(t *testing.T) {
+		c := newNodeBundleCache(8, 1<<20)
+		var builds atomic.Int32
+		release := make(chan struct{})
+		const n = 8
+		var wg sync.WaitGroup
+		bodies := make([][]byte, n)
+		errs := make([]error, n)
+		wg.Add(n)
+		for i := 0; i < n; i++ {
+			go func(i int) {
+				defer wg.Done()
+				bodies[i], errs[i] = c.do(context.Background(), "same", func(context.Context) ([]byte, bool, error) {
+					builds.Add(1)
+					<-release
+					return []byte("signed-bundle"), true, nil
+				})
+			}(i)
+		}
+		deadline := time.Now().Add(2 * time.Second)
+		for {
+			c.mu.Lock()
+			joined := 0
+			if f := c.flights["same"]; f != nil {
+				joined = f.joined
+			}
+			c.mu.Unlock()
+			if builds.Load() > 1 {
+				close(release)
+				wg.Wait()
+				t.Fatalf("builds = %d before followers joined", builds.Load())
+			}
+			if builds.Load() == 1 && joined == n-1 {
+				break
+			}
+			if time.Now().After(deadline) {
+				close(release)
+				wg.Wait()
+				t.Fatalf("builds=%d joined=%d", builds.Load(), joined)
+			}
+			time.Sleep(time.Millisecond)
+		}
+		close(release)
+		wg.Wait()
+		if builds.Load() != 1 {
+			t.Fatalf("builds = %d, want 1", builds.Load())
+		}
+		for i := 0; i < n; i++ {
+			if errs[i] != nil {
+				t.Fatalf("waiter %d: %v", i, errs[i])
+			}
+			if !bytes.Equal(bodies[i], []byte("signed-bundle")) {
+				t.Fatalf("waiter %d body = %q", i, bodies[i])
+			}
+		}
+	})
+
+	t.Run("canceled waiter does not drop the build", func(t *testing.T) {
+		c := newNodeBundleCache(4, 1024)
+		release := make(chan struct{})
+		var builds atomic.Int32
+		leaderErr := make(chan error, 1)
+		go func() {
+			_, err := c.do(context.Background(), "k", func(context.Context) ([]byte, bool, error) {
+				builds.Add(1)
+				<-release
+				return []byte("body"), true, nil
+			})
+			leaderErr <- err
+		}()
+		deadline := time.Now().Add(2 * time.Second)
+		for {
+			c.mu.Lock()
+			_, ok := c.flights["k"]
+			c.mu.Unlock()
+			if ok || time.Now().After(deadline) {
+				break
+			}
+			time.Sleep(time.Millisecond)
+		}
+		ctx, cancel := context.WithCancel(context.Background())
+		cancel()
+		_, err := c.do(ctx, "k", func(context.Context) ([]byte, bool, error) {
+			builds.Add(1)
+			return []byte("nope"), true, nil
+		})
+		if !errors.Is(err, context.Canceled) {
+			t.Fatalf("follower err = %v", err)
+		}
+		close(release)
+		if err := <-leaderErr; err != nil {
+			t.Fatal(err)
+		}
+		if builds.Load() != 1 {
+			t.Fatalf("builds = %d", builds.Load())
+		}
+		_, err = c.do(context.Background(), "k", func(context.Context) ([]byte, bool, error) {
+			builds.Add(1)
+			return []byte("other"), true, nil
+		})
+		if err != nil {
+			t.Fatal(err)
+		}
+		if builds.Load() != 1 {
+			t.Fatal("successful build was not cached after a canceled waiter")
+		}
+	})
+
+	t.Run("errors are not cached", func(t *testing.T) {
+		c := newNodeBundleCache(4, 1024)
+		calls := 0
+		boom := errors.New("httpapi: bundle build failed")
+		_, err := c.do(context.Background(), "k", func(context.Context) ([]byte, bool, error) {
+			calls++
+			return nil, true, boom
+		})
+		if !errors.Is(err, boom) {
+			t.Fatalf("err = %v", err)
+		}
+		_, err = c.do(context.Background(), "miss", func(context.Context) ([]byte, bool, error) {
+			calls++
+			return nil, false, errNodeBundleRevisionMissing
+		})
+		if !errors.Is(err, errNodeBundleRevisionMissing) {
+			t.Fatalf("err = %v", err)
+		}
+		got, err := c.do(context.Background(), "k", func(context.Context) ([]byte, bool, error) {
+			calls++
+			return []byte("ok"), true, nil
+		})
+		if err != nil {
+			t.Fatal(err)
+		}
+		if string(got) != "ok" || calls != 3 {
+			t.Fatalf("calls=%d body=%q", calls, got)
+		}
+		_, err = c.do(context.Background(), "k", func(context.Context) ([]byte, bool, error) {
+			calls++
+			return []byte("other"), true, nil
+		})
+		if err != nil || calls != 3 {
+			t.Fatalf("cached error path rebuilt: calls=%d err=%v", calls, err)
+		}
+	})
+
+	t.Run("unstorable success is not cached", func(t *testing.T) {
+		c := newNodeBundleCache(4, 1024)
+		calls := 0
+		for i := 0; i < 2; i++ {
+			got, err := c.do(context.Background(), "k", func(context.Context) ([]byte, bool, error) {
+				calls++
+				return []byte("fresh"), false, nil
+			})
+			if err != nil || string(got) != "fresh" {
+				t.Fatalf("i=%d err=%v body=%q", i, err, got)
+			}
+		}
+		if calls != 2 {
+			t.Fatalf("calls = %d, want 2", calls)
+		}
+	})
+}
+
+func TestNodeBundleCacheEvictsIdleAndRespectsCaps(t *testing.T) {
+	t.Run("entry cap evicts oldest idle", func(t *testing.T) {
+		c := newNodeBundleCache(2, 100)
+		put := func(key, body string) {
+			t.Helper()
+			got, err := c.do(context.Background(), key, func(context.Context) ([]byte, bool, error) {
+				return []byte(body), true, nil
+			})
+			if err != nil || string(got) != body {
+				t.Fatalf("put %s: %v %q", key, err, got)
+			}
+		}
+		put("a", "A")
+		put("b", "B")
+		if _, err := c.do(context.Background(), "a", func(context.Context) ([]byte, bool, error) {
+			t.Fatal("recent entry was rebuilt")
+			return nil, false, nil
+		}); err != nil {
+			t.Fatal(err)
+		}
+		put("c", "C")
+		assertBundleCacheAccounting(t, c)
+		if _, ok := c.items["b"]; ok {
+			t.Fatal("idle entry b was kept")
+		}
+		if _, ok := c.items["a"]; !ok || c.items["c"] == nil {
+			t.Fatalf("kept %v", cacheKeys(c))
+		}
+		if len(c.items) > 2 {
+			t.Fatalf("entry cap exceeded: %d", len(c.items))
+		}
+	})
+
+	t.Run("byte cap", func(t *testing.T) {
+		c := newNodeBundleCache(8, 10)
+		put := func(key string, n int) {
+			t.Helper()
+			body := bytes.Repeat([]byte("x"), n)
+			got, err := c.do(context.Background(), key, func(context.Context) ([]byte, bool, error) {
+				return body, true, nil
+			})
+			if err != nil || !bytes.Equal(got, body) {
+				t.Fatalf("put %s: %v len=%d", key, err, len(got))
+			}
+			assertBundleCacheAccounting(t, c)
+		}
+		put("a", 4)
+		put("b", 4)
+		if _, err := c.do(context.Background(), "a", func(context.Context) ([]byte, bool, error) {
+			t.Fatal("rebuilt a")
+			return nil, false, nil
+		}); err != nil {
+			t.Fatal(err)
+		}
+		put("c", 4)
+		assertBundleCacheAccounting(t, c)
+		if _, ok := c.items["b"]; ok {
+			t.Fatal("byte cap kept idle entry b")
+		}
+		if len(c.items) != 2 || c.total != 8 {
+			t.Fatalf("n=%d total=%d", len(c.items), c.total)
+		}
+		put("d", 6)
+		put("e", 6)
+		assertBundleCacheAccounting(t, c)
+		if len(c.items) != 1 || c.total != 6 {
+			t.Fatalf("after oversized pair n=%d total=%d keys=%v", len(c.items), c.total, cacheKeys(c))
+		}
+	})
+
+	t.Run("entry larger than the byte cap is not stored", func(t *testing.T) {
+		c := newNodeBundleCache(4, 8)
+		calls := 0
+		big := bytes.Repeat([]byte("y"), 9)
+		for i := 0; i < 2; i++ {
+			got, err := c.do(context.Background(), "big", func(context.Context) ([]byte, bool, error) {
+				calls++
+				return big, true, nil
+			})
+			if err != nil || !bytes.Equal(got, big) {
+				t.Fatalf("i=%d err=%v", i, err)
+			}
+		}
+		if calls != 2 {
+			t.Fatalf("oversize body was cached, calls=%d", calls)
+		}
+		if len(c.items) != 0 || c.total != 0 {
+			t.Fatalf("stored oversize n=%d total=%d", len(c.items), c.total)
+		}
+	})
+}
+
+func TestNodeBundleOverlayFingerprintInputs(t *testing.T) {
+	m := store.NewMemory()
+	m.SeedDemo()
+	tenant, _, _, rev, speaker := m.DemoIDs()
+
+	base, err := nodeBundleOverlayFingerprint(m, tenant, speaker, rev)
+	if err != nil {
+		t.Fatal(err)
+	}
+	again, err := nodeBundleOverlayFingerprint(m, tenant, speaker, rev)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if base != again {
+		t.Fatal("fingerprint changed without an overlay input change")
+	}
+
+	sp, err := m.GetSpeaker(tenant, speaker)
+	if err != nil {
+		t.Fatal(err)
+	}
+	dispatch := store.MergeSpeakerMetaJSON(sp.MetaJSON, store.SpeakerMeta{
+		LastDispatchAt:     "2026-01-01T00:00:00Z",
+		LastDispatchStatus: "ok",
+		SyncStatus:         "synced",
+	})
+	if _, err := m.UpdateSpeaker(tenant, speaker, &store.SpeakerPatch{MetaJSON: &dispatch}); err != nil {
+		t.Fatal(err)
+	}
+	if got, err := nodeBundleOverlayFingerprint(m, tenant, speaker, rev); err != nil || got != base {
+		t.Fatalf("dispatch meta changed fingerprint: %v %s", err, got)
+	}
+
+	peers := m.ListPeers(tenant)
+	if len(peers) == 0 {
+		t.Fatal("expected demo peers")
+	}
+	state := "Active"
+	if _, err := m.UpdatePeer(tenant, peers[0].ID, &store.PeerPatch{SessionState: &state}); err != nil {
+		t.Fatal(err)
+	}
+	if got, err := nodeBundleOverlayFingerprint(m, tenant, speaker, rev); err != nil || got != base {
+		t.Fatalf("session state changed fingerprint: %v %s", err, got)
+	}
+
+	if err := m.PatchGlobalSettings(tenant, map[string]any{"bird_router_id": "203.0.113.7"}); err != nil {
+		t.Fatal(err)
+	}
+	routed, err := nodeBundleOverlayFingerprint(m, tenant, speaker, rev)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if routed == base {
+		t.Fatal("router id setting did not change the overlay fingerprint")
+	}
+
+	nodeMeta := store.SpeakerMetaJSON(store.SpeakerMeta{NodeIPv4: "198.51.100.9"})
+	if _, err := m.UpdateSpeaker(tenant, speaker, &store.SpeakerPatch{MetaJSON: &nodeMeta}); err != nil {
+		t.Fatal(err)
+	}
+	nodeFP, err := nodeBundleOverlayFingerprint(m, tenant, speaker, rev)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if nodeFP == routed {
+		t.Fatal("node ipv4 did not change the overlay fingerprint")
+	}
+
+	sourceMeta := store.SpeakerMetaJSON(store.SpeakerMeta{
+		NodeIPv4:          "198.51.100.9",
+		BirdBgpSourceIPv4: "203.0.113.40",
+	})
+	if _, err := m.UpdateSpeaker(tenant, speaker, &store.SpeakerPatch{MetaJSON: &sourceMeta}); err != nil {
+		t.Fatal(err)
+	}
+	sourceFP, err := nodeBundleOverlayFingerprint(m, tenant, speaker, rev)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if sourceFP == nodeFP {
+		t.Fatal("source ipv4 did not change the overlay fingerprint")
+	}
+
+	nodeOnly := store.SpeakerMetaJSON(store.SpeakerMeta{
+		NodeIPv4:          "198.51.100.8",
+		BirdBgpSourceIPv4: "203.0.113.40",
+	})
+	if _, err := m.UpdateSpeaker(tenant, speaker, &store.SpeakerPatch{MetaJSON: &nodeOnly}); err != nil {
+		t.Fatal(err)
+	}
+	if got, err := nodeBundleOverlayFingerprint(m, tenant, speaker, rev); err != nil || got != sourceFP {
+		t.Fatalf("unused node ipv4 changed fingerprint: %v %s vs %s", err, got, sourceFP)
+	}
+
+	v6 := store.SpeakerMetaJSON(store.SpeakerMeta{
+		NodeIPv4:          "198.51.100.8",
+		BirdBgpSourceIPv4: "203.0.113.40",
+		BirdBgpSourceIPv6: "2001:db8::40",
+	})
+	if _, err := m.UpdateSpeaker(tenant, speaker, &store.SpeakerPatch{MetaJSON: &v6}); err != nil {
+		t.Fatal(err)
+	}
+	v6FP, err := nodeBundleOverlayFingerprint(m, tenant, speaker, rev)
+	if err != nil {
+		t.Fatal(err)
+	}
+	// Source v6 changes local v6. When no peer policy overrides it, the rendered
+	// peers fragment can stay the same. A difference means the overlay emitted it;
+	// equality means a rebuild would not change the bundle.
+	_ = v6FP
+
+	if _, err := m.CreatePeer(tenant, &store.BGPPeer{
+		Name:      "cache-extra",
+		Neighbor:  "203.0.113.50",
+		RemoteASN: 64512,
+		Enabled:   true,
+	}); err != nil {
+		t.Fatal(err)
+	}
+	withPeer, err := nodeBundleOverlayFingerprint(m, tenant, speaker, rev)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if withPeer == v6FP {
+		t.Fatal("added peer did not change the overlay fingerprint")
+	}
+
+	sp2, err := m.CreateSpeaker(tenant, &store.Speaker{
+		Role:     "replica",
+		Endpoint: "https://203.0.113.2",
+		MetaJSON: v6,
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	other, err := nodeBundleOverlayFingerprint(m, tenant, sp2.ID, rev)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if other == withPeer {
+		t.Fatal("speakers with the same addresses shared an overlay fingerprint")
+	}
+
+	otherRev, err := m.CreateRollbackRevision(tenant, rev)
+	if err != nil {
+		t.Fatal(err)
+	}
+	revFP, err := nodeBundleOverlayFingerprint(m, tenant, speaker, otherRev)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if revFP == withPeer {
+		t.Fatal("revision id did not change the overlay fingerprint")
+	}
+}
+
+func TestHandleNodeBundleCache(t *testing.T) {
+	srv, ts, ids := startNodeBundleServer(t)
+	if srv.bundleCache.maxN != nodeBundleCacheMaxEntries || srv.bundleCache.maxBytes != nodeBundleCacheMaxBytes {
+		t.Fatalf("caps n=%d bytes=%d", srv.bundleCache.maxN, srv.bundleCache.maxBytes)
+	}
+	var builds atomic.Int32
+	srv.nodeBundleBuild = func(revisionID, speakerID string, fragments map[string]string, priv ed25519.PrivateKey) ([]byte, error) {
+		builds.Add(1)
+		return bundle.BuildGzippedTar(revisionID, speakerID, fragments, priv)
+	}
+	client := ts.Client()
+
+	status, raw := pullNodeBundle(t, client, ts.URL, ids.speaker, "00000000-0000-0000-0000-000000000099")
+	if status != http.StatusNotFound {
+		t.Fatalf("missing revision status %d: %s", status, raw)
+	}
+	if builds.Load() != 0 {
+		t.Fatal("missing revision built a bundle")
+	}
+
+	body1 := mustPullNodeBundle(t, client, ts.URL, ids.speaker, ids.rev)
+	body2 := mustPullNodeBundle(t, client, ts.URL, ids.speaker, ids.rev)
+	if builds.Load() != 1 {
+		t.Fatalf("second pull rebuilt: builds=%d", builds.Load())
+	}
+	if !bytes.Equal(body1, body2) {
+		t.Fatal("cache hit was not byte-for-byte identical")
+	}
+	verifyNodeBundle(t, srv, body2)
+
+	sp2, err := srv.Store().CreateSpeaker(ids.tenant, &store.Speaker{
+		Role:     "replica",
+		Endpoint: "https://203.0.113.2",
+		MetaJSON: store.SpeakerMetaJSON(store.SpeakerMeta{NodeIPv4: "203.0.113.2"}),
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	bodyB := mustPullNodeBundle(t, client, ts.URL, sp2.ID, ids.rev)
+	if builds.Load() != 2 {
+		t.Fatalf("second speaker builds=%d", builds.Load())
+	}
+	if bytes.Equal(body1, bodyB) {
+		t.Fatal("different speakers returned the same bundle")
+	}
+	if !strings.Contains(bundleFile(t, bodyB, "bird.conf"), "203.0.113.2") {
+		t.Fatal("speaker B bundle missing its router id")
+	}
+	body3 := mustPullNodeBundle(t, client, ts.URL, ids.speaker, ids.rev)
+	if builds.Load() != 2 || !bytes.Equal(body1, body3) {
+		t.Fatalf("speaker A collided with B: builds=%d equal=%v", builds.Load(), bytes.Equal(body1, body3))
+	}
+
+	meta := store.SpeakerMetaJSON(store.SpeakerMeta{NodeIPv4: "198.51.100.9"})
+	if _, err := srv.Store().UpdateSpeaker(ids.tenant, ids.speaker, &store.SpeakerPatch{MetaJSON: &meta}); err != nil {
+		t.Fatal(err)
+	}
+	body4 := mustPullNodeBundle(t, client, ts.URL, ids.speaker, ids.rev)
+	if builds.Load() != 3 {
+		t.Fatalf("changed node ipv4 builds=%d, want 3", builds.Load())
+	}
+	if bytes.Equal(body1, body4) {
+		t.Fatal("changed overlay input reused the previous bundle")
+	}
+	conf := bundleFile(t, body4, "bird.conf")
+	if !strings.Contains(conf, "198.51.100.9") {
+		t.Fatalf("rebuilt bird.conf missing node ipv4: %s", conf)
+	}
+	verifyNodeBundle(t, srv, body4)
+	body5 := mustPullNodeBundle(t, client, ts.URL, ids.speaker, ids.rev)
+	if builds.Load() != 3 || !bytes.Equal(body4, body5) {
+		t.Fatal("rebuilt bundle was not cached")
+	}
+
+	source := store.SpeakerMetaJSON(store.SpeakerMeta{
+		NodeIPv4:          "198.51.100.9",
+		BirdBgpSourceIPv4: "203.0.113.40",
+	})
+	if _, err := srv.Store().UpdateSpeaker(ids.tenant, ids.speaker, &store.SpeakerPatch{MetaJSON: &source}); err != nil {
+		t.Fatal(err)
+	}
+	body6 := mustPullNodeBundle(t, client, ts.URL, ids.speaker, ids.rev)
+	if builds.Load() != 4 {
+		t.Fatalf("changed source ipv4 builds=%d, want 4", builds.Load())
+	}
+	conf = bundleFile(t, body6, "bird.conf")
+	if !strings.Contains(conf, "203.0.113.40") {
+		t.Fatalf("source ipv4 not applied: %s", conf)
+	}
+	verifyNodeBundle(t, srv, body6)
+
+	otherRev, err := srv.Store().CreateRollbackRevision(ids.tenant, ids.rev)
+	if err != nil {
+		t.Fatal(err)
+	}
+	bodyRev := mustPullNodeBundle(t, client, ts.URL, ids.speaker, otherRev)
+	if builds.Load() != 5 {
+		t.Fatalf("other revision builds=%d", builds.Load())
+	}
+	if bytes.Equal(body6, bodyRev) {
+		t.Fatal("revisions shared a bundle body")
+	}
+	cached := mustPullNodeBundle(t, client, ts.URL, ids.speaker, ids.rev)
+	if builds.Load() != 5 || !bytes.Equal(body6, cached) {
+		t.Fatal("original revision was evicted or rebuilt")
+	}
+}
+
+func TestHandleNodeBundleSingleflight(t *testing.T) {
+	srv, ts, ids := startNodeBundleServer(t)
+	var builds atomic.Int32
+	release := make(chan struct{})
+	srv.nodeBundleBuild = func(revisionID, speakerID string, fragments map[string]string, priv ed25519.PrivateKey) ([]byte, error) {
+		builds.Add(1)
+		<-release
+		return bundle.BuildGzippedTar(revisionID, speakerID, fragments, priv)
+	}
+	client := ts.Client()
+	const n = 2
+	bodies := make([][]byte, n)
+	errs := make([]error, n)
+	var wg sync.WaitGroup
+	wg.Add(n)
+	for i := 0; i < n; i++ {
+		go func(i int) {
+			defer wg.Done()
+			bodies[i], errs[i] = pullNodeBundleOK(client, ts.URL, ids.speaker, ids.rev)
+		}(i)
+	}
+	deadline := time.Now().Add(3 * time.Second)
+	joined := 0
+	for {
+		if builds.Load() > 1 {
+			close(release)
+			wg.Wait()
+			t.Fatalf("concurrent pulls built %d times", builds.Load())
+		}
+		srv.bundleCache.mu.Lock()
+		joined = 0
+		for _, f := range srv.bundleCache.flights {
+			joined += f.joined
+		}
+		srv.bundleCache.mu.Unlock()
+		if builds.Load() == 1 && joined >= 1 {
+			break
+		}
+		if time.Now().After(deadline) {
+			close(release)
+			wg.Wait()
+			t.Fatalf("timed out builds=%d joined=%d", builds.Load(), joined)
+		}
+		time.Sleep(time.Millisecond)
+	}
+	close(release)
+	wg.Wait()
+	if builds.Load() != 1 {
+		t.Fatalf("builds = %d, want 1", builds.Load())
+	}
+	for i := 0; i < n; i++ {
+		if errs[i] != nil {
+			t.Fatal(errs[i])
+		}
+	}
+	if !bytes.Equal(bodies[0], bodies[1]) {
+		t.Fatal("singleflight returned different bodies")
+	}
+	verifyNodeBundle(t, srv, bodies[0])
+}
+
+type nodeBundleIDs struct {
+	tenant, rev, speaker string
+}
+
+func startNodeBundleServer(t *testing.T) (*Server, *httptest.Server, nodeBundleIDs) {
+	t.Helper()
+	srv, err := New(Options{
+		InsecureDev:   true,
+		SeedDemo:      true,
+		BundleSeedHex: testBundleSeed,
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(srv.Close)
+	tenant, _, _, rev, speaker := srv.Store().DemoIDs()
+	mustSetTestAPIKeys(t, srv, "nodekey|"+tenant+"|node")
+	ts := httptest.NewServer(srv.Handler())
+	t.Cleanup(ts.Close)
+	return srv, ts, nodeBundleIDs{tenant: tenant, rev: rev, speaker: speaker}
+}
+
+func mustPullNodeBundle(t *testing.T, client *http.Client, base, speaker, rev string) []byte {
+	t.Helper()
+	status, body := pullNodeBundle(t, client, base, speaker, rev)
+	if status != http.StatusOK {
+		t.Fatalf("status %d: %s", status, body)
+	}
+	return body
+}
+
+func pullNodeBundle(t *testing.T, client *http.Client, base, speaker, rev string) (int, []byte) {
+	t.Helper()
+	req, err := http.NewRequest(http.MethodGet, base+"/v1/speakers/"+speaker+"/bundle/"+rev, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	req.Header.Set("Authorization", "Bearer nodekey")
+	resp, err := client.Do(req)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() { _ = resp.Body.Close() }()
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return resp.StatusCode, body
+}
+
+func pullNodeBundleOK(client *http.Client, base, speaker, rev string) ([]byte, error) {
+	req, err := http.NewRequest(http.MethodGet, base+"/v1/speakers/"+speaker+"/bundle/"+rev, nil)
+	if err != nil {
+		return nil, err
+	}
+	req.Header.Set("Authorization", "Bearer nodekey")
+	resp, err := client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = resp.Body.Close() }()
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, err
+	}
+	if resp.StatusCode != http.StatusOK {
+		return nil, errors.New(resp.Status + ": " + string(body))
+	}
+	return body, nil
+}
+
+func verifyNodeBundle(t *testing.T, srv *Server, raw []byte) {
+	t.Helper()
+	pub, err := base64.StdEncoding.DecodeString(srv.BundlePublicKeyBase64())
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := signing.VerifyGzippedTar(raw, ed25519.PublicKey(pub)); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func bundleFile(t *testing.T, gz []byte, name string) string {
+	t.Helper()
+	gr, err := gzip.NewReader(bytes.NewReader(gz))
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() { _ = gr.Close() }()
+	tr := tar.NewReader(gr)
+	for {
+		hdr, err := tr.Next()
+		if errors.Is(err, io.EOF) {
+			break
+		}
+		if err != nil {
+			t.Fatal(err)
+		}
+		if hdr.Name != name {
+			continue
+		}
+		var b strings.Builder
+		if _, err := io.Copy(&b, tr); err != nil {
+			t.Fatal(err)
+		}
+		return b.String()
+	}
+	t.Fatalf("archive missing %s", name)
+	return ""
+}
+
+func assertBundleCacheAccounting(t *testing.T, c *nodeBundleCache) {
+	t.Helper()
+	if len(c.items) != len(c.lru) {
+		t.Fatalf("items %d lru %d", len(c.items), len(c.lru))
+	}
+	sum := 0
+	seen := make(map[string]bool, len(c.lru))
+	for _, k := range c.lru {
+		e, ok := c.items[k]
+		if !ok {
+			t.Fatalf("lru key %q missing from items", k)
+		}
+		if seen[k] {
+			t.Fatalf("duplicate lru key %q", k)
+		}
+		seen[k] = true
+		sum += len(e.body)
+	}
+	if sum != c.total {
+		t.Fatalf("total=%d sum=%d", c.total, sum)
+	}
+	if c.maxN > 0 && len(c.items) > c.maxN {
+		t.Fatalf("entries %d over cap %d", len(c.items), c.maxN)
+	}
+	if c.maxBytes > 0 && c.total > c.maxBytes {
+		t.Fatalf("bytes %d over cap %d", c.total, c.maxBytes)
+	}
+}
+
+func cacheKeys(c *nodeBundleCache) []string {
+	keys := make([]string, 0, len(c.items))
+	for k := range c.items {
+		keys = append(keys, k)
+	}
+	return keys
+}

internal/httpapi/node_legacy_compatibility_test.go

@@ -0,0 +1,98 @@
+package httpapi
+
+import (
+	"crypto/ed25519"
+	"encoding/base64"
+	"encoding/json"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"evobgp/internal/bundle"
+)
+
+func TestLegacyCompatibilityNodePullAndAuth(t *testing.T) {
+	srv, err := New(Options{SeedDemo: true, BundleSeedHex: testBundleSeed})
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer srv.Close()
+	tenant, _, _, revision, speaker := srv.Store().DemoIDs()
+	mustSetTestAPIKeys(t, srv, "legacy-node|"+tenant+"|node,operator|"+tenant+"|operator,other-node|other-tenant|node")
+	if err := srv.Store().PublishRevisionForSpeaker(speaker, revision); err != nil {
+		t.Fatal(err)
+	}
+	handler := srv.Handler()
+	call := func(path, token string) (int, []byte, string) {
+		t.Helper()
+		req, err := http.NewRequest(http.MethodGet, path, nil)
+		if err != nil {
+			t.Fatal(err)
+		}
+		if token != "" {
+			req.Header.Set("Authorization", "Bearer "+token)
+		}
+		w := httptest.NewRecorder()
+		handler.ServeHTTP(w, req)
+		resp := w.Result()
+		defer func() { _ = resp.Body.Close() }()
+		body, err := io.ReadAll(resp.Body)
+		if err != nil {
+			t.Fatal(err)
+		}
+		return resp.StatusCode, body, resp.Header.Get("Content-Type")
+	}
+	latestPath := "/v1/speakers/" + speaker + "/revisions/latest"
+	bundlePath := "/v1/speakers/" + speaker + "/bundle/" + revision
+	for _, path := range []string{latestPath, bundlePath} {
+		for _, tc := range []struct {
+			token string
+			want  int
+		}{{"", http.StatusUnauthorized}, {"operator", http.StatusForbidden}, {"other-node", http.StatusForbidden}} {
+			status, _, _ := call(path, tc.token)
+			if status != tc.want {
+				t.Fatalf("%s token %q: status %d, want %d", path, tc.token, status, tc.want)
+			}
+		}
+	}
+	status, raw, _ := call(latestPath, "legacy-node")
+	if status != http.StatusOK {
+		t.Fatalf("legacy latest: status %d: %s", status, raw)
+	}
+	var latest struct {
+		RevisionID  string `json:"revision_id"`
+		PublishedAt string `json:"published_at"`
+	}
+	if err := json.Unmarshal(raw, &latest); err != nil {
+		t.Fatal(err)
+	}
+	if latest.RevisionID != revision || latest.PublishedAt == "" {
+		t.Fatalf("legacy latest response: %+v", latest)
+	}
+	status, raw, contentType := call("/v1/bundle/signing-public-key", "operator")
+	if status != http.StatusOK || contentType != "application/json" {
+		t.Fatalf("public key status %d, type %q: %s", status, contentType, raw)
+	}
+	var key struct {
+		PublicKeyBase64 string `json:"public_key_base64"`
+	}
+	if err := json.Unmarshal(raw, &key); err != nil {
+		t.Fatal(err)
+	}
+	pub, err := base64.StdEncoding.DecodeString(key.PublicKeyBase64)
+	if err != nil || len(pub) != ed25519.PublicKeySize {
+		t.Fatalf("legacy public key: len=%d err=%v", len(pub), err)
+	}
+	status, raw, contentType = call(bundlePath, "legacy-node")
+	if status != http.StatusOK || contentType != "application/gzip" {
+		t.Fatalf("legacy bundle: status %d, type %q: %s", status, contentType, raw)
+	}
+	verified, err := bundle.VerifyGzippedTar(raw, ed25519.PublicKey(pub))
+	if err != nil {
+		t.Fatalf("legacy signed bundle verification: %v", err)
+	}
+	if verified.Manifest.RevisionID != revision || verified.Manifest.SpeakerID != speaker || len(verified.Files) == 0 {
+		t.Fatalf("legacy bundle manifest: %+v", verified.Manifest)
+	}
+}

internal/httpapi/operation_limits.go

@@ -0,0 +1,105 @@
+package httpapi
+
+import (
+	"net/http"
+	"os"
+	"strconv"
+	"sync"
+	"time"
+)
+
+type expensiveOperation int
+
+const (
+	lookupOperation expensiveOperation = iota
+	cdnPreviewOperation
+	revisionPreviewOperation
+)
+
+type tenantWork struct {
+	start  time.Time
+	counts [3]int
+	active int
+}
+
+// operationLimiter is deliberately local to one API process. It never queues
+// waiters or evicts active tenants to admit new identities.
+type operationLimiter struct {
+	mu                                             sync.Mutex
+	tenants                                        map[string]*tenantWork
+	active, maxActive, maxTenantActive, maxTenants int
+	rates                                          [3]int
+	nextSweep                                      time.Time
+}
+
+func newOperationLimiter() *operationLimiter {
+	return &operationLimiter{
+		tenants: make(map[string]*tenantWork), maxTenants: 4096,
+		maxActive:       operationEnvLimit("EVOBGP_API_EXPENSIVE_MAX_CONCURRENT", 4, 64),
+		maxTenantActive: operationEnvLimit("EVOBGP_API_EXPENSIVE_TENANT_CONCURRENT", 2, 64),
+		rates:           [3]int{operationEnvLimit("EVOBGP_API_LOOKUP_PER_MINUTE", 120, 100000), operationEnvLimit("EVOBGP_API_CDN_PREVIEW_PER_MINUTE", 30, 100000), operationEnvLimit("EVOBGP_API_REVISION_PREVIEW_PER_MINUTE", 30, 100000)},
+	}
+}
+
+func operationEnvLimit(name string, fallback, maximum int) int {
+	n, err := strconv.Atoi(os.Getenv(name))
+	if err != nil || n < 1 || n > maximum {
+		return fallback
+	}
+	return n
+}
+
+func (l *operationLimiter) acquire(tenant string, op expensiveOperation, now time.Time) (func(), int) {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	if !now.Before(l.nextSweep) {
+		for key, entry := range l.tenants {
+			if entry.active == 0 && !now.Before(entry.start.Add(time.Minute)) {
+				delete(l.tenants, key)
+			}
+		}
+		l.nextSweep = now.Add(time.Second)
+	}
+	if l.active >= l.maxActive {
+		return nil, 1
+	}
+	entry := l.tenants[tenant]
+	if entry == nil {
+		if len(l.tenants) >= l.maxTenants {
+			return nil, 1
+		}
+		entry = &tenantWork{start: now}
+		l.tenants[tenant] = entry
+	}
+	if !now.Before(entry.start.Add(time.Minute)) {
+		entry.start = now
+		entry.counts = [3]int{}
+	}
+	if entry.active >= l.maxTenantActive {
+		return nil, 1
+	}
+	if entry.counts[op] >= l.rates[op] {
+		wait := entry.start.Add(time.Minute).Sub(now)
+		return nil, max(1, int((wait+time.Second-1)/time.Second))
+	}
+	entry.counts[op]++
+	entry.active++
+	l.active++
+	var once sync.Once
+	return func() { once.Do(func() { l.mu.Lock(); defer l.mu.Unlock(); entry.active--; l.active-- }) }, 0
+}
+
+func (s *Server) acquireOperation(w http.ResponseWriter, r *http.Request, tenant string, op expensiveOperation) (func(), bool) {
+	if r.Context().Err() != nil {
+		writeProblem(w, http.StatusRequestTimeout, "Request Timeout", "request cancelled")
+		return nil, false
+	}
+	s.operationOnce.Do(func() { s.operationLimiter = newOperationLimiter() })
+	release, retry := s.operationLimiter.acquire(tenant, op, time.Now())
+	if release == nil {
+		w.Header().Set("Retry-After", strconv.Itoa(retry))
+		writeProblem(w, http.StatusTooManyRequests, "Too Many Requests", "preview/lookup quota exceeded; retry later")
+		return nil, false
+	}
+	return release, true
+}

internal/httpapi/operation_limits_test.go

@@ -0,0 +1,143 @@
+package httpapi
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"sync"
+	"testing"
+	"time"
+
+	"evobgp/internal/store"
+)
+
+func TestOperationLimitsSharedCapacityAndWindows(t *testing.T) {
+	l := newOperationLimiter()
+	l.maxActive = 2
+	l.maxTenantActive = 1
+	l.rates = [3]int{1, 2, 2}
+	now := time.Now()
+	a, _ := l.acquire("a", lookupOperation, now)
+	if a == nil {
+		t.Fatal("first denied")
+	}
+	if r, _ := l.acquire("a", cdnPreviewOperation, now); r != nil {
+		t.Fatal("tenant concurrency bypass")
+	}
+	b, _ := l.acquire("b", revisionPreviewOperation, now)
+	if b == nil {
+		t.Fatal("other tenant denied")
+	}
+	if r, _ := l.acquire("c", lookupOperation, now); r != nil {
+		t.Fatal("global concurrency bypass")
+	}
+	a()
+	a()
+	b()
+	if l.active != 0 {
+		t.Fatal("release not idempotent")
+	}
+	if r, retry := l.acquire("a", lookupOperation, now.Add(time.Second)); r != nil || retry != 59 {
+		t.Fatalf("rate bypass/retry admitted=%t seconds=%d", r != nil, retry)
+	}
+	c, _ := l.acquire("a", cdnPreviewOperation, now.Add(time.Second))
+	if c == nil {
+		t.Fatal("independent operation denied")
+	}
+	c()
+	d, _ := l.acquire("a", lookupOperation, now.Add(time.Minute))
+	if d == nil {
+		t.Fatal("expired window not reset")
+	}
+	d()
+}
+
+func TestOperationTenantMapBoundWithoutActiveEviction(t *testing.T) {
+	l := newOperationLimiter()
+	l.maxTenants = 1
+	now := time.Now()
+	a, _ := l.acquire("a", lookupOperation, now)
+	if r, _ := l.acquire("b", lookupOperation, now.Add(2*time.Minute)); r != nil {
+		t.Fatal("active entry evicted")
+	}
+	a()
+	b, _ := l.acquire("b", lookupOperation, now.Add(3*time.Minute))
+	if b == nil || len(l.tenants) != 1 {
+		t.Fatal("idle expiry did not free bounded state")
+	}
+	b()
+}
+
+func TestOperationConcurrentAdmissionBound(t *testing.T) {
+	l := newOperationLimiter()
+	l.maxActive = 4
+	l.maxTenantActive = 4
+	l.rates = [3]int{100, 100, 100}
+	results := make(chan func(), 32)
+	var wg sync.WaitGroup
+	for range 32 {
+		wg.Add(1)
+		go func() { defer wg.Done(); r, _ := l.acquire("tenant", lookupOperation, time.Now()); results <- r }()
+	}
+	wg.Wait()
+	close(results)
+	accepted := 0
+	for r := range results {
+		if r != nil {
+			accepted++
+			r()
+		}
+	}
+	if accepted != 4 || l.active != 0 {
+		t.Fatalf("accepted=%d active=%d", accepted, l.active)
+	}
+}
+
+type waitingRevisionStore struct {
+	store.Backend
+	entered chan struct{}
+	unblock chan struct{}
+}
+
+func (s waitingRevisionStore) GetRevision(tenant, id string) (*store.Revision, error) {
+	s.entered <- struct{}{}
+	<-s.unblock
+	return s.Backend.GetRevision(tenant, id)
+}
+
+func TestRevisionQuotaHeldUntilWorkReturns(t *testing.T) {
+	t.Setenv("EVOBGP_API_EXPENSIVE_TENANT_CONCURRENT", "1")
+	t.Setenv("EVOBGP_API_REVISION_PREVIEW_PER_MINUTE", "30")
+	m := store.NewMemory()
+	m.SeedDemo()
+	tenant, _, _, revision, _ := m.DemoIDs()
+	entered := make(chan struct{}, 2)
+	unblock := make(chan struct{})
+	s := &Server{store: waitingRevisionStore{Backend: m, entered: entered, unblock: unblock}}
+	request := func() *http.Request {
+		r := httptest.NewRequest(http.MethodGet, "/v1/revisions/"+revision+"/preview", nil)
+		r.SetPathValue("revision_id", revision)
+		return r.WithContext(context.WithValue(r.Context(), authCtxKey, Auth{TenantID: tenant, Role: "operator"}))
+	}
+	ctx, cancel := context.WithCancel(request().Context())
+	defer cancel()
+	first := httptest.NewRecorder()
+	done := make(chan struct{})
+	go func() { defer close(done); s.handleRevisionPreview(first, request().WithContext(ctx)) }()
+	<-entered
+	cancel() // Backend deliberately ignores cancellation: capacity stays reserved.
+	second := httptest.NewRecorder()
+	s.handleRevisionPreview(second, request())
+	if second.Code != http.StatusTooManyRequests || second.Header().Get("Retry-After") == "" {
+		close(unblock)
+		<-done
+		t.Fatalf("quota: %d %s", second.Code, second.Body.String())
+	}
+	close(unblock)
+	<-done
+	third := httptest.NewRecorder()
+	s.handleRevisionPreview(third, request())
+	if third.Code != http.StatusOK {
+		t.Fatalf("release failed: %d %s", third.Code, third.Body.String())
+	}
+}

internal/httpapi/peers_discovery.go

@@ -6,7 +6,6 @@ import (
 	"strings"
 	"time"
 
-	"evobgp/internal/birdfmt"
 	"evobgp/internal/store"
 )
 
@@ -44,46 +43,6 @@ func peerDiscoveryJSON(d *store.BGPPeerDiscovery) map[string]any {
 	return m
 }
 
-// syncPeerDiscoveriesFromLive upserts pending discoveries from dynamic BGP sessions.
-func (s *Server) syncPeerDiscoveriesFromLive(tenantID string, views []speakerBGPLive) {
-	if s.store == nil || tenantID == "" {
-		return
-	}
-	knownNeighbors := map[string]struct{}{}
-	for _, p := range s.store.ListPeers(tenantID) {
-		if p == nil {
-			continue
-		}
-		if n, ok := store.NormalizePeerNeighborString(p.Neighbor); ok {
-			knownNeighbors[n] = struct{}{}
-		}
-	}
-	now := time.Now().UTC()
-	for _, v := range views {
-		for _, sess := range v.Sessions {
-			if !birdfmt.IsDynamicDiscoverySession(sess.Name) {
-				continue
-			}
-			neighbor, ok := store.NormalizePeerNeighborString(sess.Neighbor)
-			if !ok {
-				continue
-			}
-			if _, known := knownNeighbors[neighbor]; known {
-				continue
-			}
-			_, _ = s.store.UpsertPeerDiscovery(tenantID, &store.PeerDiscoveryUpsert{
-				SpeakerID:    v.SpeakerID,
-				NeighborID:   strings.TrimSpace(sess.NeighborID),
-				Neighbor:     neighbor,
-				RemoteASN:    sess.NeighborAS,
-				ProtocolName: sess.Name,
-				SessionState: sess.State,
-				SeenAt:       now,
-			})
-		}
-	}
-}
-
 func (s *Server) handleListPeerDiscoveries(w http.ResponseWriter, r *http.Request) {
 	a, ok := authFromContext(r.Context())
 	if !ok {
@@ -97,11 +56,6 @@ func (s *Server) handleListPeerDiscoveries(w http.ResponseWriter, r *http.Reques
 	if status == "" {
 		status = store.PeerDiscoveryPending
 	}
-	// Refresh live discovery when listing pending.
-	if status == store.PeerDiscoveryPending || status == "all" {
-		views := s.collectSpeakerBGPLive(r.Context(), a.TenantID, true)
-		s.syncPeerDiscoveriesFromLive(a.TenantID, views)
-	}
 	listStatus := status
 	if status == "all" {
 		listStatus = ""
@@ -138,7 +92,7 @@ func (s *Server) handleApprovePeerDiscovery(w http.ResponseWriter, r *http.Reque
 		writeStoreErr(w, err)
 		return
 	}
-	s.enqueuePeerReconcile(a.TenantID, "peer_discovery_approve")
+	s.enqueuePeerReconcile(w, r, a.TenantID, "peer_discovery_approve")
 	s.recordCRUDAudit(r, a, "bgp.peer.discovery.approve", "Approved discovered peer "+peer.Neighbor, peer.ID, map[string]any{
 		"peer_id":      peer.ID,
 		"discovery_id": disc.ID,

internal/httpapi/peers_live.go

@@ -1,47 +1,98 @@
 package httpapi
 
 import (
-	"context"
 	"net/netip"
-	"os"
 	"strings"
-	"sync"
 	"time"
 
 	"evobgp/internal/birdfmt"
-	"evobgp/internal/nodedispatch"
 	"evobgp/internal/store"
 )
 
-const peerLiveCacheTTL = 15 * time.Second
-
 type speakerBGPLive struct {
 	SpeakerID           string
 	Label               string
 	Sessions            []birdfmt.BGPSession
 	Error               string
+	Polled              bool
+	ObservedAt          time.Time
+	ReceivedAt          time.Time
+	LastSuccessAt       time.Time
+	Source              string
+	ObservationOverride *liveObservation
 }
 
-type peerLiveCacheEntry struct {
+type liveObservation struct {
-	at    time.Time
+	ObservedAt    *string `json:"observed_at"`
-	views []speakerBGPLive
+	ReceivedAt    *string `json:"received_at"`
+	LastSuccessAt *string `json:"last_success_at"`
+	ExpiresAt     *string `json:"expires_at,omitempty"`
+	Source        string  `json:"source"`
+	Quality       string  `json:"quality"`
+	ErrorCode     string  `json:"error_code,omitempty"`
 }
 
-var peerLiveCache sync.Map // tenantID -> peerLiveCacheEntry
+func liveTime(t time.Time) *string {
+	if t.IsZero() {
+		return nil
+	}
+	v := t.UTC().Format(time.RFC3339Nano)
+	return &v
+}
+
+func (v speakerBGPLive) observed() bool {
+	return v.Polled || v.Error != "" || len(v.Sessions) > 0
+}
+
+func (v speakerBGPLive) successful() bool {
+	return v.observed() && v.Error == ""
+}
+
+func (v speakerBGPLive) observation() liveObservation {
+	if v.ObservationOverride != nil {
+		return *v.ObservationOverride
+	}
+	o := liveObservation{ObservedAt: liveTime(v.ObservedAt), ReceivedAt: liveTime(v.ReceivedAt), LastSuccessAt: liveTime(v.LastSuccessAt), Source: v.Source, Quality: "unknown"}
+	if o.Source == "" {
+		o.Source = "none"
+	}
+	switch {
+	case v.successful():
+		o.Quality = "ok"
+	case v.Error != "" && !v.LastSuccessAt.IsZero():
+		o.Quality = "stale"
+		o.ErrorCode = liveErrorCode(v.Error)
+	case v.Error != "":
+		o.Quality = "unreachable"
+		o.ErrorCode = liveErrorCode(v.Error)
+	default:
+		o.ErrorCode = "not_polled"
+	}
+	return o
+}
+
+func liveErrorCode(err string) string {
+	if strings.Contains(strings.ToLower(err), "not configured") || strings.Contains(strings.ToLower(err), "missing") {
+		return "not_configured"
+	}
+	return "poll_failed"
+}
 
 type peerSessionOnSpeaker struct {
 	SpeakerID   string          `json:"speaker_id"`
 	Label       string          `json:"label"`
 	State       string          `json:"state"`
 	PollError   string          `json:"poll_error,omitempty"`
+	Observation liveObservation `json:"observation"`
 }
 
 type liveSpeakerPoll struct {
 	SpeakerID    string          `json:"speaker_id"`
 	Label        string          `json:"label"`
 	OK           bool            `json:"ok"`
-	SessionCount int    `json:"session_count"`
+	SessionCount *int            `json:"session_count"`
 	PollError    string          `json:"poll_error,omitempty"`
+	Observation  liveObservation `json:"observation"`
 }
 
 func speakerDisplayLabel(sp *store.Speaker) string {
@@ -65,95 +116,21 @@ func speakerDisplayLabel(sp *store.Speaker) string {
 	return sp.ID
 }
 
-func masterSpeakerID(speakers []*store.Speaker) string {
-	for _, sp := range speakers {
-		if sp != nil && strings.EqualFold(strings.TrimSpace(sp.Role), "master") {
-			return sp.ID
-		}
-	}
-	return ""
-}
-
-func (s *Server) collectSpeakerBGPLive(ctx context.Context, tenantID string, fresh bool) []speakerBGPLive {
-	if !fresh {
-		if v, ok := peerLiveCache.Load(tenantID); ok {
-			ent := v.(peerLiveCacheEntry)
-			if time.Since(ent.at) < peerLiveCacheTTL {
-				return ent.views
-			}
-		}
-	}
-	speakers := s.store.ListSpeakersForTenant(tenantID)
-	views := make([]speakerBGPLive, 0, len(speakers)+1)
-
-	if sock := strings.TrimSpace(os.Getenv("EVOBGP_BIRDC_SOCKET")); sock != "" {
-		v := speakerBGPLive{Label: "CP (local BIRD)"}
-		if mid := masterSpeakerID(speakers); mid != "" {
-			v.SpeakerID = mid
-			for _, sp := range speakers {
-				if sp != nil && sp.ID == mid {
-					v.Label = speakerDisplayLabel(sp)
-					break
-				}
-			}
-		}
-		out, err := birdfmt.ShowProtocols(ctx, sock, strings.TrimSpace(os.Getenv("EVOBGP_BIRDC_BIN")))
-		if err != nil {
-			v.Error = err.Error()
-		} else {
-			v.Sessions = birdfmt.ParseBGPSessions(out)
-		}
-		views = append(views, v)
-	}
-
-	opts := nodedispatch.Options{Timeout: 8 * time.Second}
-	type resWrap struct {
-		sp  *store.Speaker
-		res nodedispatch.BirdProtocolsResult
-	}
-	ch := make(chan resWrap, len(speakers))
-	var wg sync.WaitGroup
-	for _, sp := range speakers {
-		if sp == nil {
-			continue
-		}
-		meta := store.ParseSpeakerMeta(sp.MetaJSON)
-		if !store.SpeakerNeedsRemoteDispatch(sp.Role, meta) {
-			continue
-		}
-		wg.Add(1)
-		go func(speaker *store.Speaker) {
-			defer wg.Done()
-			ch <- resWrap{
-				sp:  speaker,
-				res: nodedispatch.FetchBirdProtocols(ctx, speaker, opts),
-			}
-		}(sp)
-	}
-	wg.Wait()
-	close(ch)
-	for rw := range ch {
-		views = append(views, speakerBGPLive{
-			SpeakerID: rw.sp.ID,
-			Label:     speakerDisplayLabel(rw.sp),
-			Sessions:  rw.res.Sessions,
-			Error:     rw.res.Error,
-		})
-	}
-
-	peerLiveCache.Store(tenantID, peerLiveCacheEntry{at: time.Now(), views: views})
-	return views
-}
-
 func liveSpeakerPollJSON(views []speakerBGPLive) []liveSpeakerPoll {
 	out := make([]liveSpeakerPoll, 0, len(views))
 	for _, v := range views {
+		var sessionCount *int
+		if v.successful() {
+			count := len(v.Sessions)
+			sessionCount = &count
+		}
 		out = append(out, liveSpeakerPoll{
 			SpeakerID:    v.SpeakerID,
 			Label:        v.Label,
-			OK:           v.Error == "",
+			OK:           v.successful(),
-			SessionCount: len(v.Sessions),
+			SessionCount: sessionCount,
 			PollError:    v.Error,
+			Observation:  v.observation(),
 		})
 	}
 	return out
@@ -183,11 +160,12 @@ func matchPeerOnSpeakers(peer *store.BGPPeer, views []speakerBGPLive) (
 	protoName := birdfmt.PeerProtocolName(peer.ID)
 
 	for _, v := range views {
-		if v.Error != "" && len(v.Sessions) == 0 {
+		if !v.successful() {
 			on = append(on, peerSessionOnSpeaker{
 				SpeakerID:   v.SpeakerID,
 				Label:       v.Label,
 				PollError:   v.Error,
+				Observation: v.observation(),
 			})
 			continue
 		}
@@ -197,6 +175,7 @@ func matchPeerOnSpeakers(peer *store.BGPPeer, views []speakerBGPLive) (
 				SpeakerID:   v.SpeakerID,
 				Label:       v.Label,
 				State:       "absent",
+				Observation: v.observation(),
 			})
 			continue
 		}
@@ -204,6 +183,7 @@ func matchPeerOnSpeakers(peer *store.BGPPeer, views []speakerBGPLive) (
 			SpeakerID:   v.SpeakerID,
 			Label:       v.Label,
 			State:       sess.State,
+			Observation: v.observation(),
 		}
 		on = append(on, hit)
 		if strings.EqualFold(strings.TrimSpace(sess.State), "Established") {
@@ -273,20 +253,109 @@ func sessionStateRank(state string) int {
 }
 
 func applyPeerLiveFields(row map[string]any, peer *store.BGPPeer, views []speakerBGPLive) {
-	state, connID, connLabel, establishedOn, on, mismatch := matchPeerOnSpeakers(peer, views)
+	expected := expectedPeerViews(peer, views)
+	state, _, _, _, _, _ := matchPeerOnSpeakers(peer, expected)
+	_, connID, connLabel, establishedOn, on, mismatch := matchPeerOnSpeakers(peer, views)
 	row["session_on_speakers"] = on
 	row["established_on_speakers"] = establishedOn
 	row["session_conflict"] = false
 	row["session_mismatch"] = mismatch
+	row["last_known_session_state"] = row["session_state"]
+	row["session_state"] = nil
 	if state != "" {
 		row["session_state"] = state
 	}
+	if state == "" && len(expected) > 0 && allSpeakerPollsSuccessful(expected) {
+		row["session_state"] = "absent"
+	}
+	row["live_observation"] = peerLiveObservation(expected)
 	if connLabel != "" {
 		row["connected_speaker_label"] = connLabel
 	}
 	row["connected_speaker_id"] = peerLiveSpeakerIDOrNull(connID)
 }
 
+func expectedPeerViews(peer *store.BGPPeer, views []speakerBGPLive) []speakerBGPLive {
+	if peer == nil || peer.SpeakerID == nil || strings.TrimSpace(*peer.SpeakerID) == "" {
+		return views
+	}
+	want := strings.TrimSpace(*peer.SpeakerID)
+	expected := make([]speakerBGPLive, 0, 1)
+	for _, v := range views {
+		if strings.EqualFold(v.SpeakerID, want) {
+			expected = append(expected, v)
+		}
+	}
+	return expected
+}
+
+func allSpeakerPollsSuccessful(views []speakerBGPLive) bool {
+	for _, v := range views {
+		if !v.successful() {
+			return false
+		}
+	}
+	return true
+}
+
+func peerLiveObservation(views []speakerBGPLive) liveObservation {
+	if len(views) == 0 {
+		return speakerBGPLive{}.observation()
+	}
+	result := liveObservation{Source: views[0].Source, Quality: "unknown"}
+	if result.Source == "" {
+		result.Source = "none"
+	}
+	var latestObserved, latestReceived, latestSuccess time.Time
+	allOK, allFailed, anyStale := true, true, false
+	for _, v := range views {
+		o := v.observation()
+		if v.ObservedAt.After(latestObserved) {
+			latestObserved = v.ObservedAt
+		}
+		if v.ReceivedAt.After(latestReceived) {
+			latestReceived = v.ReceivedAt
+		}
+		if v.LastSuccessAt.After(latestSuccess) {
+			latestSuccess = v.LastSuccessAt
+		}
+		allOK = allOK && v.successful()
+		allFailed = allFailed && o.Quality == "unreachable"
+		anyStale = anyStale || o.Quality == "stale"
+		if result.ErrorCode == "" && o.ErrorCode != "" {
+			result.ErrorCode = o.ErrorCode
+		}
+		if o.ExpiresAt != nil {
+			expires, err := time.Parse(time.RFC3339Nano, *o.ExpiresAt)
+			if err == nil {
+				previous := time.Time{}
+				if result.ExpiresAt != nil {
+					previous, _ = time.Parse(time.RFC3339Nano, *result.ExpiresAt)
+				}
+				if previous.IsZero() || expires.Before(previous) {
+					result.ExpiresAt = o.ExpiresAt
+				}
+			}
+		}
+	}
+	result.ObservedAt = liveTime(latestObserved)
+	result.ReceivedAt = liveTime(latestReceived)
+	result.LastSuccessAt = liveTime(latestSuccess)
+	if len(views) > 1 {
+		result.Source = "mixed"
+	}
+	switch {
+	case allOK:
+		result.Quality = "ok"
+		result.ErrorCode = ""
+	case anyStale:
+		result.Quality = "stale"
+	case allFailed:
+		result.Quality = "unreachable"
+	}
+	return result
+}
+
 func peerLiveSpeakerIDOrNull(id string) any {
 	if strings.TrimSpace(id) == "" {
 		return nil

internal/httpapi/peers_live_test.go

@@ -2,11 +2,84 @@ package httpapi
 
 import (
 	"testing"
+	"time"
 
 	"evobgp/internal/birdfmt"
 	"evobgp/internal/store"
 )
 
+func TestApplyPeerLiveFieldsObservationStates(t *testing.T) {
+	peer := &store.BGPPeer{ID: "aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee", Neighbor: "198.51.100.2"}
+	now := time.Date(2026, 10, 26, 10, 0, 0, 0, time.UTC)
+	cases := []struct {
+		name        string
+		views       []speakerBGPLive
+		wantState   any
+		wantQuality string
+	}{
+		{"no collector", nil, nil, "unknown"},
+		{"successful empty poll", []speakerBGPLive{{Polled: true, Source: "agent_bird_protocols", ObservedAt: now, ReceivedAt: now, LastSuccessAt: now}}, "absent", "ok"},
+		{"first failure", []speakerBGPLive{{Polled: true, Error: "timeout", Source: "agent_bird_protocols", ObservedAt: now}}, nil, "unreachable"},
+		{"failure after success", []speakerBGPLive{{Polled: true, Error: "timeout", Source: "agent_bird_protocols", ObservedAt: now, LastSuccessAt: now.Add(-time.Minute)}}, nil, "stale"},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			row := map[string]any{"session_state": "Established"}
+			applyPeerLiveFields(row, peer, tc.views)
+			if row["session_state"] != tc.wantState || row["last_known_session_state"] != "Established" {
+				t.Fatalf("session_state=%v last_known=%v", row["session_state"], row["last_known_session_state"])
+			}
+			observation := row["live_observation"].(liveObservation)
+			if observation.Quality != tc.wantQuality {
+				t.Fatalf("observation=%+v", observation)
+			}
+		})
+	}
+}
+
+func TestApplyPeerLiveFieldsExpectedSpeakerCoverage(t *testing.T) {
+	now := time.Date(2026, 10, 26, 10, 0, 0, 0, time.UTC)
+	assigned := "speaker-a"
+	peer := &store.BGPPeer{ID: "aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee", Neighbor: "198.51.100.2", SpeakerID: &assigned}
+	viewAError := speakerBGPLive{SpeakerID: assigned, Polled: true, Error: "timeout", ObservedAt: now, Source: "agent_bird_protocols"}
+	viewAEmpty := speakerBGPLive{SpeakerID: assigned, Polled: true, ObservedAt: now, LastSuccessAt: now, Source: "agent_bird_protocols"}
+	viewBEmpty := speakerBGPLive{SpeakerID: "speaker-b", Polled: true, ObservedAt: now.Add(time.Minute), LastSuccessAt: now.Add(time.Minute), Source: "agent_bird_protocols"}
+	viewBError := speakerBGPLive{SpeakerID: "speaker-b", Polled: true, Error: "timeout", ObservedAt: now.Add(time.Minute), Source: "agent_bird_protocols"}
+	cases := []struct {
+		name         string
+		assigned     bool
+		views        []speakerBGPLive
+		wantState    any
+		wantQuality  string
+		wantObserved string
+	}{
+		{"assigned timeout unrelated success", true, []speakerBGPLive{viewAError, viewBEmpty}, nil, "unreachable", now.Format(time.RFC3339Nano)},
+		{"assigned success unrelated failure", true, []speakerBGPLive{viewAEmpty, viewBError}, "absent", "ok", now.Format(time.RFC3339Nano)},
+		{"global mixed outcome", false, []speakerBGPLive{viewAEmpty, viewBError}, nil, "unknown", now.Add(time.Minute).Format(time.RFC3339Nano)},
+		{"assigned speaker not observed", true, []speakerBGPLive{viewBEmpty}, nil, "unknown", ""},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			candidate := *peer
+			if !tc.assigned {
+				candidate.SpeakerID = nil
+			}
+			row := map[string]any{"session_state": "Established"}
+			applyPeerLiveFields(row, &candidate, tc.views)
+			obs := row["live_observation"].(liveObservation)
+			if row["session_state"] != tc.wantState || obs.Quality != tc.wantQuality {
+				t.Fatalf("state=%v observation=%+v", row["session_state"], obs)
+			}
+			if tc.wantObserved == "" && obs.ObservedAt != nil {
+				t.Fatalf("unexpected timestamp %v", *obs.ObservedAt)
+			}
+			if tc.wantObserved != "" && (obs.ObservedAt == nil || *obs.ObservedAt != tc.wantObserved) {
+				t.Fatalf("observed_at=%v want=%q", obs.ObservedAt, tc.wantObserved)
+			}
+		})
+	}
+}
+
 func TestMatchPeerOnSpeakers_establishedOnReplica(t *testing.T) {
 	peer := &store.BGPPeer{
 		ID:       "aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee",

internal/httpapi/ratelimit.go

@@ -9,23 +9,33 @@ import (
 	"time"
 )
 
-// authRateLimiter is a simple per-IP token bucket for public auth-ish endpoints.
+const maxAuthRateLimitClients = 4096
+
+type authRateHit struct {
+	start time.Time
+	count int
+}
+
+// authRateLimiter keeps a bounded fixed-window counter per client IP.
 type authRateLimiter struct {
 	mu         sync.Mutex
-	hits   map[string][]time.Time
+	hits       map[string]authRateHit
 	limit      int
 	window     time.Duration
+	maxClients int
+	nextSweep  time.Time
 }
 
 func newAuthRateLimiter() *authRateLimiter {
 	limit := 60
-	if n, err := strconv.Atoi(strings.TrimSpace(os.Getenv("EVOBGP_AUTH_RATE_LIMIT"))); err == nil && n > 0 {
+	if n, err := strconv.Atoi(strings.TrimSpace(os.Getenv("EVOBGP_AUTH_RATE_LIMIT"))); err == nil && n > 0 && n <= 1000 {
 		limit = n
 	}
 	return &authRateLimiter{
-		hits:   make(map[string][]time.Time),
+		hits:       make(map[string]authRateHit),
 		limit:      limit,
 		window:     time.Minute,
+		maxClients: maxAuthRateLimitClients,
 	}
 }
 
@@ -33,23 +43,40 @@ func (l *authRateLimiter) allow(ip string) bool {
 	if l == nil {
 		return true
 	}
-	now := time.Now()
+	return l.allowAt(ip, time.Now())
+}
+
+func (l *authRateLimiter) allowAt(ip string, now time.Time) bool {
 	l.mu.Lock()
 	defer l.mu.Unlock()
-	cut := now.Add(-l.window)
+	if !l.nextSweep.IsZero() && !now.Before(l.nextSweep) {
-	arr := l.hits[ip]
+		l.nextSweep = time.Time{}
-	kept := arr[:0]
+		for key, hit := range l.hits {
-	for _, t := range arr {
+			expiry := hit.start.Add(l.window)
-		if t.After(cut) {
+			if !now.Before(expiry) {
-			kept = append(kept, t)
+				delete(l.hits, key)
+			} else if l.nextSweep.IsZero() || expiry.Before(l.nextSweep) {
+				l.nextSweep = expiry
 			}
 		}
-	if len(kept) >= l.limit {
+	}
-		l.hits[ip] = kept
+	hit, exists := l.hits[ip]
+	if !exists && len(l.hits) >= l.maxClients {
+		// Never evict an active counter: a new identity must wait for expiry.
 		return false
 	}
-	kept = append(kept, now)
+	if exists && hit.count >= l.limit {
-	l.hits[ip] = kept
+		return false
+	}
+	if !exists {
+		hit.start = now
+		expiry := now.Add(l.window)
+		if l.nextSweep.IsZero() || expiry.Before(l.nextSweep) {
+			l.nextSweep = expiry
+		}
+	}
+	hit.count++
+	l.hits[ip] = hit
 	return true
 }
 

internal/httpapi/request_body.go

@@ -0,0 +1,100 @@
+package httpapi
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"io"
+	"net"
+	"net/http"
+	"os"
+	"strconv"
+	"strings"
+)
+
+const requestBodyLimit int64 = 8 << 20
+
+// withRequestBodyLimit must run inside authMiddleware. Reading the complete
+// bounded body before dispatch prevents a valid JSON prefix from committing a
+// mutation before a later size/read error is discovered. Content-Type is not a
+// security boundary; CSV retains its additional row/field checks in its handler.
+func withRequestBodyLimit(next http.Handler) http.Handler {
+	return withBodyLimitAndConcurrency(requestBodyLimit, requestBodyConcurrency(), next)
+}
+
+func withBodyLimit(limit int64, next http.Handler) http.Handler {
+	return withBodyLimitAndConcurrency(limit, 4, next)
+}
+
+func requestBodyConcurrency() int {
+	n, err := strconv.Atoi(strings.TrimSpace(os.Getenv("EVOBGP_API_BODY_MAX_CONCURRENT")))
+	if err == nil && n >= 1 && n <= 64 {
+		return n
+	}
+	return 4
+}
+
+func withBodyLimitAndConcurrency(limit int64, concurrency int, next http.Handler) http.Handler {
+	slots := make(chan struct{}, concurrency)
+	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		original := r.Body
+		if original != nil {
+			defer func() { _ = original.Close() }()
+		}
+		if r.ContentLength > limit {
+			writeRequestBodyTooLarge(w)
+			return
+		}
+		if r.Context().Err() != nil {
+			writeProblem(w, http.StatusRequestTimeout, "Request Timeout", "request body read cancelled")
+			return
+		}
+		if r.Body == nil || r.Body == http.NoBody {
+			next.ServeHTTP(w, r)
+			return
+		}
+		select {
+		case slots <- struct{}{}:
+			defer func() { <-slots }()
+		default:
+			w.Header().Set("Retry-After", "1")
+			writeProblem(w, http.StatusTooManyRequests, "Too Many Requests", "request body processing capacity is full")
+			return
+		}
+		bounded := http.MaxBytesReader(w, r.Body, limit)
+		data, err := io.ReadAll(requestContextReader{ctx: r.Context(), reader: bounded})
+		var tooLarge *http.MaxBytesError
+		if errors.As(err, &tooLarge) {
+			writeRequestBodyTooLarge(w)
+			return
+		}
+		var timeout net.Error
+		if r.Context().Err() != nil || errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) || (errors.As(err, &timeout) && timeout.Timeout()) {
+			writeProblem(w, http.StatusRequestTimeout, "Request Timeout", "request body read cancelled")
+			return
+		}
+		if err != nil {
+			writeProblem(w, http.StatusBadRequest, "Bad Request", "could not read request body")
+			return
+		}
+		r.Body = io.NopCloser(bytes.NewReader(data))
+		r.ContentLength = int64(len(data))
+		next.ServeHTTP(w, r)
+	})
+}
+
+func writeRequestBodyTooLarge(w http.ResponseWriter) {
+	writeProblem(w, http.StatusRequestEntityTooLarge, "Content Too Large", "request body exceeds the 8 MiB limit")
+}
+
+type requestContextReader struct {
+	ctx    context.Context
+	reader io.Reader
+}
+
+func (r requestContextReader) Read(p []byte) (int, error) {
+	if err := r.ctx.Err(); err != nil {
+		return 0, err
+	}
+	return r.reader.Read(p)
+}

internal/httpapi/request_body_test.go

@@ -0,0 +1,273 @@
+package httpapi
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+)
+
+func TestRequestBodyKnownAndUnknownBoundsBeforeMutation(t *testing.T) {
+	const limit = 32
+	for _, unknown := range []bool{false, true} {
+		for _, contentType := range []string{"", "application/json", "text/plain"} {
+			for _, extra := range []int{0, 1} {
+				label := "known"
+				if unknown {
+					label = "unknown"
+				}
+				t.Run(label+"/"+contentType+"/"+string(rune('0'+extra)), func(t *testing.T) {
+					mutations := 0
+					h := withBodyLimit(limit, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+						var input map[string]any
+						if err := json.NewDecoder(r.Body).Decode(&input); err != nil {
+							t.Error(err)
+						}
+						mutations++
+						w.WriteHeader(http.StatusCreated)
+					}))
+					body := `{"x":1}` + strings.Repeat(" ", limit-len(`{"x":1}`)+extra)
+					req := httptest.NewRequest(http.MethodPost, "/v1/modules", strings.NewReader(body))
+					if unknown {
+						req.ContentLength = -1
+					}
+					if contentType != "" {
+						req.Header.Set("Content-Type", contentType)
+					}
+					w := httptest.NewRecorder()
+					h.ServeHTTP(w, req)
+					if extra == 0 {
+						if w.Code != http.StatusCreated || mutations != 1 {
+							t.Fatalf("exact limit rejected: %d mutations=%d", w.Code, mutations)
+						}
+					} else if w.Code != http.StatusRequestEntityTooLarge || mutations != 0 {
+						t.Fatalf("oversized tail mutated: %d mutations=%d", w.Code, mutations)
+					}
+				})
+			}
+		}
+	}
+}
+
+type checkedRequestBody struct {
+	reads  int
+	closed bool
+	reader io.Reader
+}
+
+func (b *checkedRequestBody) Read(p []byte) (int, error) { b.reads++; return b.reader.Read(p) }
+func (b *checkedRequestBody) Close() error               { b.closed = true; return nil }
+
+type failingRequestReader struct{}
+
+func (failingRequestReader) Read([]byte) (int, error) {
+	return 0, errors.New("untrusted detailed read failure")
+}
+
+func TestRequestBodyErrorsDoNotMutate(t *testing.T) {
+	for _, cancelled := range []bool{false, true} {
+		calls := 0
+		b := &checkedRequestBody{reader: io.MultiReader(strings.NewReader(`{"x":1}`), failingRequestReader{})}
+		req := httptest.NewRequest(http.MethodPost, "/v1/modules", nil)
+		req.Body = b
+		req.ContentLength = -1
+		if cancelled {
+			ctx, cancel := context.WithCancel(req.Context())
+			cancel()
+			req = req.WithContext(ctx)
+		}
+		w := httptest.NewRecorder()
+		withRequestBodyLimit(http.HandlerFunc(func(http.ResponseWriter, *http.Request) { calls++ })).ServeHTTP(w, req)
+		want := http.StatusBadRequest
+		if cancelled {
+			want = http.StatusRequestTimeout
+		}
+		if w.Code != want || calls != 0 || !b.closed || strings.Contains(w.Body.String(), "untrusted detailed") {
+			t.Fatalf("error handling status=%d calls=%d body=%s", w.Code, calls, w.Body.String())
+		}
+		if cancelled && b.reads != 0 {
+			t.Fatal("read body after cancellation")
+		}
+	}
+}
+
+func TestRequestBodyAuthenticationRunsFirst(t *testing.T) {
+	b := &checkedRequestBody{reader: failingRequestReader{}}
+	req := httptest.NewRequest(http.MethodPost, "/v1/modules", nil)
+	req.Body = b
+	req.ContentLength = requestBodyLimit + 1
+	s := &Server{}
+	called := false
+	h := s.authMiddleware(withRequestBodyLimit(http.HandlerFunc(func(http.ResponseWriter, *http.Request) { called = true })))
+	w := httptest.NewRecorder()
+	h.ServeHTTP(w, req)
+	if w.Code != http.StatusUnauthorized || b.reads != 0 || called {
+		t.Fatalf("body processed before auth: %d reads=%d", w.Code, b.reads)
+	}
+}
+
+func TestRequestBodyAppliesToGETAndIgnoresClaimedZeroLength(t *testing.T) {
+	for _, length := range []int64{-1, 0, requestBodyLimit + 1} {
+		req := httptest.NewRequest(http.MethodGet, "/v1/jobs", strings.NewReader(strings.Repeat("x", int(requestBodyLimit)+1)))
+		req.ContentLength = length
+		called := false
+		w := httptest.NewRecorder()
+		withRequestBodyLimit(http.HandlerFunc(func(http.ResponseWriter, *http.Request) { called = true })).ServeHTTP(w, req)
+		if w.Code != http.StatusRequestEntityTooLarge || called {
+			t.Fatalf("GET size bypass length=%d code=%d", length, w.Code)
+		}
+	}
+}
+
+func TestRequestBodyHandlerWiringAndNoMutation(t *testing.T) {
+	s, err := New(Options{SeedDemo: true, InsecureDev: true, BundleSeedHex: testBundleSeed})
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer s.Close()
+	tenant, _, _, _, _ := s.store.DemoIDs()
+	before := len(s.store.ListModules(tenant))
+	h := s.Handler()
+	prefix := `{"type":"IP_RANGES","name":"bounded-create","enabled":false}`
+	for _, extra := range []int{1, 0} {
+		body := prefix + strings.Repeat(" ", int(requestBodyLimit)-len(prefix)+extra)
+		req := httptest.NewRequest(http.MethodPost, "/v1/modules", strings.NewReader(body))
+		req.ContentLength = -1
+		req.Header.Set("Authorization", "Bearer dev")
+		w := httptest.NewRecorder()
+		h.ServeHTTP(w, req)
+		if extra == 1 {
+			if w.Code != http.StatusRequestEntityTooLarge || len(s.store.ListModules(tenant)) != before {
+				t.Fatalf("oversized JSON committed: %d %s", w.Code, w.Body.String())
+			}
+		} else if w.Code != http.StatusCreated || len(s.store.ListModules(tenant)) != before+1 {
+			t.Fatalf("8MiB body rejected: %d %s", w.Code, w.Body.String())
+		}
+	}
+}
+
+func TestRequestBodyLeavesStreamingResponseUnbuffered(t *testing.T) {
+	w := httptest.NewRecorder()
+	req := httptest.NewRequest(http.MethodGet, "/v1/runtime-logs/stream", nil)
+	withRequestBodyLimit(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		f, ok := w.(http.Flusher)
+		if !ok {
+			t.Fatal("response lost Flusher")
+		}
+		w.Header().Set("Content-Type", "text/event-stream")
+		_, _ = io.WriteString(w, "data: ready\n\n")
+		f.Flush()
+	})).ServeHTTP(w, req)
+	if !w.Flushed || w.Body.String() != "data: ready\n\n" {
+		t.Fatal("stream response buffered")
+	}
+}
+
+type blockedAdmissionBody struct {
+	entered chan struct{}
+	resume  chan struct{}
+	sent    bool
+}
+
+func (b *blockedAdmissionBody) Read(p []byte) (int, error) {
+	if b.sent {
+		return 0, io.EOF
+	}
+	b.sent = true
+	close(b.entered)
+	<-b.resume
+	p[0] = 'x'
+	return 1, nil
+}
+func (*blockedAdmissionBody) Close() error { return nil }
+
+func TestRequestBodyAdmissionRejectsWithoutWaitingAndHoldsThroughHandler(t *testing.T) {
+	readEntered, readResume := make(chan struct{}), make(chan struct{})
+	handlerEntered, handlerResume := make(chan struct{}), make(chan struct{})
+	var once sync.Once
+	h := withBodyLimitAndConcurrency(32, 1, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/empty" {
+			once.Do(func() { close(handlerEntered) })
+			<-handlerResume
+		}
+		w.WriteHeader(http.StatusNoContent)
+	}))
+	first := httptest.NewRequest(http.MethodPost, "/first", nil)
+	first.Body = &blockedAdmissionBody{entered: readEntered, resume: readResume}
+	first.ContentLength = 0 // A claimed zero must not bypass body admission.
+	done := make(chan struct{})
+	go func() { defer close(done); h.ServeHTTP(httptest.NewRecorder(), first) }()
+	select {
+	case <-readEntered:
+	case <-time.After(3 * time.Second):
+		t.Fatal("first body did not start")
+	}
+	reject := func() {
+		t.Helper()
+		w := httptest.NewRecorder()
+		req := httptest.NewRequest(http.MethodPost, "/second", strings.NewReader("x"))
+		h.ServeHTTP(w, req)
+		if w.Code != http.StatusTooManyRequests || w.Header().Get("Retry-After") != "1" {
+			t.Fatalf("admission did not reject: %d %s", w.Code, w.Body.String())
+		}
+	}
+	reject()
+	// An empty legacy GET remains available while all body slots are occupied.
+	empty := httptest.NewRecorder()
+	h.ServeHTTP(empty, httptest.NewRequest(http.MethodGet, "/empty", nil))
+	if empty.Code != http.StatusNoContent {
+		t.Fatal("empty GET consumed body slot")
+	}
+	close(readResume)
+	select {
+	case <-handlerEntered:
+	case <-time.After(3 * time.Second):
+		t.Fatal("handler did not start")
+	}
+	reject() // The fully read buffer is still retained by the running handler.
+	close(handlerResume)
+	select {
+	case <-done:
+	case <-time.After(3 * time.Second):
+		t.Fatal("first request did not finish")
+	}
+	w := httptest.NewRecorder()
+	h.ServeHTTP(w, httptest.NewRequest(http.MethodPost, "/third", strings.NewReader("x")))
+	if w.Code != http.StatusNoContent {
+		t.Fatalf("body slot not released: %d", w.Code)
+	}
+}
+
+func TestRequestBodyAdmissionReleasesOnReadError(t *testing.T) {
+	h := withBodyLimitAndConcurrency(32, 1, http.HandlerFunc(func(http.ResponseWriter, *http.Request) { t.Fatal("read error reached handler") }))
+	for range 2 {
+		req := httptest.NewRequest(http.MethodPost, "/bad", nil)
+		req.Body = io.NopCloser(failingRequestReader{})
+		req.ContentLength = -1
+		w := httptest.NewRecorder()
+		h.ServeHTTP(w, req)
+		if w.Code != http.StatusBadRequest {
+			t.Fatalf("read failure retained slot: %d", w.Code)
+		}
+	}
+}
+
+func TestRequestBodyAdmissionEnvironment(t *testing.T) {
+	for _, tc := range []struct {
+		raw  string
+		want int
+	}{{"", 4}, {"1", 1}, {"64", 64}, {"65", 4}, {"0", 4}, {"-1", 4}, {"invalid", 4}} {
+		t.Run(tc.raw, func(t *testing.T) {
+			t.Setenv("EVOBGP_API_BODY_MAX_CONCURRENT", tc.raw)
+			if got := requestBodyConcurrency(); got != tc.want {
+				t.Fatalf("concurrency=%d want%d", got, tc.want)
+			}
+		})
+	}
+}

internal/httpapi/routes.go

@@ -8,14 +8,12 @@ import (
 	"errors"
 	"io"
 	"net/http"
-	"os"
 	"sort"
 	"strconv"
 	"strings"
 	"time"
 
 	"evobgp/internal/birdfmt"
-	"evobgp/internal/bundle"
 	"evobgp/internal/jobs"
 	"evobgp/internal/observability"
 	"evobgp/internal/pipeline"
@@ -39,7 +37,7 @@ func (s *Server) Handler() http.Handler {
 	// Firewall subsystem moved to the standalone EvoFirewall service; see docs/firewall.md.
 	// Registered on the public mux so it wins over the "/v1/" subtree below regardless of auth.
 	s.mux.HandleFunc("/v1/firewall/", s.handleFirewallGone)
-	s.mux.Handle("/v1/", s.authMiddleware(wrappedV1))
+	s.mux.Handle("/v1/", s.authMiddleware(withRequestBodyLimit(wrappedV1)))
 	return s.withCORS(observability.HTTPMiddleware(s.mux))
 }
 
@@ -60,6 +58,7 @@ func (s *Server) registerV1(m *http.ServeMux) {
 	m.HandleFunc("GET /modules/{module_id}", s.handleGetModule)
 	m.HandleFunc("GET /peers", s.handleListPeers)
 	m.HandleFunc("GET /speakers", s.handleListSpeakers)
+	m.HandleFunc("POST /telemetry/refresh", s.handleTelemetryRefresh)
 	m.HandleFunc("GET /bundle/signing-public-key", s.handleBundleSigningPublicKey)
 	m.HandleFunc("POST /modules/{module_id}/refresh", s.handleModuleRefresh)
 	m.HandleFunc("POST /tenant/refresh", s.handleTenantRefresh)
@@ -79,6 +78,7 @@ func (s *Server) registerV1(m *http.ServeMux) {
 	m.HandleFunc("GET /jobs/{job_id}", s.handleGetJob)
 	m.HandleFunc("GET /jobs/{job_id}/report", s.handleGetJobReport)
 	m.HandleFunc("POST /jobs/{job_id}/cancel", s.handleCancelJob)
+	m.HandleFunc("POST /jobs/{job_id}/retry", s.handleRetryJob)
 	m.HandleFunc("GET /speakers/{speaker_id}/revisions/latest", s.handleNodeLatestRevision)
 	m.HandleFunc("GET /speakers/{speaker_id}/bundle/{revision_id}", s.handleNodeBundle)
 	m.HandleFunc("POST /nodes/enroll", s.handleNodeEnroll)
@@ -295,12 +295,10 @@ func (s *Server) handleListPeers(w http.ResponseWriter, r *http.Request) {
 	allPeers := s.store.ListPeers(a.TenantID)
 	allPeers = store.FilterOwned(allPeers, func(p *store.BGPPeer) string { return p.CreatedByUserID }, a.Kind, a.IsAdmin, a.UserID)
 	page, next, more := store.PaginateOffset(allPeers, r.URL.Query().Get("cursor"), parseListLimit(r))
-	fresh := r != nil && strings.EqualFold(strings.TrimSpace(r.URL.Query().Get("live")), "1")
+	liveViews, _, err := s.loadTelemetry(r.Context(), a.TenantID, s.store.ListSpeakersForTenant(a.TenantID))
-	ctx, cancel := context.WithTimeout(r.Context(), 12*time.Second)
+	if err != nil {
-	defer cancel()
+		writeProblem(w, http.StatusServiceUnavailable, "Service Unavailable", "telemetry snapshots unavailable")
-	liveViews := s.collectSpeakerBGPLive(ctx, a.TenantID, fresh)
+		return
-	if fresh {
-		s.syncPeerDiscoveriesFromLive(a.TenantID, liveViews)
 	}
 	items := make([]map[string]any, 0, len(page))
 	for _, p := range page {
@@ -316,30 +314,6 @@ func (s *Server) handleListPeers(w http.ResponseWriter, r *http.Request) {
 	})
 }
 
-func (s *Server) liveBGPProtocolStates(r *http.Request) map[string]string {
-	if r != nil && strings.EqualFold(strings.TrimSpace(r.URL.Query().Get("live")), "1") {
-		return s.liveBGPProtocolStatesFresh(r.Context())
-	}
-	if cached, ok := observability.CachedBirdProtocolStates(90 * time.Second); ok {
-		return cached
-	}
-	return s.liveBGPProtocolStatesFresh(r.Context())
-}
-
-func (s *Server) liveBGPProtocolStatesFresh(ctx context.Context) map[string]string {
-	sock := strings.TrimSpace(os.Getenv("EVOBGP_BIRDC_SOCKET"))
-	if sock == "" {
-		return map[string]string{}
-	}
-	out, err := birdfmt.ShowProtocols(ctx, sock, strings.TrimSpace(os.Getenv("EVOBGP_BIRDC_BIN")))
-	if err != nil {
-		return map[string]string{}
-	}
-	states := birdfmt.ParseBGPProtocolStates(out)
-	observability.SetBirdProtocolStates(states)
-	return states
-}
-
 // parseBGPProtocolStates parses `birdc show protocols all` summary rows into protocol_name -> state.
 func parseBGPProtocolStates(output string) map[string]string {
 	out := make(map[string]string)
@@ -403,9 +377,12 @@ func (s *Server) handleListSpeakers(w http.ResponseWriter, r *http.Request) {
 	fresh := r != nil && strings.EqualFold(strings.TrimSpace(r.URL.Query().Get("live")), "1")
 	var liveByID map[string]map[string]any
 	if fresh {
-		ctx, cancel := context.WithTimeout(r.Context(), 12*time.Second)
+		var err error
-		defer cancel()
+		_, liveByID, err = s.loadTelemetry(r.Context(), a.TenantID, speakers)
-		liveByID = s.collectSpeakerLiveStatus(ctx, a.TenantID, true, speakers)
+		if err != nil {
+			writeProblem(w, http.StatusServiceUnavailable, "Service Unavailable", "telemetry snapshots unavailable")
+			return
+		}
 	}
 	items := make([]map[string]any, 0, len(speakers))
 	for _, sp := range speakers {
@@ -448,12 +425,12 @@ func (s *Server) handleModuleRefresh(w http.ResponseWriter, r *http.Request) {
 		idemPtr = &idem
 	}
 	mid := mod.ID
-	j, _, err := s.jobs.Enqueue(a.TenantID, jobs.KindModuleRefresh, idemPtr, &mid, map[string]any{
+	j, _, err := s.jobs.EnqueueContext(r.Context(), a.TenantID, jobs.KindModuleRefresh, idemPtr, &mid, map[string]any{
 		"module_id": moduleID,
 		"trigger":   "api",
 	})
 	if err != nil {
-		writeInternalError(w, "internal", err)
+		writeJobEnqueueError(w, err)
 		return
 	}
 	w.Header().Set("Location", "/v1/jobs/"+j.ID)
@@ -480,6 +457,17 @@ func (s *Server) handleTenantRefresh(w http.ResponseWriter, r *http.Request) {
 		}
 	}
 	moduleIDs := body.ModuleIDs
+	sort.Strings(body.ModuleIDs)
+	requestIDs := append([]string{}, body.ModuleIDs...)
+	idem := strings.TrimSpace(r.Header.Get("Idempotency-Key"))
+	var idemPtr *string
+	if idem != "" {
+		idemPtr = &idem
+	}
+	request := map[string]any{"module_ids": requestIDs, "trigger": "api"}
+	if s.replayJobResponse(w, r, store.JobEnqueue{TenantID: a.TenantID, Kind: jobs.KindTenantRefresh, IdempotencyKey: idemPtr, Request: request}) {
+		return
+	}
 	if len(moduleIDs) == 0 {
 		now := time.Now().UTC()
 		for _, mod := range s.store.ListModules(a.TenantID) {
@@ -488,21 +476,16 @@ func (s *Server) handleTenantRefresh(w http.ResponseWriter, r *http.Request) {
 			}
 		}
 	}
-	if len(moduleIDs) == 0 {
+	if len(moduleIDs) == 0 && idem == "" {
 		writeJSON(w, http.StatusNoContent, map[string]any{"message": "no modules due for refresh"})
 		return
 	}
-	idem := strings.TrimSpace(r.Header.Get("Idempotency-Key"))
+	j, _, err := s.jobs.EnqueueRequestContext(r.Context(), a.TenantID, jobs.KindTenantRefresh, idemPtr, nil, map[string]any{
-	var idemPtr *string
-	if idem != "" {
-		idemPtr = &idem
-	}
-	j, _, err := s.jobs.Enqueue(a.TenantID, jobs.KindTenantRefresh, idemPtr, nil, map[string]any{
 		"module_ids": moduleIDs,
 		"trigger":    "api",
-	})
+	}, request)
 	if err != nil {
-		writeInternalError(w, "internal", err)
+		writeJobEnqueueError(w, err)
 		return
 	}
 	w.Header().Set("Location", "/v1/jobs/"+j.ID)
@@ -510,7 +493,7 @@ func (s *Server) handleTenantRefresh(w http.ResponseWriter, r *http.Request) {
 	writeJSON(w, http.StatusAccepted, map[string]any{
 		"job_id":      snap["job_id"],
 		"status":      snap["status"],
-		"module_ids":  moduleIDs,
+		"module_ids":  j.Snapshot()["meta"].(map[string]any)["module_ids"],
 		"modules_due": len(moduleIDs),
 	})
 }
@@ -560,8 +543,8 @@ func revisionJSON(rev *store.Revision) map[string]any {
 	return m
 }
 
-// enqueueModuleRefreshIfEnabled queues module_refresh when the module exists and is enabled (best-effort, no HTTP error).
+// enqueueModuleRefreshIfEnabled reports scheduling failure after a committed CRUD mutation.
-func (s *Server) enqueueModuleRefreshIfEnabled(tenantID, moduleID, trigger string) {
+func (s *Server) enqueueModuleRefreshIfEnabled(w http.ResponseWriter, r *http.Request, tenantID, moduleID, trigger string) {
 	if s.jobs == nil {
 		return
 	}
@@ -570,22 +553,35 @@ func (s *Server) enqueueModuleRefreshIfEnabled(tenantID, moduleID, trigger strin
 		return
 	}
 	mid := moduleID
-	key := "module_refresh:" + moduleID
+	// Each mutation is a new logical refresh. Reusing a module-wide key would
-	_, _, _ = s.jobs.Enqueue(tenantID, jobs.KindModuleRefresh, &key, &mid, map[string]any{
+	// incorrectly replay an older terminal job after later configuration changes.
+	_, _, err = s.jobs.EnqueueContext(r.Context(), tenantID, jobs.KindModuleRefresh, nil, &mid, map[string]any{
 		"module_id": moduleID,
 		"trigger":   trigger,
 	})
+	s.reportAutomaticEnqueue(w, r, trigger, moduleID, err)
 }
 
 // enqueuePeerReconcile queues fast peer-only reconcile/render (best-effort, no HTTP error).
-func (s *Server) enqueuePeerReconcile(tenantID, trigger string) {
+func (s *Server) enqueuePeerReconcile(w http.ResponseWriter, r *http.Request, tenantID, trigger string) {
 	if s.jobs == nil {
 		return
 	}
-	_, _, _ = s.jobs.Enqueue(tenantID, jobs.KindPeerReconcile, nil, nil, map[string]any{
+	_, _, err := s.jobs.EnqueueContext(r.Context(), tenantID, jobs.KindPeerReconcile, nil, nil, map[string]any{
 		"trigger":   trigger,
 		"job_title": "Обновление BGP пиров",
 	})
+	s.reportAutomaticEnqueue(w, r, trigger, "", err)
+}
+
+func (s *Server) reportAutomaticEnqueue(w http.ResponseWriter, r *http.Request, trigger, target string, err error) {
+	if err == nil {
+		return
+	}
+	w.Header().Set("X-EvoBGP-Job-Status", "enqueue-failed")
+	if a, ok := authFromContext(r.Context()); ok {
+		s.recordCRUDAudit(r, a, "bgp.job.enqueue_failed", "Configuration saved; automatic job was not queued", target, map[string]any{"trigger": trigger, "error": err.Error()})
+	}
 }
 
 func (s *Server) handleGetRevision(w http.ResponseWriter, r *http.Request) {
@@ -614,6 +610,11 @@ func (s *Server) handleRevisionPreview(w http.ResponseWriter, r *http.Request) {
 	if !s.requirePerm(w, a, "bgp:operations:read") {
 		return
 	}
+	release, ok := s.acquireOperation(w, r, a.TenantID, revisionPreviewOperation)
+	if !ok {
+		return
+	}
+	defer release()
 	rev, err := s.store.GetRevision(a.TenantID, r.PathValue("revision_id"))
 	if err != nil {
 		writeProblem(w, http.StatusNotFound, "Not Found", "revision not found")
@@ -706,11 +707,11 @@ func (s *Server) handleRevisionRollback(w http.ResponseWriter, r *http.Request)
 	if idem != "" {
 		idemPtr = &idem
 	}
-	j, _, err := s.jobs.Enqueue(a.TenantID, jobs.KindRevisionRollback, idemPtr, nil, map[string]any{
+	j, _, err := s.jobs.EnqueueContext(r.Context(), a.TenantID, jobs.KindRevisionRollback, idemPtr, nil, map[string]any{
 		"source_revision_id": revID,
 	})
 	if err != nil {
-		writeInternalError(w, "internal", err)
+		writeJobEnqueueError(w, err)
 		return
 	}
 	w.Header().Set("Location", "/v1/jobs/"+j.ID)
@@ -751,13 +752,13 @@ func (s *Server) handleApply(w http.ResponseWriter, r *http.Request) {
 	if idem != "" {
 		idemPtr = &idem
 	}
-	j, _, err := s.jobs.Enqueue(a.TenantID, jobs.KindDeployApply, idemPtr, nil, map[string]any{
+	j, _, err := s.jobs.EnqueueContext(r.Context(), a.TenantID, jobs.KindDeployApply, idemPtr, nil, map[string]any{
 		"revision_id": revID,
 		"strategy":    body.Strategy,
 		"trigger":     "api",
 	})
 	if err != nil {
-		writeInternalError(w, "internal", err)
+		writeJobEnqueueError(w, err)
 		return
 	}
 	w.Header().Set("Location", "/v1/jobs/"+j.ID)
@@ -802,13 +803,13 @@ func (s *Server) handleSpeakerApply(w http.ResponseWriter, r *http.Request) {
 	if idem != "" {
 		idemPtr = &idem
 	}
-	j, _, err := s.jobs.Enqueue(a.TenantID, jobs.KindDeployApply, idemPtr, nil, map[string]any{
+	j, _, err := s.jobs.EnqueueContext(r.Context(), a.TenantID, jobs.KindDeployApply, idemPtr, nil, map[string]any{
 		"revision_id": revID,
 		"speaker_id":  spkID,
 		"trigger":     "api",
 	})
 	if err != nil {
-		writeInternalError(w, "internal", err)
+		writeJobEnqueueError(w, err)
 		return
 	}
 	w.Header().Set("Location", "/v1/jobs/"+j.ID)
@@ -830,9 +831,9 @@ func (s *Server) handleBirdReload(w http.ResponseWriter, r *http.Request) {
 	if idem != "" {
 		idemPtr = &idem
 	}
-	j, _, err := s.jobs.Enqueue(a.TenantID, jobs.KindBirdReload, idemPtr, nil, nil)
+	j, _, err := s.jobs.EnqueueContext(r.Context(), a.TenantID, jobs.KindBirdReload, idemPtr, nil, nil)
 	if err != nil {
-		writeInternalError(w, "internal", err)
+		writeJobEnqueueError(w, err)
 		return
 	}
 	w.Header().Set("Location", "/v1/jobs/"+j.ID)
@@ -867,7 +868,11 @@ func (s *Server) handleListJobs(w http.ResponseWriter, r *http.Request) {
 	status := r.URL.Query().Get("status")
 	kind := r.URL.Query().Get("kind")
 	cursor := r.URL.Query().Get("cursor")
-	list, next, more := s.jobs.List(a.TenantID, status, kind, cursor, limit)
+	list, next, more, err := s.jobs.ListContext(r.Context(), a.TenantID, status, kind, cursor, limit)
+	if err != nil {
+		writeInternalError(w, "jobs_list", err)
+		return
+	}
 	items := make([]map[string]any, 0, len(list))
 	for _, j := range list {
 		items = append(items, j.Snapshot())
@@ -886,9 +891,9 @@ func (s *Server) handleGetJob(w http.ResponseWriter, r *http.Request) {
 	if !s.requirePerm(w, a, "bgp:operations:read") {
 		return
 	}
-	j, err := s.jobs.Get(a.TenantID, r.PathValue("job_id"))
+	j, err := s.jobs.GetContext(r.Context(), a.TenantID, r.PathValue("job_id"))
 	if err != nil {
-		writeProblem(w, http.StatusNotFound, "Not Found", "job not found")
+		writeJobReadError(w, err)
 		return
 	}
 	writeJSON(w, http.StatusOK, j.Snapshot())
@@ -903,9 +908,9 @@ func (s *Server) handleGetJobReport(w http.ResponseWriter, r *http.Request) {
 	if !s.requirePerm(w, a, "bgp:operations:read") {
 		return
 	}
-	j, err := s.jobs.Get(a.TenantID, r.PathValue("job_id"))
+	j, err := s.jobs.GetContext(r.Context(), a.TenantID, r.PathValue("job_id"))
 	if err != nil {
-		writeProblem(w, http.StatusNotFound, "Not Found", "job not found")
+		writeJobReadError(w, err)
 		return
 	}
 	snap := j.Snapshot()
@@ -929,6 +934,35 @@ func (s *Server) handleGetJobReport(w http.ResponseWriter, r *http.Request) {
 			out["revision_id"] = v
 		}
 	}
+	sourceJobID := ""
+	if source, ok := meta["source_deployment_job_id"].(string); ok && source != "" {
+		// Resolve through the tenant-scoped registry before exposing linked progress.
+		if _, err := s.jobs.GetContext(r.Context(), a.TenantID, source); err != nil {
+			writeJobReadError(w, err)
+			return
+		}
+		sourceJobID = source
+		out["source_deployment_job_id"] = source
+	}
+	rows, err := s.store.ListDeploymentSpeakers(r.Context(), a.TenantID, j.ID)
+	if err == nil && len(rows) == 0 && sourceJobID != "" {
+		rows, err = s.store.ListDeploymentSpeakers(r.Context(), a.TenantID, sourceJobID)
+	}
+	if err != nil {
+		writeProblem(w, http.StatusServiceUnavailable, "Service Unavailable", "deployment progress unavailable")
+		return
+	}
+	progress := make([]map[string]any, 0, len(rows))
+	for _, d := range rows {
+		progress = append(progress, map[string]any{
+			"job_id": d.JobID, "speaker_id": d.SpeakerID, "revision_id": d.RevisionID,
+			"generation": d.Generation, "state": d.State, "confirmation": d.Confirmation,
+			"error": d.Error, "dispatch_attempts": d.DispatchAttempts, "retry_round": d.RetryRound,
+			"health_received_at": d.HealthReceivedAt, "health_revision_id": d.HealthRevisionID,
+			"health_source_at": d.HealthSourceAt, "health_error": d.HealthError,
+		})
+	}
+	out["deployment_progress"] = progress
 	writeJSON(w, http.StatusOK, out)
 }
 
@@ -941,9 +975,34 @@ func (s *Server) handleCancelJob(w http.ResponseWriter, r *http.Request) {
 	if !s.requirePerm(w, a, "bgp:operations:admin") {
 		return
 	}
-	j, err := s.jobs.RequestCancel(a.TenantID, r.PathValue("job_id"))
+	j, err := s.jobs.RequestCancelContext(r.Context(), a.TenantID, r.PathValue("job_id"))
 	if err != nil {
-		writeProblem(w, http.StatusNotFound, "Not Found", "job not found")
+		writeJobReadError(w, err)
+		return
+	}
+	writeJSON(w, http.StatusAccepted, j.Snapshot())
+}
+
+func (s *Server) handleRetryJob(w http.ResponseWriter, r *http.Request) {
+	a, ok := authFromContext(r.Context())
+	if !ok {
+		writeProblem(w, http.StatusUnauthorized, "Unauthorized", "missing auth")
+		return
+	}
+	if !s.requirePerm(w, a, "bgp:operations:admin") {
+		return
+	}
+	j, err := s.jobs.RetryContext(r.Context(), a.TenantID, r.PathValue("job_id"))
+	if err != nil {
+		if errors.Is(err, store.ErrJobQueueFull) {
+			writeJobEnqueueError(w, err)
+			return
+		}
+		if errors.Is(err, store.ErrJobConflict) {
+			writeProblem(w, http.StatusConflict, "Conflict", "job is not failed")
+			return
+		}
+		writeJobReadError(w, err)
 		return
 	}
 	writeJSON(w, http.StatusAccepted, j.Snapshot())
@@ -998,21 +1057,30 @@ func (s *Server) handleNodeBundle(w http.ResponseWriter, r *http.Request) {
 		writeProblem(w, http.StatusForbidden, "Forbidden", "speaker not in tenant scope")
 		return
 	}
-	rev, err := s.store.GetRevision(a.TenantID, rid)
+	overlayFP, err := nodeBundleOverlayFingerprint(s.store, a.TenantID, sid, rid)
-	if err != nil {
-		writeProblem(w, http.StatusNotFound, "Not Found", "revision not found")
-		return
-	}
-	frags := rev.PreviewFragments
-	overlaid, err := pipeline.OverlayFragmentsForSpeaker(s.store, a.TenantID, sid, rid, frags)
 	if err != nil {
 		writeInternalError(w, "bundle overlay", err)
 		return
 	}
-	frags = overlaid
+	// Finished gzip, including manifest and bundle.sig. Revision bytes are immutable;
-	tgz, err := bundle.BuildGzippedTar(rid, sid, frags, s.bundlePriv)
+	// the overlay fingerprint keeps speaker-specific renders from colliding.
+	tgz, err := s.bundleCache.do(r.Context(), nodeBundleCacheKey(a.TenantID, rid, sid, overlayFP), func(ctx context.Context) ([]byte, bool, error) {
+		return s.loadNodeBundle(ctx, a.TenantID, sid, rid, overlayFP)
+	})
 	if err != nil {
-		writeInternalError(w, "internal", err)
+		if errors.Is(err, errNodeBundleRevisionMissing) {
+			writeProblem(w, http.StatusNotFound, "Not Found", "revision not found")
+			return
+		}
+		if errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
+			return
+		}
+		op := "internal"
+		var opErr *nodeBundleOpError
+		if errors.As(err, &opErr) && opErr.op != "" {
+			op = opErr.op
+		}
+		writeInternalError(w, op, err)
 		return
 	}
 	w.Header().Set("Content-Type", "application/gzip")

internal/httpapi/routes_audit.go

@@ -3,7 +3,6 @@ package httpapi
 import (
 	"context"
 	"log"
-	"net"
 	"net/http"
 	"strings"
 	"time"
@@ -132,23 +131,7 @@ func (s *Server) pushAuditToPortal(entry *store.AuditEntry) {
 }
 
 func clientIP(r *http.Request) string {
-	if r == nil {
+	return clientIPWithTrustedProxies(r, trustedProxyPrefixes())
-		return ""
-	}
-	if xff := strings.TrimSpace(r.Header.Get("X-Forwarded-For")); xff != "" {
-		parts := strings.Split(xff, ",")
-		if len(parts) > 0 {
-			return strings.TrimSpace(parts[0])
-		}
-	}
-	if xrip := strings.TrimSpace(r.Header.Get("X-Real-IP")); xrip != "" {
-		return xrip
-	}
-	host, _, err := net.SplitHostPort(strings.TrimSpace(r.RemoteAddr))
-	if err != nil {
-		return strings.TrimSpace(r.RemoteAddr)
-	}
-	return host
 }
 
 // initAuditPusher wires portal push when URL and secret are configured.

internal/httpapi/routes_crud.go

@@ -296,10 +296,11 @@ func (s *Server) handlePreviewCDNSource(w http.ResponseWriter, r *http.Request)
 		writeProblem(w, http.StatusUnprocessableEntity, "Unprocessable Entity", invalidInputDetail)
 		return
 	}
-	if err := pipeline.ResolveCDNURLHost(r.Context(), u); err != nil {
+	release, ok := s.acquireOperation(w, r, a.TenantID, cdnPreviewOperation)
-		writeProblem(w, http.StatusUnprocessableEntity, "Unprocessable Entity", invalidInputDetail)
+	if !ok {
 		return
 	}
+	defer release()
 	mod, err := s.store.GetModule(a.TenantID, r.PathValue("module_id"))
 	if err != nil {
 		writeStoreErr(w, err)
@@ -309,6 +310,10 @@ func (s *Server) handlePreviewCDNSource(w http.ResponseWriter, r *http.Request)
 		writeProblem(w, http.StatusUnprocessableEntity, "Unprocessable Entity", "module type must be CDN_CIDRS")
 		return
 	}
+	if err := pipeline.ResolveCDNURLHost(r.Context(), u); err != nil {
+		writeProblem(w, http.StatusUnprocessableEntity, "Unprocessable Entity", invalidInputDetail)
+		return
+	}
 	req, err := http.NewRequestWithContext(r.Context(), http.MethodGet, u, nil)
 	if err != nil {
 		writeProblem(w, http.StatusUnprocessableEntity, "Unprocessable Entity", "invalid url")
@@ -321,11 +326,11 @@ func (s *Server) handlePreviewCDNSource(w http.ResponseWriter, r *http.Request)
 	}
 	defer func() { _ = resp.Body.Close() }()
 	if resp.StatusCode != http.StatusOK {
-		_, _ = io.Copy(io.Discard, resp.Body)
+		_, _ = io.Copy(io.Discard, io.LimitReader(resp.Body, 64<<10))
 		writeBadGateway(w, "cdn preview fetch", fmt.Errorf("upstream status: %s", resp.Status))
 		return
 	}
-	raw, err := io.ReadAll(io.LimitReader(resp.Body, 8<<20))
+	raw, err := pipeline.ReadCDNBody(resp.Body)
 	if err != nil {
 		writeBadGateway(w, "cdn preview read body", err)
 		return
@@ -379,7 +384,7 @@ func (s *Server) handlePostCDNSource(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.cdn_source.create", "Created CDN source", x.ID, map[string]any{"module_id": mid, "source_id": x.ID, "url": x.URL})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "cdn_source_create")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "cdn_source_create")
 	writeJSON(w, http.StatusCreated, cdnSourceJSON(x))
 }
 
@@ -410,7 +415,7 @@ func (s *Server) handlePatchCDNSource(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.cdn_source.update", "Updated CDN source", x.ID, map[string]any{"module_id": mid, "source_id": x.ID})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "cdn_source_patch")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "cdn_source_patch")
 	writeJSON(w, http.StatusOK, cdnSourceJSON(x))
 }
 
@@ -426,7 +431,7 @@ func (s *Server) handleDeleteCDNSource(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.cdn_source.delete", "Deleted CDN source", sourceID, map[string]any{"module_id": mid, "source_id": sourceID})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "cdn_source_delete")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "cdn_source_delete")
 	w.WriteHeader(http.StatusNoContent)
 }
 
@@ -485,7 +490,7 @@ func (s *Server) handlePostAS(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.as_entry.create", "Created AS entry", x.ID, map[string]any{"module_id": mid, "entry_id": x.ID, "asn": x.ASN})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "as_entry_create")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "as_entry_create")
 	writeJSON(w, http.StatusCreated, asEntryJSON(x))
 }
 
@@ -506,7 +511,7 @@ func (s *Server) handlePatchAS(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.as_entry.update", "Updated AS entry", x.ID, map[string]any{"module_id": mid, "entry_id": x.ID, "asn": x.ASN})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "as_entry_patch")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "as_entry_patch")
 	writeJSON(w, http.StatusOK, asEntryJSON(x))
 }
 
@@ -522,7 +527,7 @@ func (s *Server) handleDeleteAS(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.as_entry.delete", "Deleted AS entry", entryID, map[string]any{"module_id": mid, "entry_id": entryID})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "as_entry_delete")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "as_entry_delete")
 	w.WriteHeader(http.StatusNoContent)
 }
 
@@ -566,7 +571,7 @@ func (s *Server) handlePostDomain(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.domain_entry.create", "Created domain entry", x.ID, map[string]any{"module_id": mid, "entry_id": x.ID, "fqdn": x.FQDN})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "domain_entry_create")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "domain_entry_create")
 	writeJSON(w, http.StatusCreated, domainEntryJSON(x))
 }
 
@@ -587,7 +592,7 @@ func (s *Server) handlePatchDomain(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.domain_entry.update", "Updated domain entry", x.ID, map[string]any{"module_id": mid, "entry_id": x.ID, "fqdn": x.FQDN})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "domain_entry_patch")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "domain_entry_patch")
 	writeJSON(w, http.StatusOK, domainEntryJSON(x))
 }
 
@@ -603,7 +608,7 @@ func (s *Server) handleDeleteDomain(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.domain_entry.delete", "Deleted domain entry", entryID, map[string]any{"module_id": mid, "entry_id": entryID})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "domain_entry_delete")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "domain_entry_delete")
 	w.WriteHeader(http.StatusNoContent)
 }
 
@@ -647,7 +652,7 @@ func (s *Server) handlePostIPRange(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.ip_range.create", "Created IP range entry", x.ID, map[string]any{"module_id": mid, "entry_id": x.ID, "prefix": x.Prefix})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "ip_range_create")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "ip_range_create")
 	writeJSON(w, http.StatusCreated, ipRangeJSON(x))
 }
 
@@ -668,7 +673,7 @@ func (s *Server) handlePatchIPRange(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.ip_range.update", "Updated IP range entry", x.ID, map[string]any{"module_id": mid, "entry_id": x.ID, "prefix": x.Prefix})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "ip_range_patch")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "ip_range_patch")
 	writeJSON(w, http.StatusOK, ipRangeJSON(x))
 }
 
@@ -684,7 +689,7 @@ func (s *Server) handleDeleteIPRange(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.ip_range.delete", "Deleted IP range entry", entryID, map[string]any{"module_id": mid, "entry_id": entryID})
-	s.enqueueModuleRefreshIfEnabled(a.TenantID, mid, "ip_range_delete")
+	s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, mid, "ip_range_delete")
 	w.WriteHeader(http.StatusNoContent)
 }
 
@@ -777,8 +782,12 @@ func (s *Server) handleImportModuleEntriesCSV(w http.ResponseWriter, r *http.Req
 		return
 	}
 	moduleID := r.PathValue("module_id")
-	res, err := importer.ImportModuleEntriesCSV(s.store, a.TenantID, moduleID, r.Body)
+	res, err := importer.ImportModuleEntriesCSV(r.Context(), s.store, a.TenantID, moduleID, r.Body)
 	if err != nil {
+		if errors.Is(err, importer.ErrCSVTooLarge) {
+			writeProblem(w, http.StatusRequestEntityTooLarge, "Content Too Large", "csv exceeds 8 MiB, 10000 data rows or 64 fields per record; split the import into smaller requests")
+			return
+		}
 		if errors.Is(err, store.ErrInvalidInput) {
 			writeProblem(w, http.StatusUnprocessableEntity, "Unprocessable Entity", "csv is empty")
 			return
@@ -802,11 +811,11 @@ func (s *Server) handleImportModuleEntriesCSV(w http.ResponseWriter, r *http.Req
 	if res.Imported > 0 {
 		switch res.ModuleType {
 		case "AS_PREFIXES":
-			s.enqueueModuleRefreshIfEnabled(a.TenantID, moduleID, "as_entry_import_csv")
+			s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, moduleID, "as_entry_import_csv")
 		case "DOMAINS":
-			s.enqueueModuleRefreshIfEnabled(a.TenantID, moduleID, "domain_entry_import_csv")
+			s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, moduleID, "domain_entry_import_csv")
 		case "IP_RANGES":
-			s.enqueueModuleRefreshIfEnabled(a.TenantID, moduleID, "ip_range_import_csv")
+			s.enqueueModuleRefreshIfEnabled(w, r, a.TenantID, moduleID, "ip_range_import_csv")
 		}
 	}
 	writeJSON(w, http.StatusOK, map[string]any{
@@ -1055,7 +1064,7 @@ func (s *Server) handlePostPeer(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.peer.create", "Created BGP peer "+x.Name, x.ID, map[string]any{"peer_id": x.ID, "neighbor": x.Neighbor})
-	s.enqueuePeerReconcile(a.TenantID, "peer_create")
+	s.enqueuePeerReconcile(w, r, a.TenantID, "peer_create")
 	writeJSON(w, http.StatusCreated, peerJSON(x))
 }
 
@@ -1099,7 +1108,7 @@ func (s *Server) handlePatchPeer(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.peer.update", "Updated BGP peer "+x.Name, x.ID, map[string]any{"peer_id": x.ID, "neighbor": x.Neighbor})
-	s.enqueuePeerReconcile(a.TenantID, "peer_patch")
+	s.enqueuePeerReconcile(w, r, a.TenantID, "peer_patch")
 	writeJSON(w, http.StatusOK, peerJSON(x))
 }
 
@@ -1120,7 +1129,7 @@ func (s *Server) handleDeletePeer(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	s.recordCRUDAudit(r, a, "bgp.peer.delete", "Deleted BGP peer", peerID, map[string]any{"peer_id": peerID})
-	s.enqueuePeerReconcile(a.TenantID, "peer_delete")
+	s.enqueuePeerReconcile(w, r, a.TenantID, "peer_delete")
 	w.WriteHeader(http.StatusNoContent)
 }
 
@@ -1295,7 +1304,7 @@ func (s *Server) handlePatchSettings(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	if peerDiscoverySettingsChanged(body) {
-		s.enqueuePeerReconcile(a.TenantID, "peer_discovery_settings")
+		s.enqueuePeerReconcile(w, r, a.TenantID, "peer_discovery_settings")
 	}
 	s.recordCRUDAudit(r, a, "bgp.settings.update", "Updated tenant settings", a.TenantID, map[string]any{"keys": settingsAuditKeys(body)})
 	writeJSON(w, http.StatusOK, map[string]string{"status": "ok"})

internal/httpapi/routes_crud_csv_test.go

@@ -1,6 +1,7 @@
 package httpapi
 
 import (
+	"context"
 	"encoding/csv"
 	"encoding/json"
 	"io"
@@ -8,8 +9,60 @@ import (
 	"net/http/httptest"
 	"strings"
 	"testing"
+
+	"evobgp/internal/importer"
+	"evobgp/internal/store"
 )
 
+func TestCSVImportLimitsBeforeMutations(t *testing.T) {
+	m := store.NewMemory()
+	m.SeedDemo()
+	tenant, _, _, _, _ := m.DemoIDs()
+	mod, err := m.CreateModule(tenant, &store.Module{Type: "AS_PREFIXES", Name: "csv-limits"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	module := mod.ID
+	s := &Server{store: m}
+	before, err := m.ListASEntries(tenant, module)
+	if err != nil {
+		t.Fatal(err)
+	}
+	header := "asn,community\n"
+	validPrefix := header + "64512,\n"
+	oversized := validPrefix + strings.Repeat("\n", importer.MaxCSVBytes-len(validPrefix)) + "64513,\n"
+	for _, tc := range []struct {
+		name, body    string
+		unknownLength bool
+		status        int
+	}{
+		{"exact", header + strings.Repeat("\n", importer.MaxCSVBytes-len(header)), false, http.StatusOK},
+		{"oversize", oversized, false, http.StatusRequestEntityTooLarge},
+		{"unknown_length", oversized, true, http.StatusRequestEntityTooLarge},
+		{"rows", header + strings.Repeat("64512,\n", importer.MaxCSVRows+1), true, http.StatusRequestEntityTooLarge},
+		{"fields", header + "64512," + strings.Repeat(",", importer.MaxCSVFields), false, http.StatusRequestEntityTooLarge},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			req := httptest.NewRequest(http.MethodPost, "/v1/modules/"+module+"/entries.csv", strings.NewReader(tc.body))
+			if tc.unknownLength {
+				req.ContentLength = -1
+				req.TransferEncoding = []string{"chunked"}
+			}
+			req.SetPathValue("module_id", module)
+			req = req.WithContext(context.WithValue(req.Context(), authCtxKey, Auth{TenantID: tenant, Role: "operator"}))
+			w := httptest.NewRecorder()
+			s.handleImportModuleEntriesCSV(w, req)
+			if w.Code != tc.status {
+				t.Fatalf("status=%d body=%s", w.Code, w.Body.String())
+			}
+			after, e := m.ListASEntries(tenant, module)
+			if e != nil || len(after) != len(before) {
+				t.Fatalf("mutations on rejected/header-only request: count=%d err=%v", len(after), e)
+			}
+		})
+	}
+}
+
 func TestModuleEntriesCSVImportExportIPRanges(t *testing.T) {
 	srv, err := New(Options{
 		InsecureDev: true,

internal/httpapi/routes_lookup.go

@@ -24,6 +24,11 @@ func (s *Server) handleLookup(w http.ResponseWriter, r *http.Request) {
 		writeProblem(w, http.StatusBadRequest, "Bad Request", "query parameter q is required")
 		return
 	}
+	release, ok := s.acquireOperation(w, r, a.TenantID, lookupOperation)
+	if !ok {
+		return
+	}
+	defer release()
 	res, err := lookup.Lookup(r.Context(), s.store, a.TenantID, q)
 	if err != nil {
 		if errors.Is(err, store.ErrInvalidInput) {

internal/httpapi/routes_lookup_test.go

@@ -1,6 +1,7 @@
 package httpapi
 
 import (
+	"context"
 	"encoding/json"
 	"io"
 	"net/http"
@@ -31,7 +32,7 @@ func TestLookupMembershipHTTP(t *testing.T) {
 	}); err != nil {
 		t.Fatal(err)
 	}
-	if err := srv.Store().SetModulePrefixSnapshot(tenant, modIP, "t", []store.PrefixRow{
+	if err := srv.Store().SetModulePrefixSnapshot(context.Background(), tenant, modIP, "t", []store.PrefixRow{
 		{Prefix: "198.51.100.0/24", CommunityID: &cid, Source: "ip_range"},
 	}); err != nil {
 		t.Fatal(err)

internal/httpapi/routes_maintenance.go

@@ -242,10 +242,6 @@ func (s *Server) enqueueMaintenancePolicy(w http.ResponseWriter, r *http.Request
 		return
 	}
 	kind := "maintenance_policy_run"
-	if !s.checkPgMaintRateLimit(a.TenantID, kind+":"+policyID) {
-		writeProblem(w, http.StatusTooManyRequests, "Too Many Requests", "wait before repeating this maintenance operation")
-		return
-	}
 	idem := strings.TrimSpace(r.Header.Get("Idempotency-Key"))
 	var idemPtr *string
 	if idem != "" {
@@ -255,11 +251,18 @@ func (s *Server) enqueueMaintenancePolicy(w http.ResponseWriter, r *http.Request
 	if dryRun {
 		title = "Maintenance policy dry-run"
 	}
-	j, _, err := s.jobs.Enqueue(a.TenantID, jobs.KindMaintenancePolicyRun, idemPtr, nil, map[string]any{
+	if s.replayJobResponse(w, r, store.JobEnqueue{TenantID: a.TenantID, Kind: jobs.KindMaintenancePolicyRun, IdempotencyKey: idemPtr, Meta: map[string]any{"policy_id": policyID, "dry_run": dryRun, "actor_prefix": actorPrefix(a), "job_title": title}}) {
+		return
+	}
+	if !s.checkPgMaintRateLimit(a.TenantID, kind+":"+policyID) {
+		writeProblem(w, http.StatusTooManyRequests, "Too Many Requests", "wait before repeating this maintenance operation")
+		return
+	}
+	j, _, err := s.jobs.EnqueueContext(r.Context(), a.TenantID, jobs.KindMaintenancePolicyRun, idemPtr, nil, map[string]any{
 		"policy_id": policyID, "dry_run": dryRun, "actor_prefix": actorPrefix(a), "job_title": title,
 	})
 	if err != nil {
-		writeInternalError(w, "maintenance_policy_enqueue", err)
+		writeJobEnqueueError(w, err)
 		return
 	}
 	w.Header().Set("Location", "/v1/jobs/"+j.ID)

internal/httpapi/routes_postgres_maintenance.go

@@ -11,6 +11,7 @@ import (
 
 	"evobgp/internal/jobs"
 	"evobgp/internal/pgmonitor"
+	"evobgp/internal/store"
 )
 
 var (
@@ -68,19 +69,22 @@ func (s *Server) enqueuePostgresMaint(w http.ResponseWriter, r *http.Request, a
 	if !s.requirePostgres(w) || !s.requireOperatorStrict(w, a) {
 		return
 	}
-	if !s.checkPgMaintRateLimit(a.TenantID, kind) {
-		writeProblem(w, http.StatusTooManyRequests, "Too Many Requests", "wait before repeating this maintenance operation")
-		return
-	}
 	idem := strings.TrimSpace(r.Header.Get("Idempotency-Key"))
 	var idemPtr *string
 	if idem != "" {
 		idemPtr = &idem
 	}
 	meta["actor_prefix"] = actorPrefix(a)
-	j, _, err := s.jobs.Enqueue(a.TenantID, kind, idemPtr, nil, meta)
+	if s.replayJobResponse(w, r, store.JobEnqueue{TenantID: a.TenantID, Kind: kind, IdempotencyKey: idemPtr, Meta: meta}) {
+		return
+	}
+	if !s.checkPgMaintRateLimit(a.TenantID, kind) {
+		writeProblem(w, http.StatusTooManyRequests, "Too Many Requests", "wait before repeating this maintenance operation")
+		return
+	}
+	j, _, err := s.jobs.EnqueueContext(r.Context(), a.TenantID, kind, idemPtr, nil, meta)
 	if err != nil {
-		writeInternalError(w, "postgres_maint_enqueue", err)
+		writeJobEnqueueError(w, err)
 		return
 	}
 	w.Header().Set("Location", "/v1/jobs/"+j.ID)
@@ -184,12 +188,12 @@ func (s *Server) handlePostgresCleanup(w http.ResponseWriter, r *http.Request) {
 	if idem != "" {
 		idemPtr = &idem
 	}
-	j, _, err := s.jobs.Enqueue(a.TenantID, jobs.KindMaintenancePolicyRun, idemPtr, nil, map[string]any{
+	j, _, err := s.jobs.EnqueueContext(r.Context(), a.TenantID, jobs.KindMaintenancePolicyRun, idemPtr, nil, map[string]any{
 		"policy_id": policyID, "dry_run": body.DryRun, "actor_prefix": actorPrefix(a),
 		"job_title": "PostgreSQL cleanup (deprecated path)",
 	})
 	if err != nil {
-		writeInternalError(w, "postgres_maint_enqueue", err)
+		writeJobEnqueueError(w, err)
 		return
 	}
 	w.Header().Set("Location", "/v1/jobs/"+j.ID)

internal/httpapi/server.go

@@ -8,15 +8,17 @@ import (
 	"errors"
 	"net/http"
 	"strings"
+	"sync"
 	"time"
 
 	"evobgp/internal/audit"
 	"evobgp/internal/jobs"
+	"evobgp/internal/logging"
 	"evobgp/internal/maintenance"
 	"evobgp/internal/pgmonitor"
-	"evobgp/internal/repository"
 	"evobgp/internal/runtimelogs"
 	"evobgp/internal/store"
+	"evobgp/internal/telemetry"
 
 	"github.com/jackc/pgx/v5/pgxpool"
 )
@@ -30,7 +32,11 @@ type Server struct {
 	maintStats              *maintenance.DBStatsProvider
 	jobs                    *jobs.Registry
 	authLimiter             *authRateLimiter
+	operationOnce           sync.Once
+	operationLimiter        *operationLimiter
 	bundlePriv              ed25519.PrivateKey
+	bundleCache             *nodeBundleCache
+	nodeBundleBuild         nodeBundleBuilder
 	keyResolver             *apiKeyResolver
 	firewallResolver        *firewallTokenResolver
 	bundleSeedHex           string
@@ -39,6 +45,11 @@ type Server struct {
 	runtimeLogs             *runtimelogs.Service
 	runtimeLogsPolicyTenant string
 	mux                     *http.ServeMux
+	backgroundMu            sync.Mutex
+	backgroundStarted       bool
+	backgroundClosed        bool
+	backgroundCancel        context.CancelFunc
+	telemetryCollector      *telemetry.Collector
 
 	// Portal / dual-auth (JWT) configuration.
 	jwtSecret         string
@@ -118,6 +129,7 @@ func New(opts Options) (*Server, error) {
 		maintStats:              maintStats,
 		jobs:                    reg,
 		bundlePriv:              priv,
+		bundleCache:             newNodeBundleCache(nodeBundleCacheMaxEntries, nodeBundleCacheMaxBytes),
 		keyResolver:             resolver,
 		firewallResolver:        fwResolver,
 		bundleSeedHex:           strings.TrimSpace(opts.BundleSeedHex),
@@ -143,6 +155,29 @@ func New(opts Options) (*Server, error) {
 
 // Close releases database resources.
 func (s *Server) Close() {
+	if s == nil {
+		return
+	}
+	s.backgroundMu.Lock()
+	if s.backgroundClosed {
+		s.backgroundMu.Unlock()
+		return
+	}
+	s.backgroundClosed = true
+	cancel := s.backgroundCancel
+	collector := s.telemetryCollector
+	s.backgroundMu.Unlock()
+	if cancel != nil {
+		cancel()
+	}
+	if collector != nil {
+		_ = collector.Close(context.Background())
+	}
+	// Do not close the pool while a worker is still persisting its final state.
+	if s.jobs != nil {
+		s.jobs.RequestCancelAll()
+		_ = s.jobs.Drain(context.Background())
+	}
 	if s.pgPool != nil {
 		s.pgPool.Close()
 	}
@@ -154,26 +189,44 @@ func (s *Server) Store() store.Backend { return s.store }
 // Jobs exposes the in-process async job registry (for scheduler / evobgp-all).
 func (s *Server) Jobs() *jobs.Registry { return s.jobs }
 
-// StartBackground starts PostgreSQL monitoring and maintenance schedulers until ctx is cancelled.
+// StartBackground starts shared schedulers and telemetry polling once until ctx is cancelled.
 func (s *Server) StartBackground(ctx context.Context) {
-	if s != nil && s.pgPool != nil {
+	if s == nil {
-		pgmonitor.StartScheduler(ctx, s.pgPool)
+		return
+	}
+	s.backgroundMu.Lock()
+	defer s.backgroundMu.Unlock()
+	if s.backgroundClosed || s.backgroundStarted {
+		return
+	}
+	s.backgroundStarted = true
+	backgroundCtx, cancel := context.WithCancel(ctx)
+	s.backgroundCancel = cancel
+	if s.store != nil {
+		s.telemetryCollector = telemetry.Start(backgroundCtx, s.store, telemetry.ConfigFromEnv())
+	}
+	if s.pgPool != nil {
+		pgmonitor.StartScheduler(backgroundCtx, s.pgPool)
 		if s.jobs != nil {
-			jobs.StartDurableQueueLoop(ctx, s.jobs, repository.NewJobAuditWriter(s.pgPool))
+			jobs.StartDurableQueueLoop(backgroundCtx, s.jobs)
+			jobs.StartDeploymentReconciliationLoop(backgroundCtx, s.jobs, s.store)
 		}
 	}
 	if s != nil && s.maintConfig != nil && s.jobs != nil {
-		maintenance.StartScheduler(ctx, s.maintConfig, func(policyID string, dryRun bool, idem string) {
+		maintenance.StartScheduler(backgroundCtx, s.maintConfig, func(policyID string, dryRun bool, idem string) {
 			key := idem
-			_, _, _ = s.jobs.Enqueue("", jobs.KindMaintenancePolicyRun, &key, nil, map[string]any{
+			_, _, err := s.jobs.EnqueueContext(backgroundCtx, store.SystemJobTenantID, jobs.KindMaintenancePolicyRun, &key, nil, map[string]any{
 				"policy_id": policyID,
 				"dry_run":   dryRun,
 				"trigger":   "scheduler",
 			})
+			if err != nil {
+				logging.With("component", "maintenance.scheduler").Error("enqueue failed", "policy_id", policyID, "err", err)
+			}
 		}, 30*time.Second)
 	}
 	if s != nil && s.runtimeLogs != nil && s.store != nil {
-		runtimelogs.StartAutoCleanupScheduler(ctx, runtimelogs.SchedulerDeps{
+		runtimelogs.StartAutoCleanupScheduler(backgroundCtx, runtimelogs.SchedulerDeps{
 			Service:      s.runtimeLogs,
 			Store:        s.store,
 			PolicyTenant: s.runtimeLogsPolicyTenant,

internal/httpapi/speakers.go

@@ -6,7 +6,11 @@ import (
 	"encoding/hex"
 	"encoding/json"
 	"errors"
+	"net"
 	"net/http"
+	"net/netip"
+	"net/url"
+	"strconv"
 	"strings"
 	"time"
 
@@ -165,25 +169,66 @@ func publicControlPlaneURL(r *http.Request, override string) string {
 	if s := strings.TrimSpace(override); s != "" {
 		return strings.TrimRight(s, "/")
 	}
-	if origin := strings.TrimSpace(r.Header.Get("Origin")); strings.HasPrefix(origin, "http://") || strings.HasPrefix(origin, "https://") {
+	if r == nil {
-		return strings.TrimRight(origin, "/")
-	}
-	proto := strings.TrimSpace(r.Header.Get("X-Forwarded-Proto"))
-	if proto == "" {
-		proto = "https"
-	}
-	host := strings.TrimSpace(r.Header.Get("X-Forwarded-Host"))
-	if i := strings.Index(host, ","); i >= 0 {
-		host = strings.TrimSpace(host[:i])
-	}
-	if host == "" {
-		host = strings.TrimSpace(r.Host)
-	}
-	if host == "" {
 		return ""
 	}
+	if peer, ok := peerAddress(r); ok && isTrustedProxy(peer, trustedProxyPrefixes()) {
+		protos := r.Header.Values("X-Forwarded-Proto")
+		hosts := r.Header.Values("X-Forwarded-Host")
+		if len(protos) == 1 && len(hosts) == 1 {
+			proto := strings.TrimSpace(protos[0])
+			host := strings.TrimSpace(hosts[0])
+			if (proto == "http" || proto == "https") && validControlPlaneHost(host) {
 				return proto + "://" + host
 			}
+		}
+	}
+	if !validControlPlaneHost(r.Host) {
+		return ""
+	}
+	proto := "http"
+	if r.TLS != nil {
+		proto = "https"
+	}
+	return proto + "://" + r.Host
+}
+
+func validControlPlaneHost(host string) bool {
+	if host == "" || strings.TrimSpace(host) != host || strings.HasSuffix(host, ":") || strings.ContainsAny(host, " ,/@?#\\\t\r\n") {
+		return false
+	}
+	u, err := url.Parse("http://" + host)
+	if err != nil || u.Host != host || u.User != nil || u.Path != "" || u.RawQuery != "" || u.Fragment != "" {
+		return false
+	}
+	name := u.Hostname()
+	if name == "" {
+		return false
+	}
+	if strings.HasPrefix(host, "[") {
+		if _, err := netip.ParseAddr(name); err != nil {
+			return false
+		}
+	} else if net.ParseIP(name) == nil {
+		for _, label := range strings.Split(name, ".") {
+			if label == "" || label[0] == '-' || label[len(label)-1] == '-' {
+				return false
+			}
+			for _, c := range label {
+				if !(c >= 'a' && c <= 'z' || c >= 'A' && c <= 'Z' || c >= '0' && c <= '9' || c == '-') {
+					return false
+				}
+			}
+		}
+	}
+	if port := u.Port(); port != "" {
+		n, err := strconv.Atoi(port)
+		if err != nil || n < 1 || n > 65535 {
+			return false
+		}
+	}
+	return true
+}
 
 func (s *Server) attachReplicaInstall(resp map[string]any, sp *store.Speaker, nodeToken string, req speakerCreateRequest, r *http.Request) {
 	if s == nil || sp == nil || resp == nil {

internal/httpapi/speakers_live.go

@@ -1,9 +1,7 @@
 package httpapi
 
 import (
-	"context"
 	"strings"
-	"sync"
 	"time"
 
 	"evobgp/internal/birdfmt"
@@ -11,6 +9,34 @@ import (
 	"evobgp/internal/store"
 )
 
+type agentHealthPoll struct {
+	result              nodedispatch.AgentHealthResult
+	at                  time.Time
+	lastSuccessAt       time.Time
+	observationOverride *liveObservation
+}
+
+func agentHealthObservation(poll *agentHealthPoll) liveObservation {
+	if poll == nil {
+		return liveObservation{Source: "none", Quality: "unknown", ErrorCode: "not_polled"}
+	}
+	if poll.observationOverride != nil {
+		return *poll.observationOverride
+	}
+	o := liveObservation{ObservedAt: liveTime(poll.at), ReceivedAt: liveTime(poll.at), LastSuccessAt: liveTime(poll.lastSuccessAt), Source: "agent_health", Quality: "ok"}
+	if !poll.result.OK {
+		o.Quality = "unreachable"
+		if !poll.lastSuccessAt.IsZero() {
+			o.Quality = "stale"
+		}
+		o.ErrorCode = liveErrorCode(poll.result.Error)
+		if poll.result.Error == "agent reported ok=false" {
+			o.ErrorCode = "reported_unhealthy"
+		}
+	}
+	return o
+}
+
 func countBGPSessions(sessions []birdfmt.BGPSession) (total, established int) {
 	total = len(sessions)
 	for _, s := range sessions {
@@ -22,29 +48,48 @@ func countBGPSessions(sessions []birdfmt.BGPSession) (total, established int) {
 }
 
 func speakerLiveStatusJSON(sp *store.Speaker, view speakerBGPLive, health *nodedispatch.AgentHealthResult) map[string]any {
+	var poll *agentHealthPoll
+	if health != nil {
+		poll = &agentHealthPoll{result: *health}
+	}
+	return speakerLiveStatusWithPoll(sp, view, poll)
+}
+
+func speakerLiveStatusWithPoll(sp *store.Speaker, view speakerBGPLive, health *agentHealthPoll) map[string]any {
 	total, established := countBGPSessions(view.Sessions)
+	bgpObservation := view.observation()
+	agentObservation := agentHealthObservation(health)
+	if sp != nil && strings.EqualFold(strings.TrimSpace(sp.Role), "master") && health == nil {
+		agentObservation = bgpObservation
+	}
 	m := map[string]any{
 		"label":              view.Label,
-		"bgp_poll_ok":        view.Error == "",
+		"bgp_poll_ok":        view.successful(),
-		"bgp_sessions_total": total,
+		"bgp_sessions_total": nil,
-		"bgp_established":    established,
+		"bgp_established":    nil,
+		"bgp_observation":    bgpObservation,
+		"agent_observation":  agentObservation,
+	}
+	if view.successful() {
+		m["bgp_sessions_total"] = total
+		m["bgp_established"] = established
 	}
 	if view.Error != "" {
 		m["bgp_poll_error"] = view.Error
 	}
 	if health != nil {
-		m["agent_ok"] = health.OK
+		m["agent_ok"] = health.result.OK
-		if health.Error != "" {
+		if health.result.Error != "" {
-			m["agent_error"] = health.Error
+			m["agent_error"] = health.result.Error
 		}
-		if health.LastSyncAt != "" {
+		if health.result.LastSyncAt != "" {
-			m["agent_last_sync_at"] = health.LastSyncAt
+			m["agent_last_sync_at"] = health.result.LastSyncAt
 		}
-		if health.LastAppliedRevisionID != "" {
+		if health.result.LastAppliedRevisionID != "" {
-			m["agent_last_applied_revision_id"] = health.LastAppliedRevisionID
+			m["agent_last_applied_revision_id"] = health.result.LastAppliedRevisionID
 		}
 	} else if sp != nil && strings.EqualFold(strings.TrimSpace(sp.Role), "master") {
-		m["agent_ok"] = view.Error == ""
+		m["agent_ok"] = view.successful()
 		if view.Error != "" {
 			m["agent_error"] = view.Error
 		}
@@ -52,7 +97,7 @@ func speakerLiveStatusJSON(sp *store.Speaker, view speakerBGPLive, health *noded
 		m["agent_ok"] = false
 		m["agent_error"] = "agent health not polled"
 	}
-	if len(view.Sessions) > 0 {
+	if view.successful() {
 		sess := make([]map[string]any, 0, len(view.Sessions))
 		for _, s := range view.Sessions {
 			row := map[string]any{
@@ -68,62 +113,3 @@ func speakerLiveStatusJSON(sp *store.Speaker, view speakerBGPLive, health *noded
 	}
 	return m
 }
-
-func (s *Server) collectSpeakerLiveStatus(ctx context.Context, tenantID string, fresh bool, speakers []*store.Speaker) map[string]map[string]any {
-	views := s.collectSpeakerBGPLive(ctx, tenantID, fresh)
-	viewByID := make(map[string]speakerBGPLive, len(views))
-	for _, v := range views {
-		if v.SpeakerID != "" {
-			viewByID[v.SpeakerID] = v
-		}
-	}
-
-	opts := nodedispatch.Options{Timeout: 8 * time.Second}
-	type healthWrap struct {
-		id string
-		h  nodedispatch.AgentHealthResult
-	}
-	healthCh := make(chan healthWrap, len(speakers))
-	var wg sync.WaitGroup
-	for _, sp := range speakers {
-		if sp == nil {
-			continue
-		}
-		meta := store.ParseSpeakerMeta(sp.MetaJSON)
-		if !store.SpeakerNeedsRemoteDispatch(sp.Role, meta) {
-			continue
-		}
-		wg.Add(1)
-		go func(speaker *store.Speaker) {
-			defer wg.Done()
-			healthCh <- healthWrap{
-				id: speaker.ID,
-				h:  nodedispatch.FetchAgentHealth(ctx, speaker, opts),
-			}
-		}(sp)
-	}
-	wg.Wait()
-	close(healthCh)
-	healthByID := make(map[string]nodedispatch.AgentHealthResult, len(speakers))
-	for hw := range healthCh {
-		healthByID[hw.id] = hw.h
-	}
-
-	out := make(map[string]map[string]any, len(speakers))
-	for _, sp := range speakers {
-		if sp == nil {
-			continue
-		}
-		view, ok := viewByID[sp.ID]
-		if !ok {
-			view = speakerBGPLive{SpeakerID: sp.ID, Label: speakerDisplayLabel(sp)}
-		}
-		var hp *nodedispatch.AgentHealthResult
-		if h, ok := healthByID[sp.ID]; ok {
-			hCopy := h
-			hp = &hCopy
-		}
-		out[sp.ID] = speakerLiveStatusJSON(sp, view, hp)
-	}
-	return out
-}

internal/httpapi/speakers_live_test.go

@@ -2,12 +2,29 @@ package httpapi
 
 import (
 	"testing"
+	"time"
 
 	"evobgp/internal/birdfmt"
 	"evobgp/internal/nodedispatch"
 	"evobgp/internal/store"
 )
 
+func TestSpeakerLiveStatusJSONEmptyAndUnpolled(t *testing.T) {
+	sp := &store.Speaker{ID: "master", Role: "master"}
+	unpolled := speakerLiveStatusJSON(sp, speakerBGPLive{SpeakerID: sp.ID}, nil)
+	if unpolled["agent_ok"] != false || unpolled["bgp_poll_ok"] != false || unpolled["bgp_sessions_total"] != nil {
+		t.Fatalf("unpolled=%#v", unpolled)
+	}
+	if unpolled["bgp_observation"].(liveObservation).Quality != "unknown" {
+		t.Fatalf("unpolled observation=%#v", unpolled["bgp_observation"])
+	}
+	now := time.Date(2026, 10, 26, 10, 0, 0, 0, time.UTC)
+	empty := speakerLiveStatusJSON(sp, speakerBGPLive{SpeakerID: sp.ID, Polled: true, ObservedAt: now, ReceivedAt: now, LastSuccessAt: now, Source: "cp_birdc"}, nil)
+	if empty["agent_ok"] != true || empty["bgp_poll_ok"] != true || empty["bgp_sessions_total"] != 0 {
+		t.Fatalf("empty=%#v", empty)
+	}
+}
+
 func TestCountBGPSessions(t *testing.T) {
 	total, est := countBGPSessions([]birdfmt.BGPSession{
 		{Name: "p1", State: "Established"},

internal/httpapi/telemetry.go

@@ -0,0 +1,154 @@
+package httpapi
+
+import (
+	"context"
+	"net/http"
+	"time"
+
+	"evobgp/internal/nodedispatch"
+	"evobgp/internal/store"
+	"evobgp/internal/telemetry"
+)
+
+func (s *Server) loadTelemetry(ctx context.Context, tenant string, speakers []*store.Speaker) ([]speakerBGPLive, map[string]map[string]any, error) {
+	ctx, cancel := context.WithTimeout(ctx, 2*time.Second)
+	defer cancel()
+	snapshots, err := s.store.ListTelemetry(ctx, tenant)
+	if err != nil {
+		return nil, nil, err
+	}
+	views, statuses := hydrateTelemetry(speakers, snapshots, time.Now(), telemetry.ConfigFromEnv().StaleAfter)
+	return views, statuses, nil
+}
+
+// snapshotObservation uses control-plane receipt time, never the remote clock.
+func snapshotObservation(o store.TelemetryObservation, now time.Time, staleAfter time.Duration, invalid string) liveObservation {
+	v := liveObservation{ObservedAt: liveTime(o.ObservedAt), ReceivedAt: liveTime(o.ReceivedAt), LastSuccessAt: liveTime(o.LastSuccessAt), Source: o.Source, Quality: "unknown"}
+	if !o.ReceivedAt.IsZero() {
+		v.ExpiresAt = liveTime(o.ReceivedAt.Add(staleAfter))
+	}
+	if v.Source == "" {
+		v.Source = "none"
+	}
+	switch {
+	case invalid != "":
+		v.ErrorCode = invalid
+	case o.ReceivedAt.IsZero():
+		v.ErrorCode = "not_polled"
+	case o.ReceivedAt.After(now) || now.Sub(o.ReceivedAt) > staleAfter:
+		v.Quality, v.ErrorCode = "stale", "snapshot_expired"
+	case o.Error != "":
+		v.Quality, v.ErrorCode = "unreachable", liveErrorCode(o.Error)
+		if !o.LastSuccessAt.IsZero() {
+			v.Quality = "stale"
+		}
+	default:
+		v.Quality = "ok"
+	}
+	return v
+}
+
+func hydrateTelemetry(speakers []*store.Speaker, snapshots []store.TelemetrySnapshot, now time.Time, staleAfter time.Duration) ([]speakerBGPLive, map[string]map[string]any) {
+	byID := make(map[string]*store.Speaker, len(speakers))
+	for _, sp := range speakers {
+		if sp != nil {
+			byID[sp.ID] = sp
+		}
+	}
+	grouped := make(map[string][]store.TelemetrySnapshot)
+	for _, snapshot := range snapshots {
+		if snapshot.SpeakerID != "" && byID[snapshot.SpeakerID] == nil {
+			continue
+		}
+		if snapshot.SpeakerID == "" && snapshot.Kind != "local_bird" {
+			continue
+		}
+		grouped[snapshot.SpeakerID] = append(grouped[snapshot.SpeakerID], snapshot)
+	}
+	views := make([]speakerBGPLive, 0, len(speakers)+1)
+	statuses := make(map[string]map[string]any, len(speakers))
+	appendView := func(id string, sp *store.Speaker) {
+		view := speakerBGPLive{SpeakerID: id, Label: speakerDisplayLabel(sp)}
+		if sp == nil {
+			view.Label = "CP (local)"
+		}
+		var health *agentHealthPoll
+		rows := grouped[id]
+		// A restarted local collector may leave an expired authority behind.
+		// Only simultaneously current authorities conflict; never prefer one
+		// of two current BIRD nodes arbitrarily.
+		if len(rows) > 1 {
+			current := make([]store.TelemetrySnapshot, 0, len(rows))
+			for _, row := range rows {
+				if row.Kind == "local_bird" && row.ConfigFingerprint == store.TelemetrySpeakerFingerprint(sp) && !row.BGP.ReceivedAt.IsZero() && !row.BGP.ReceivedAt.After(now) && now.Sub(row.BGP.ReceivedAt) <= staleAfter {
+					current = append(current, row)
+				}
+			}
+			if len(current) > 0 {
+				rows = current
+			}
+		}
+		if len(rows) > 1 {
+			view.Error = "ambiguous telemetry authorities"
+			view.ObservationOverride = &liveObservation{Source: "none", Quality: "unknown", ErrorCode: "ambiguous_authority"}
+		} else if len(rows) == 1 {
+			row := rows[0]
+			invalid := ""
+			if row.ConfigFingerprint != store.TelemetrySpeakerFingerprint(sp) {
+				invalid = "configuration_changed"
+			}
+			bgp := snapshotObservation(row.BGP.TelemetryObservation, now, staleAfter, invalid)
+			view.ObservationOverride = &bgp
+			view.Polled = !row.BGP.ReceivedAt.IsZero()
+			view.ObservedAt, view.ReceivedAt, view.LastSuccessAt, view.Source = row.BGP.ObservedAt, row.BGP.ReceivedAt, row.BGP.LastSuccessAt, row.BGP.Source
+			if bgp.Quality == "ok" {
+				view.Sessions = row.BGP.Sessions
+			} else {
+				view.Error = bgp.ErrorCode
+			}
+			if row.Kind == "remote_agent" {
+				h := snapshotObservation(row.Health.TelemetryObservation, now, staleAfter, invalid)
+				result := nodedispatch.AgentHealthResult{OK: row.Health.OK && h.Quality == "ok", Error: row.Health.Error, LastSyncAt: row.Health.LastSyncAt, LastAppliedRevisionID: row.Health.LastAppliedRevisionID}
+				if h.Quality != "ok" {
+					result.Error = h.ErrorCode
+				}
+				if h.Quality == "ok" && !result.OK {
+					h.Quality, h.ErrorCode = "unreachable", "reported_unhealthy"
+				}
+				health = &agentHealthPoll{result: result, observationOverride: &h}
+			}
+		}
+		views = append(views, view)
+		if sp != nil {
+			statuses[id] = speakerLiveStatusWithPoll(sp, view, health)
+		}
+	}
+	for _, sp := range speakers {
+		if sp != nil {
+			appendView(sp.ID, sp)
+		}
+	}
+	if len(grouped[""]) > 0 {
+		appendView("", nil)
+	}
+	return views, statuses
+}
+
+func (s *Server) handleTelemetryRefresh(w http.ResponseWriter, r *http.Request) {
+	a, ok := authFromContext(r.Context())
+	if !ok {
+		writeProblem(w, http.StatusUnauthorized, "Unauthorized", "missing auth")
+		return
+	}
+	if !s.requirePerm(w, a, "bgp:network:read") {
+		return
+	}
+	ctx, cancel := context.WithTimeout(r.Context(), 2*time.Second)
+	defer cancel()
+	requested, err := s.store.RequestTelemetryRefresh(ctx, a.TenantID, 10*time.Second)
+	if err != nil {
+		writeProblem(w, http.StatusServiceUnavailable, "Service Unavailable", "telemetry refresh unavailable")
+		return
+	}
+	writeJSON(w, http.StatusAccepted, map[string]any{"requested": requested})
+}

internal/httpapi/telemetry_http_test.go

@@ -0,0 +1,126 @@
+package httpapi
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"evobgp/internal/nodedispatch"
+	"evobgp/internal/store"
+)
+
+type telemetryHTTPStore struct {
+	store.Backend
+	discoveryWrites int
+	snapshotErr     error
+	refreshCalls    int
+}
+
+func (s *telemetryHTTPStore) UpsertPeerDiscovery(tenant string, in *store.PeerDiscoveryUpsert) (*store.BGPPeerDiscovery, error) {
+	s.discoveryWrites++
+	return s.Backend.UpsertPeerDiscovery(tenant, in)
+}
+func (s *telemetryHTTPStore) ListTelemetry(ctx context.Context, tenant string) ([]store.TelemetrySnapshot, error) {
+	if s.snapshotErr != nil {
+		return nil, s.snapshotErr
+	}
+	return s.Backend.ListTelemetry(ctx, tenant)
+}
+func (s *telemetryHTTPStore) RequestTelemetryRefresh(ctx context.Context, tenant string, cooldown time.Duration) (bool, error) {
+	s.refreshCalls++
+	return s.Backend.RequestTelemetryRefresh(ctx, tenant, cooldown)
+}
+func telemetryRequest(method, path, tenant string) *http.Request {
+	r := httptest.NewRequest(method, path, nil)
+	return r.WithContext(context.WithValue(r.Context(), authCtxKey, Auth{TenantID: tenant, Role: "viewer"}))
+}
+
+func TestTelemetryGETsAndRefreshNeverPollAgents(t *testing.T) {
+	var calls atomic.Int64
+	agent := httptest.NewTLSServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		calls.Add(1)
+		_, _ = w.Write([]byte(`{"ok":true,"sessions":[]}`))
+	}))
+	defer agent.Close()
+	t.Setenv("EVOBGP_NODE_DISPATCH_INSECURE_TLS", "1")
+	mem := store.NewMemory()
+	mem.SeedDemo()
+	tenant, _, _, _, _ := mem.DemoIDs()
+	sp, err := mem.CreateSpeaker(tenant, &store.Speaker{Role: "replica", MetaJSON: store.SpeakerMetaJSON(store.SpeakerMeta{AgentDomain: strings.TrimPrefix(agent.URL, "https://"), AgentSecret: "test"})})
+	if err != nil {
+		t.Fatal(err)
+	}
+	// Positive control: the legacy adapter reaches the fake configured agent.
+	if result := nodedispatch.FetchAgentHealth(context.Background(), sp, nodedispatch.Options{HTTPClient: agent.Client()}); !result.OK {
+		t.Fatal(result)
+	}
+	calls.Store(0)
+	st := &telemetryHTTPStore{Backend: mem}
+	server := &Server{store: st}
+	claims, err := mem.ClaimTelemetry(context.Background(), "collector", 1, time.Minute, "")
+	if err != nil || len(claims) != 1 {
+		t.Fatalf("claim: %v %v", claims, err)
+	}
+	for i := 0; i < 20; i++ {
+		for _, tc := range []struct {
+			path    string
+			handler http.HandlerFunc
+		}{
+			{"/v1/peers", server.handleListPeers}, {"/v1/peers?live=1", server.handleListPeers},
+			{"/v1/speakers?live=1", server.handleListSpeakers}, {"/v1/peers/discovered?status=all", server.handleListPeerDiscoveries},
+		} {
+			w := httptest.NewRecorder()
+			tc.handler(w, telemetryRequest(http.MethodGet, tc.path, tenant))
+			if w.Code != http.StatusOK {
+				t.Fatalf("%s: %d %s", tc.path, w.Code, w.Body.String())
+			}
+		}
+		w := httptest.NewRecorder()
+		server.handleTelemetryRefresh(w, telemetryRequest(http.MethodPost, "/v1/telemetry/refresh", tenant))
+		if w.Code != http.StatusAccepted {
+			t.Fatalf("refresh: %d %s", w.Code, w.Body.String())
+		}
+		want := `"requested":false`
+		if i == 0 {
+			want = `"requested":true`
+		}
+		if !strings.Contains(w.Body.String(), want) {
+			t.Fatalf("refresh coalescing: %s", w.Body.String())
+		}
+	}
+	if calls.Load() != 0 || st.discoveryWrites != 0 {
+		t.Fatalf("GET/refresh caused IO: agent=%d discovery=%d", calls.Load(), st.discoveryWrites)
+	}
+}
+
+func TestTelemetryReadFailureAndRefreshPermission(t *testing.T) {
+	mem := store.NewMemory()
+	mem.SeedDemo()
+	tenant, _, _, _, _ := mem.DemoIDs()
+	st := &telemetryHTTPStore{Backend: mem, snapshotErr: errors.New("private database failure")}
+	server := &Server{store: st}
+	for _, handler := range []http.HandlerFunc{server.handleListPeers, server.handleListSpeakers} {
+		w := httptest.NewRecorder()
+		handler(w, telemetryRequest(http.MethodGet, "/v1/resource?live=1", tenant))
+		if w.Code != http.StatusServiceUnavailable || strings.Contains(w.Body.String(), "private") {
+			t.Fatalf("snapshot failure: %d %s", w.Code, w.Body.String())
+		}
+	}
+	r := httptest.NewRequest(http.MethodPost, "/v1/telemetry/refresh", nil)
+	w := httptest.NewRecorder()
+	server.handleTelemetryRefresh(w, r)
+	if w.Code != http.StatusUnauthorized || st.refreshCalls != 0 {
+		t.Fatal("unauthenticated refresh admitted")
+	}
+	r = r.WithContext(context.WithValue(r.Context(), authCtxKey, Auth{TenantID: tenant, Kind: AuthKindJWT}))
+	w = httptest.NewRecorder()
+	server.handleTelemetryRefresh(w, r)
+	if w.Code != http.StatusForbidden || st.refreshCalls != 0 {
+		t.Fatal("unprivileged refresh admitted")
+	}
+}

internal/httpapi/telemetry_lifecycle_test.go

@@ -0,0 +1,30 @@
+package httpapi
+
+import (
+	"context"
+	"testing"
+)
+
+func TestTelemetryBackgroundStartsOnceAndStopsBeforeClose(t *testing.T) {
+	s, err := New(Options{SeedDemo: true})
+	if err != nil {
+		t.Fatal(err)
+	}
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	s.StartBackground(ctx)
+	first := s.telemetryCollector
+	if first == nil {
+		t.Fatal("telemetry collector not started")
+	}
+	s.StartBackground(ctx)
+	if s.telemetryCollector != first {
+		t.Fatal("duplicate telemetry collector")
+	}
+	s.Close()
+	s.StartBackground(ctx)
+	if s.telemetryCollector != first {
+		t.Fatal("collector started after close")
+	}
+	s.Close()
+}

internal/httpapi/telemetry_test.go

@@ -0,0 +1,102 @@
+package httpapi
+
+import (
+	"evobgp/internal/birdfmt"
+	"evobgp/internal/store"
+	"testing"
+	"time"
+)
+
+func TestTelemetryHydrationFreshnessAndConfiguration(t *testing.T) {
+	now := time.Now().UTC()
+	sp := &store.Speaker{ID: "replica", Role: "replica", Endpoint: "old"}
+	observation := store.TelemetryObservation{ObservedAt: now.Add(-time.Hour), ReceivedAt: now, LastSuccessAt: now, Source: "agent"}
+	base := store.TelemetrySnapshot{SpeakerID: sp.ID, Kind: "remote_agent", ConfigFingerprint: store.TelemetrySpeakerFingerprint(sp), BGP: store.TelemetryBGP{TelemetryObservation: observation, Sessions: []birdfmt.BGPSession{{Name: "peer", State: "Established"}}}, Health: store.TelemetryHealth{TelemetryObservation: observation, OK: true}}
+	cases := []struct {
+		name        string
+		mutate      func(*store.TelemetrySnapshot)
+		bgp, health bool
+		quality     string
+	}{
+		{"fresh receipt despite remote clock", func(*store.TelemetrySnapshot) {}, true, true, "ok"},
+		{"expired BGP independent of health", func(s *store.TelemetrySnapshot) { s.BGP.ReceivedAt = now.Add(-2 * time.Minute) }, false, true, "stale"},
+		{"expired health independent of BGP", func(s *store.TelemetrySnapshot) { s.Health.ReceivedAt = now.Add(-2 * time.Minute) }, true, false, "ok"},
+		{"future receipt is not fresh", func(s *store.TelemetrySnapshot) { s.BGP.ReceivedAt = now.Add(time.Hour) }, false, true, "stale"},
+		{"configuration changed", func(s *store.TelemetrySnapshot) { s.ConfigFingerprint = "old" }, false, false, "unknown"},
+		{"failed BGP", func(s *store.TelemetrySnapshot) { s.BGP.Error = "timeout" }, false, true, "stale"},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			snapshot := base
+			tc.mutate(&snapshot)
+			views, statuses := hydrateTelemetry([]*store.Speaker{sp}, []store.TelemetrySnapshot{snapshot}, now, time.Minute)
+			status := statuses[sp.ID]
+			if status["bgp_poll_ok"] != tc.bgp || status["agent_ok"] != tc.health || views[0].observation().Quality != tc.quality {
+				t.Fatalf("status=%+v observation=%+v", status, views[0].observation())
+			}
+			if !tc.bgp && (status["bgp_established"] != nil || len(views[0].Sessions) != 0) {
+				t.Fatal("stale BGP exposed live sessions")
+			}
+		})
+	}
+}
+
+func TestTelemetryHydrationMissingAndAmbiguousLocal(t *testing.T) {
+	now := time.Now()
+	sp := &store.Speaker{ID: "master", Role: "master"}
+	for _, rows := range [][]store.TelemetrySnapshot{nil, {{SpeakerID: sp.ID, Kind: "local_bird", LocalAuthority: "a"}, {SpeakerID: sp.ID, Kind: "local_bird", LocalAuthority: "b"}}} {
+		views, statuses := hydrateTelemetry([]*store.Speaker{sp}, rows, now, time.Minute)
+		if views[0].observation().Quality != "unknown" || statuses[sp.ID]["agent_ok"] == true || statuses[sp.ID]["bgp_poll_ok"] == true {
+			t.Fatalf("unexpected healthy master: %+v", statuses)
+		}
+	}
+}
+
+func TestTelemetryHydrationLocalAuthorityRestart(t *testing.T) {
+	now := time.Now()
+	sp := &store.Speaker{ID: "master", Role: "master"}
+	fresh := store.TelemetrySnapshot{SpeakerID: sp.ID, Kind: "local_bird", LocalAuthority: "new", ConfigFingerprint: store.TelemetrySpeakerFingerprint(sp), BGP: store.TelemetryBGP{TelemetryObservation: store.TelemetryObservation{ReceivedAt: now}}}
+	old := fresh
+	old.LocalAuthority = "old"
+	old.BGP.ReceivedAt = now.Add(-2 * time.Minute)
+	_, status := hydrateTelemetry([]*store.Speaker{sp}, []store.TelemetrySnapshot{old, fresh}, now, time.Minute)
+	if status[sp.ID]["bgp_poll_ok"] != true {
+		t.Fatal("expired authority hides current local snapshot")
+	}
+	old.BGP.ReceivedAt = now
+	_, status = hydrateTelemetry([]*store.Speaker{sp}, []store.TelemetrySnapshot{old, fresh}, now, time.Minute)
+	if status[sp.ID]["bgp_poll_ok"] != false {
+		t.Fatal("two current local authorities are ambiguous")
+	}
+}
+
+func TestTelemetryPeerObservationPreservesUnknownConfiguration(t *testing.T) {
+	now := time.Now()
+	sp := &store.Speaker{ID: "replica", Role: "replica"}
+	row := store.TelemetrySnapshot{SpeakerID: sp.ID, Kind: "remote_agent", ConfigFingerprint: "previous", BGP: store.TelemetryBGP{TelemetryObservation: store.TelemetryObservation{ReceivedAt: now, LastSuccessAt: now}}}
+	views, _ := hydrateTelemetry([]*store.Speaker{sp}, []store.TelemetrySnapshot{row}, now, time.Minute)
+	observation := peerLiveObservation(views)
+	if observation.Quality != "unknown" || observation.ErrorCode != "configuration_changed" {
+		t.Fatalf("changed config is not a failed current poll: %+v", observation)
+	}
+}
+
+func TestTelemetryExpiryUsesConfiguredIntervalAndEarliestSource(t *testing.T) {
+	now := time.Now().UTC()
+	old := now.Add(-30 * time.Second)
+	a := snapshotObservation(store.TelemetryObservation{ReceivedAt: old}, now, 5*time.Minute, "")
+	b := snapshotObservation(store.TelemetryObservation{ReceivedAt: now}, now, 5*time.Minute, "")
+	want := old.Add(5 * time.Minute).Format(time.RFC3339Nano)
+	if a.ExpiresAt == nil || *a.ExpiresAt != want {
+		t.Fatalf("expiry ignores configured interval: %+v", a)
+	}
+	views := []speakerBGPLive{{Polled: true, ObservationOverride: &b}, {Polled: true, ObservationOverride: &a}}
+	combined := peerLiveObservation(views)
+	if combined.ExpiresAt == nil || *combined.ExpiresAt != want {
+		t.Fatalf("aggregate extends oldest source: %+v", combined)
+	}
+	missing := snapshotObservation(store.TelemetryObservation{}, now, time.Minute, "")
+	if missing.ExpiresAt != nil {
+		t.Fatal("unobserved source has expiry")
+	}
+}

internal/httpapi/trusted_proxy.go

@@ -0,0 +1,91 @@
+package httpapi
+
+import (
+	"net"
+	"net/http"
+	"net/netip"
+	"os"
+	"strings"
+)
+
+// trustedProxyPrefixes returns no trusted peers when any configured CIDR is invalid.
+// An empty configuration deliberately treats every connection as direct.
+func trustedProxyPrefixes() []netip.Prefix {
+	raw := strings.TrimSpace(os.Getenv("EVOBGP_TRUSTED_PROXY_CIDRS"))
+	if raw == "" {
+		return nil
+	}
+	parts := strings.Split(raw, ",")
+	prefixes := make([]netip.Prefix, 0, len(parts))
+	for _, part := range parts {
+		prefix, err := netip.ParsePrefix(strings.TrimSpace(part))
+		if err != nil {
+			return nil
+		}
+		if prefix.Addr().Is4In6() {
+			if prefix.Bits() < 96 {
+				return nil
+			}
+			prefix = netip.PrefixFrom(prefix.Addr().Unmap(), prefix.Bits()-96)
+		}
+		prefixes = append(prefixes, prefix.Masked())
+	}
+	return prefixes
+}
+
+func peerAddress(r *http.Request) (netip.Addr, bool) {
+	if r == nil {
+		return netip.Addr{}, false
+	}
+	host, _, err := net.SplitHostPort(strings.TrimSpace(r.RemoteAddr))
+	if err != nil {
+		host = strings.TrimSpace(r.RemoteAddr)
+	}
+	ip, err := netip.ParseAddr(host)
+	if err != nil {
+		return netip.Addr{}, false
+	}
+	return ip.WithZone("").Unmap(), true
+}
+
+func isTrustedProxy(ip netip.Addr, prefixes []netip.Prefix) bool {
+	for _, prefix := range prefixes {
+		if prefix.Contains(ip) || prefix.Contains(ip.Unmap()) {
+			return true
+		}
+	}
+	return false
+}
+
+func clientIPWithTrustedProxies(r *http.Request, prefixes []netip.Prefix) string {
+	peer, ok := peerAddress(r)
+	if !ok {
+		if r == nil {
+			return ""
+		}
+		return strings.TrimSpace(r.RemoteAddr)
+	}
+	if !isTrustedProxy(peer, prefixes) {
+		return peer.String()
+	}
+	values := r.Header.Values("X-Forwarded-For")
+	if len(values) != 1 || strings.TrimSpace(values[0]) == "" {
+		return peer.String()
+	}
+	hops := strings.Split(values[0], ",")
+	current := peer
+	for i := len(hops) - 1; i >= 0; i-- {
+		if !isTrustedProxy(current, prefixes) {
+			return current.String()
+		}
+		next, err := netip.ParseAddr(strings.TrimSpace(hops[i]))
+		if err != nil {
+			return peer.String()
+		}
+		if next.Zone() != "" {
+			return peer.String()
+		}
+		current = next.Unmap()
+	}
+	return current.String()
+}

internal/httpapi/trusted_proxy_test.go

@@ -0,0 +1,129 @@
+package httpapi
+
+import (
+	"fmt"
+	"net/http/httptest"
+	"net/netip"
+	"testing"
+	"time"
+)
+
+func TestClientIPTrustedProxyChain(t *testing.T) {
+	prefixes := []netip.Prefix{netip.MustParsePrefix("10.0.0.0/8")}
+	tests := []struct {
+		name   string
+		remote string
+		xff    string
+		want   string
+	}{
+		{"untrusted peer ignores spoof", "198.51.100.10:1234", "203.0.113.9", "198.51.100.10"},
+		{"rightmost untrusted hop", "10.0.0.2:1234", "203.0.113.9, 198.51.100.8", "198.51.100.8"},
+		{"trusted chain", "10.0.0.2:1234", "203.0.113.9, 10.0.0.3", "203.0.113.9"},
+		{"mapped peer", "[::ffff:10.0.0.2]:1234", "203.0.113.9", "203.0.113.9"},
+		{"mapped forwarded client", "10.0.0.2:1234", "::ffff:203.0.113.9", "203.0.113.9"},
+		{"malformed nearest hop", "10.0.0.2:1234", "203.0.113.9, bogus", "10.0.0.2"},
+		{"malformed farther hop ignored", "10.0.0.2:1234", "bogus, 198.51.100.8", "198.51.100.8"},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			r := httptest.NewRequest("GET", "http://example.test/", nil)
+			r.RemoteAddr = tt.remote
+			r.Header.Set("X-Forwarded-For", tt.xff)
+			if got := clientIPWithTrustedProxies(r, prefixes); got != tt.want {
+				t.Fatalf("client IP = %q, want %q", got, tt.want)
+			}
+		})
+	}
+}
+
+func TestClientIPConfigurationFailsClosed(t *testing.T) {
+	r := httptest.NewRequest("GET", "http://example.test/", nil)
+	r.RemoteAddr = "10.0.0.2:1234"
+	r.Header.Set("X-Forwarded-For", "203.0.113.9")
+	r.Header.Set("X-Real-IP", "203.0.113.8")
+	if got := clientIPWithTrustedProxies(r, nil); got != "10.0.0.2" {
+		t.Fatalf("direct client IP = %q", got)
+	}
+	t.Setenv("EVOBGP_TRUSTED_PROXY_CIDRS", "10.0.0.0/8, invalid")
+	if got := clientIP(r); got != "10.0.0.2" {
+		t.Fatalf("invalid CIDR client IP = %q", got)
+	}
+	t.Setenv("EVOBGP_TRUSTED_PROXY_CIDRS", "::ffff:10.0.0.2/128")
+	if got := clientIP(r); got != "203.0.113.9" {
+		t.Fatalf("mapped CIDR client IP = %q", got)
+	}
+}
+
+func TestPublicControlPlaneURLTrust(t *testing.T) {
+	t.Setenv("EVOBGP_TRUSTED_PROXY_CIDRS", "10.0.0.0/8")
+	r := httptest.NewRequest("POST", "http://internal.local:8080/v1/speakers", nil)
+	r.Header.Set("Origin", "https://evil.example")
+	r.Header.Set("X-Forwarded-Proto", "https")
+	r.Header.Set("X-Forwarded-Host", "bgp.example.test")
+	r.RemoteAddr = "198.51.100.8:1234"
+	if got := publicControlPlaneURL(r, ""); got != "http://internal.local:8080" {
+		t.Fatalf("direct URL = %q", got)
+	}
+	r.RemoteAddr = "10.0.0.2:1234"
+	if got := publicControlPlaneURL(r, ""); got != "https://bgp.example.test" {
+		t.Fatalf("proxied URL = %q", got)
+	}
+	if got := publicControlPlaneURL(r, "https://explicit.example/"); got != "https://explicit.example" {
+		t.Fatalf("override URL = %q", got)
+	}
+	for _, bad := range []struct{ proto, host string }{
+		{"javascript", "bgp.example.test"},
+		{"https,http", "bgp.example.test"},
+		{"https", "evil.example/path"},
+		{"https", "[email protected]"},
+		{"https", "evil.example,bgp.example.test"},
+		{"https", "bgp.example.test:99999"},
+		{"https", "bgp.example.test:"},
+	} {
+		r.Header.Set("X-Forwarded-Proto", bad.proto)
+		r.Header.Set("X-Forwarded-Host", bad.host)
+		if got := publicControlPlaneURL(r, ""); got != "http://internal.local:8080" {
+			t.Errorf("bad forwarded headers (%q, %q): %q", bad.proto, bad.host, got)
+		}
+	}
+	r.Header.Set("X-Forwarded-Proto", "https")
+	r.Header.Set("X-Forwarded-Host", "bgp.example.test")
+	r.Header.Add("X-Forwarded-Host", "evil.example")
+	if got := publicControlPlaneURL(r, ""); got != "http://internal.local:8080" {
+		t.Fatalf("duplicate forwarded host = %q", got)
+	}
+}
+
+func TestAuthRateLimiterManyIdentities(t *testing.T) {
+	l := &authRateLimiter{hits: make(map[string]authRateHit), limit: 1, window: time.Minute, maxClients: 32}
+	now := time.Unix(1000, 0)
+	for i := 0; i < 1000; i++ {
+		allowed := l.allowAt(fmt.Sprintf("ip-%d", i), now)
+		if allowed != (i < l.maxClients) {
+			t.Fatalf("identity %d allowed = %t", i, allowed)
+		}
+	}
+	if len(l.hits) != l.maxClients || l.allowAt("ip-0", now) {
+		t.Fatalf("active counters changed under cardinality pressure: %d", len(l.hits))
+	}
+}
+
+func TestAuthRateLimiterCapacityAndExpiry(t *testing.T) {
+	l := &authRateLimiter{hits: make(map[string]authRateHit), limit: 2, window: time.Minute, maxClients: 2}
+	now := time.Unix(1000, 0)
+	if !l.allowAt("a", now) || !l.allowAt("b", now) || !l.allowAt("a", now) {
+		t.Fatal("initial requests rejected")
+	}
+	if l.allowAt("a", now) || l.allowAt("c", now) {
+		t.Fatal("active limit or capacity bypassed")
+	}
+	if len(l.hits) != 2 {
+		t.Fatalf("map size = %d", len(l.hits))
+	}
+	if !l.allowAt("c", now.Add(time.Minute)) {
+		t.Fatal("expired entries did not release capacity")
+	}
+	if len(l.hits) != 1 {
+		t.Fatalf("expired entries remained: %d", len(l.hits))
+	}
+}

internal/httpclient/body.go

@@ -0,0 +1,31 @@
+package httpclient
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+)
+
+// ErrResponseTooLarge means the complete decoded HTTP response exceeded its budget.
+var ErrResponseTooLarge = errors.New("httpclient: response body exceeds limit")
+
+// ReadResponseBody reads a complete bounded body, including one extra byte to
+// distinguish an exact limit from truncation. Standard Transport gzip decoding
+// occurs before this reader, so the limit covers decompressed data as well.
+func ReadResponseBody(resp *http.Response, limit int64) ([]byte, error) {
+	if limit <= 0 || limit == int64(^uint64(0)>>1) {
+		return nil, fmt.Errorf("httpclient: invalid response limit")
+	}
+	if resp.ContentLength > limit {
+		return nil, ErrResponseTooLarge
+	}
+	body, err := io.ReadAll(io.LimitReader(resp.Body, limit+1))
+	if err != nil {
+		return nil, err
+	}
+	if int64(len(body)) > limit {
+		return nil, ErrResponseTooLarge
+	}
+	return body, nil
+}

internal/httpclient/body_test.go

@@ -0,0 +1,22 @@
+package httpclient
+
+import (
+	"errors"
+	"io"
+	"net/http"
+	"strings"
+	"testing"
+)
+
+func TestReadResponseBodyDoesNotAcceptValidPrefixTruncation(t *testing.T) {
+	for _, length := range []int64{-1, 9} {
+		resp := &http.Response{ContentLength: length, Body: io.NopCloser(strings.NewReader("{}       "))}
+		if _, err := ReadResponseBody(resp, 8); !errors.Is(err, ErrResponseTooLarge) {
+			t.Fatalf("length=%d got %v", length, err)
+		}
+	}
+	resp := &http.Response{ContentLength: -1, Body: io.NopCloser(strings.NewReader("{}      "))}
+	if b, err := ReadResponseBody(resp, 8); err != nil || len(b) != 8 {
+		t.Fatalf("exact limit failed: %q %v", b, err)
+	}
+}

internal/httpclient/circuit.go

@@ -1,8 +1,13 @@
 package httpclient
 
 import (
+	"errors"
+	"fmt"
+	"strings"
 	"sync"
 	"time"
+
+	"evobgp/internal/observability"
 )
 
 const (
@@ -16,14 +21,77 @@ type hostBreaker struct {
 	openUntil time.Time
 }
 
-var hostBreakers sync.Map // string -> *hostBreaker
+// ErrBreakerCapacity rejects new upstream identities when the bounded cache is full.
+var ErrBreakerCapacity = errors.New("httpclient: upstream circuit capacity exhausted")
 
-func breakerForHost(host string) *hostBreaker {
+type breakerEntry struct {
+	breaker  *hostBreaker
+	lastUsed time.Time
+	active   int
+}
+
+type breakerCache struct {
+	mu        sync.Mutex
+	entries   map[string]*breakerEntry
+	maximum   int
+	idle      time.Duration
+	nextSweep time.Time
+}
+
+var hostBreakers = breakerCache{entries: make(map[string]*breakerEntry), maximum: 1024, idle: 10 * time.Minute}
+
+func (c *breakerCache) sweep(now time.Time) {
+	for host, e := range c.entries {
+		if e.active != 0 || now.Before(e.lastUsed.Add(c.idle)) {
+			continue
+		}
+		e.breaker.mu.Lock()
+		open := now.Before(e.breaker.openUntil)
+		e.breaker.mu.Unlock()
+		if !open {
+			delete(c.entries, host)
+			observability.RemoveUpstreamBreaker(host)
+		}
+	}
+	c.nextSweep = now.Add(time.Second)
+}
+
+func (c *breakerCache) acquire(host string, now time.Time) (*hostBreaker, string, func(), error) {
+	if len(host) > 254 {
+		return nil, "", nil, fmt.Errorf("httpclient: upstream hostname too long")
+	}
+	host = strings.TrimSuffix(strings.ToLower(host), ".")
+	if len(host) > 253 {
+		return nil, "", nil, fmt.Errorf("httpclient: upstream hostname too long")
+	}
 	if host == "" {
 		host = "_"
 	}
-	v, _ := hostBreakers.LoadOrStore(host, &hostBreaker{})
+	c.mu.Lock()
-	return v.(*hostBreaker)
+	defer c.mu.Unlock()
+	if !now.Before(c.nextSweep) {
+		c.sweep(now)
+	}
+	e := c.entries[host]
+	if e == nil {
+		if len(c.entries) >= c.maximum {
+			return nil, host, nil, ErrBreakerCapacity
+		}
+		e = &breakerEntry{breaker: &hostBreaker{}}
+		c.entries[host] = e
+	}
+	e.active++
+	e.lastUsed = now
+	var once sync.Once
+	release := func() {
+		once.Do(func() {
+			c.mu.Lock()
+			defer c.mu.Unlock()
+			e.active--
+			e.lastUsed = time.Now()
+		})
+	}
+	return e.breaker, host, release, nil
 }
 
 func (b *hostBreaker) allow() bool {
@@ -51,23 +119,31 @@ func (b *hostBreaker) recordFailure() {
 
 // SnapshotBreakers returns whether each known host breaker is currently open.
 func SnapshotBreakers() map[string]bool {
+	hostBreakers.mu.Lock()
+	defer hostBreakers.mu.Unlock()
+	now := time.Now()
+	hostBreakers.sweep(now)
 	out := map[string]bool{}
-	hostBreakers.Range(func(key, value any) bool {
+	for host, e := range hostBreakers.entries {
-		host, _ := key.(string)
+		b := e.breaker
-		b, _ := value.(*hostBreaker)
-		if b == nil {
-			return true
-		}
 		b.mu.Lock()
 		open := time.Now().Before(b.openUntil)
 		b.mu.Unlock()
 		out[host] = open
-		return true
+	}
-	})
 	return out
 }
 
-// ResetHostBreakers clears all circuit breakers (tests only).
+// ResetHostBreakers clears idle circuit breakers (tests only). In-flight
+// references are retained so their completions cannot recreate orphan metrics.
 func ResetHostBreakers() {
-	hostBreakers = sync.Map{}
+	hostBreakers.mu.Lock()
+	defer hostBreakers.mu.Unlock()
+	for host, e := range hostBreakers.entries {
+		if e.active == 0 {
+			delete(hostBreakers.entries, host)
+			observability.RemoveUpstreamBreaker(host)
+		}
+	}
+	hostBreakers.nextSweep = time.Time{}
 }

internal/httpclient/circuit_cache_test.go

@@ -0,0 +1,118 @@
+package httpclient
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"net/http"
+	"strings"
+	"testing"
+	"time"
+
+	"evobgp/internal/observability"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+func TestBreakerCanonicalHostAndActiveRetention(t *testing.T) {
+	c := breakerCache{entries: make(map[string]*breakerEntry), maximum: 1, idle: time.Minute}
+	now := time.Now()
+	first, key, release, err := c.acquire("Cache-Test.Example.", now)
+	if err != nil || key != "cache-test.example" {
+		t.Fatalf("canonical key=%s err=%v", key, err)
+	}
+	second, _, release2, err := c.acquire("cache-test.example", now)
+	if err != nil || first != second || len(c.entries) != 1 {
+		t.Fatal("alias bypassed shared breaker")
+	}
+	observability.SetUpstreamBreakerOpen(key, true)
+	first.openUntil = now.Add(3 * time.Minute)
+	if _, _, _, err := c.acquire("other.example", now.Add(2*time.Minute)); !errors.Is(err, ErrBreakerCapacity) {
+		t.Fatalf("active entry evicted: %v", err)
+	}
+	release()
+	release()
+	release2()
+	if _, _, _, err := c.acquire("other.example", now.Add(2*time.Minute+2*time.Second)); !errors.Is(err, ErrBreakerCapacity) {
+		t.Fatalf("open breaker evicted: %v", err)
+	}
+	_, _, done, err := c.acquire("other.example", now.Add(4*time.Minute))
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer done()
+	if len(c.entries) != 1 || c.entries[key] != nil {
+		t.Fatal("idle closed entry retained")
+	}
+	families, err := prometheus.DefaultGatherer.Gather()
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, family := range families {
+		if strings.HasSuffix(family.GetName(), "upstream_breaker_open") {
+			for _, metric := range family.Metric {
+				for _, label := range metric.Label {
+					if label.GetName() == "host" && label.GetValue() == key {
+						t.Fatal("evicted host metric retained")
+					}
+				}
+			}
+		}
+	}
+}
+
+type breakerTransport func(*http.Request) (*http.Response, error)
+
+func (f breakerTransport) RoundTrip(r *http.Request) (*http.Response, error) { return f(r) }
+
+func TestBreakerCapacityRejectsBeforeDial(t *testing.T) {
+	ResetHostBreakers()
+	var releases []func()
+	defer func() {
+		for _, release := range releases {
+			release()
+		}
+		ResetHostBreakers()
+	}()
+	for i := 0; i < hostBreakers.maximum; i++ {
+		_, _, release, err := hostBreakers.acquire(fmt.Sprintf("capacity-%d.example", i), time.Now())
+		if err != nil {
+			t.Fatal(err)
+		}
+		releases = append(releases, release)
+	}
+	called := false
+	hc := &http.Client{Transport: breakerTransport(func(*http.Request) (*http.Response, error) { called = true; return nil, errors.New("unexpected dial") })}
+	req, _ := http.NewRequest(http.MethodGet, "https://overflow.example/", nil)
+	_, err := DoWithBreaker(context.Background(), hc, req, 1)
+	if !errors.Is(err, ErrBreakerCapacity) || called {
+		t.Fatalf("capacity err=%v dial=%t", err, called)
+	}
+	if len(SnapshotBreakers()) != hostBreakers.maximum {
+		t.Fatal("cache capacity changed")
+	}
+	ResetHostBreakers()
+	if len(SnapshotBreakers()) != hostBreakers.maximum {
+		t.Fatal("reset discarded in-flight references")
+	}
+}
+
+func TestBreakerRejectsOversizedIdentity(t *testing.T) {
+	c := breakerCache{entries: make(map[string]*breakerEntry), maximum: 1, idle: time.Minute}
+	if _, _, _, err := c.acquire(strings.Repeat("x", 254), time.Now()); err == nil || len(c.entries) != 0 {
+		t.Fatal("oversized identity retained")
+	}
+	name := strings.Repeat("a", 253)
+	first, key, release, err := c.acquire(name, time.Now())
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer release()
+	second, alias, release2, err := c.acquire(strings.ToUpper(name)+".", time.Now())
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer release2()
+	if first != second || key != alias {
+		t.Fatal("maximum length trailing-dot alias differs")
+	}
+}

internal/httpclient/httpclient.go

@@ -54,7 +54,9 @@ func DoWithRetry(ctx context.Context, hc *http.Client, req *http.Request, maxAtt
 			continue
 		}
 		if resp.StatusCode >= 500 {
-			_, _ = io.Copy(io.Discard, resp.Body)
+			// Draining is only an optimization for connection reuse, never an
+			// invitation to consume an unbounded error stream before retrying.
+			_, _ = io.Copy(io.Discard, io.LimitReader(resp.Body, 64<<10))
 			_ = resp.Body.Close()
 			lastErr = fmt.Errorf("httpclient: upstream %s", resp.Status)
 			continue
@@ -72,8 +74,11 @@ func DoWithBreaker(ctx context.Context, hc *http.Client, req *http.Request, maxA
 	if req == nil || req.URL == nil {
 		return nil, fmt.Errorf("httpclient: nil request")
 	}
-	host := req.URL.Hostname()
+	br, host, release, err := hostBreakers.acquire(req.URL.Hostname(), time.Now())
-	br := breakerForHost(host)
+	if err != nil {
+		return nil, err
+	}
+	defer release()
 	if !br.allow() {
 		observability.SetUpstreamBreakerOpen(host, true)
 		return nil, fmt.Errorf("httpclient: circuit open for %s", host)

internal/httpclient/retry_body_test.go

@@ -0,0 +1,44 @@
+package httpclient
+
+import (
+	"context"
+	"io"
+	"net/http"
+	"testing"
+)
+
+type endlessRetryBody struct {
+	read   int
+	closed bool
+}
+
+func (b *endlessRetryBody) Read(p []byte) (int, error) {
+	for i := range p {
+		p[i] = 'x'
+	}
+	b.read += len(p)
+	return len(p), nil
+}
+func (b *endlessRetryBody) Close() error { b.closed = true; return nil }
+
+type retryTransport func(*http.Request) (*http.Response, error)
+
+func (f retryTransport) RoundTrip(r *http.Request) (*http.Response, error) { return f(r) }
+
+func TestRetryDrainBoundsEndlessErrorBody(t *testing.T) {
+	b := &endlessRetryBody{}
+	c := &http.Client{Transport: retryTransport(func(*http.Request) (*http.Response, error) {
+		return &http.Response{StatusCode: 503, Status: "503 Service Unavailable", Body: b, Header: make(http.Header)}, nil
+	})}
+	req, err := http.NewRequest(http.MethodGet, "http://agent.example/bundle", nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err = DoWithRetry(context.Background(), c, req, 1); err == nil {
+		t.Fatal("upstream failure accepted")
+	}
+	if b.read != 64<<10 || !b.closed {
+		t.Fatalf("retry consumed %d bytes, closed=%v", b.read, b.closed)
+	}
+	var _ io.ReadCloser = b
+}

internal/importer/csv.go

@@ -1,7 +1,10 @@
 package importer
 
 import (
+	"bytes"
+	"context"
 	"encoding/csv"
+	"errors"
 	"fmt"
 	"io"
 	"strconv"
@@ -10,6 +13,18 @@ import (
 	"evobgp/internal/store"
 )
 
+// MaxCSVBytes is the complete request limit, including the optional header.
+const MaxCSVBytes = 8 << 20
+
+// MaxCSVRows bounds database work for a single import, excluding its header.
+const MaxCSVRows = 10000
+
+// MaxCSVFields bounds CSV parser allocations for delimiter-heavy records.
+const MaxCSVFields = 64
+
+// ErrCSVTooLarge means the entire import was rejected before creating entries.
+var ErrCSVTooLarge = errors.New("importer: csv exceeds byte, row or field limit")
+
 // Result of importing module entries from CSV.
 type Result struct {
 	Imported   int
@@ -17,19 +32,51 @@ type Result struct {
 }
 
 // ImportModuleEntriesCSV parses and creates entries for AS_PREFIXES, DOMAINS, or IP_RANGES modules.
-func ImportModuleEntriesCSV(st store.Backend, tenantID, moduleID string, body io.Reader) (*Result, error) {
+func ImportModuleEntriesCSV(ctx context.Context, st store.Backend, tenantID, moduleID string, body io.Reader) (*Result, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
 	mod, err := st.GetModule(tenantID, moduleID)
 	if err != nil {
 		return nil, err
 	}
 
-	cr := csv.NewReader(io.LimitReader(body, 8<<20))
+	// Read one byte beyond the cap: a valid prefix must never be imported as if
+	// it were the complete request. Check all resource limits before mutations.
+	raw, err := io.ReadAll(io.LimitReader(body, MaxCSVBytes+1))
+	if err != nil {
+		return nil, fmt.Errorf("importer: read csv: %w", err)
+	}
+	if len(raw) > MaxCSVBytes {
+		return nil, ErrCSVTooLarge
+	}
+	if err := checkCSVFields(ctx, raw); err != nil {
+		return nil, err
+	}
+	cr := csv.NewReader(bytes.NewReader(raw))
 	cr.TrimLeadingSpace = true
 	cr.FieldsPerRecord = -1
-	rows, err := cr.ReadAll()
+	rows := make([][]string, 0)
+	headerRows := 0
+	for {
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+		record, err := cr.Read()
+		if err == io.EOF {
+			break
+		}
 		if err != nil {
 			return nil, fmt.Errorf("importer: invalid csv: %w", err)
 		}
+		if len(rows) == 0 && isCSVHeader(record) {
+			headerRows = 1
+		}
+		if len(rows)-headerRows >= MaxCSVRows {
+			return nil, ErrCSVTooLarge
+		}
+		rows = append(rows, record)
+	}
 	if len(rows) == 0 {
 		return nil, fmt.Errorf("importer: %w", store.ErrInvalidInput)
 	}
@@ -63,18 +110,17 @@ func ImportModuleEntriesCSV(st store.Backend, tenantID, moduleID string, body io
 	}
 
 	start := 0
-	if len(rows[0]) >= 2 {
+	if isCSVHeader(rows[0]) {
-		key := strings.ToLower(strings.TrimSpace(rows[0][0]))
-		switch key {
-		case "asn", "domain", "iprange":
 		start = 1
 	}
-	}
 
 	imported := 0
 	switch mod.Type {
 	case "AS_PREFIXES":
 		for i := start; i < len(rows); i++ {
+			if err := ctx.Err(); err != nil {
+				return nil, err
+			}
 			rec := rows[i]
 			if len(rec) == 0 || (strings.TrimSpace(rec[0]) == "" && (len(rec) < 2 || strings.TrimSpace(rec[1]) == "")) {
 				continue
@@ -97,6 +143,9 @@ func ImportModuleEntriesCSV(st store.Backend, tenantID, moduleID string, body io
 		}
 	case "DOMAINS":
 		for i := start; i < len(rows); i++ {
+			if err := ctx.Err(); err != nil {
+				return nil, err
+			}
 			rec := rows[i]
 			if len(rec) == 0 || (strings.TrimSpace(rec[0]) == "" && (len(rec) < 2 || strings.TrimSpace(rec[1]) == "")) {
 				continue
@@ -119,6 +168,9 @@ func ImportModuleEntriesCSV(st store.Backend, tenantID, moduleID string, body io
 		}
 	case "IP_RANGES":
 		for i := start; i < len(rows); i++ {
+			if err := ctx.Err(); err != nil {
+				return nil, err
+			}
 			rec := rows[i]
 			if len(rec) == 0 || (strings.TrimSpace(rec[0]) == "" && (len(rec) < 2 || strings.TrimSpace(rec[1]) == "")) {
 				continue
@@ -145,3 +197,49 @@ func ImportModuleEntriesCSV(st store.Backend, tenantID, moduleID string, body io
 
 	return &Result{Imported: imported, ModuleType: mod.Type}, nil
 }
+
+func isCSVHeader(record []string) bool {
+	if len(record) < 2 {
+		return false
+	}
+	switch strings.ToLower(strings.TrimSpace(record[0])) {
+	case "asn", "domain", "iprange":
+		return true
+	default:
+		return false
+	}
+}
+
+// checkCSVFields runs before encoding/csv allocates a string/index for each field.
+// Quoted commas, escaped quotes and multiline fields don't start new fields.
+// Syntax validation remains the responsibility of encoding/csv below.
+func checkCSVFields(ctx context.Context, raw []byte) error {
+	quoted, fields := false, 1
+	for i := 0; i < len(raw); i++ {
+		if i%4096 == 0 {
+			if err := ctx.Err(); err != nil {
+				return err
+			}
+		}
+		switch raw[i] {
+		case '"':
+			if quoted && i+1 < len(raw) && raw[i+1] == '"' {
+				i++
+				continue
+			}
+			quoted = !quoted
+		case ',':
+			if !quoted {
+				fields++
+				if fields > MaxCSVFields {
+					return ErrCSVTooLarge
+				}
+			}
+		case '\n':
+			if !quoted {
+				fields = 1
+			}
+		}
+	}
+	return nil
+}

internal/importer/csv_test.go

@@ -0,0 +1,95 @@
+package importer
+
+import (
+	"context"
+	"errors"
+	"strings"
+	"testing"
+
+	"evobgp/internal/store"
+)
+
+type importBackend struct {
+	store.Backend
+	created  int
+	onCreate func()
+}
+
+func (s *importBackend) GetModule(string, string) (*store.Module, error) {
+	return &store.Module{Type: "AS_PREFIXES"}, nil
+}
+func (s *importBackend) ListCommunities(string) ([]*store.Community, error) { return nil, nil }
+func (s *importBackend) CreateASEntry(_, _ string, e *store.ASEntry) (*store.ASEntry, error) {
+	s.created++
+	if s.onCreate != nil {
+		s.onCreate()
+	}
+	return e, nil
+}
+
+func TestCSVByteLimitRejectsValidTruncatedPrefix(t *testing.T) {
+	header := "asn,community\n"
+	for _, extra := range []int{0, 1} {
+		s := &importBackend{}
+		body := header + strings.Repeat("\n", MaxCSVBytes-len(header)) + strings.Repeat("\n", extra)
+		_, err := ImportModuleEntriesCSV(context.Background(), s, "tenant", "module", strings.NewReader(body))
+		if extra == 0 && err != nil {
+			t.Fatalf("exact limit: %v", err)
+		}
+		if extra == 1 && !errors.Is(err, ErrCSVTooLarge) {
+			t.Fatalf("limit+1: %v", err)
+		}
+		if s.created != 0 {
+			t.Fatal("oversized import created entries")
+		}
+	}
+}
+
+func TestCSVRowLimitCheckedBeforeWrites(t *testing.T) {
+	for _, header := range []string{"", "asn,community\n"} {
+		for _, extra := range []int{0, 1} {
+			s := &importBackend{}
+			_, err := ImportModuleEntriesCSV(context.Background(), s, "tenant", "module", strings.NewReader(header+strings.Repeat("64512,\n", MaxCSVRows+extra)))
+			if extra == 0 && (err != nil || s.created != MaxCSVRows) {
+				t.Fatalf("exact rows: created=%d err=%v", s.created, err)
+			}
+			if extra == 1 && (!errors.Is(err, ErrCSVTooLarge) || s.created != 0) {
+				t.Fatalf("rows+1: created=%d err=%v", s.created, err)
+			}
+		}
+	}
+}
+
+func TestCSVCancellationStopsFurtherWrites(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	s := &importBackend{onCreate: cancel}
+	_, err := ImportModuleEntriesCSV(ctx, s, "tenant", "module", strings.NewReader("asn,community\n64512,\n64513,\n"))
+	if !errors.Is(err, context.Canceled) || s.created != 1 {
+		t.Fatalf("created=%d err=%v", s.created, err)
+	}
+}
+
+func TestCSVFieldLimitAndQuotedFields(t *testing.T) {
+	for _, tc := range []struct {
+		name, input string
+		oversized   bool
+	}{
+		{"exact", "64512," + strings.Repeat(",", MaxCSVFields-2) + "\n", false},
+		{"over", "64512," + strings.Repeat(",", MaxCSVFields-1) + "\n", true},
+		{"many_fields", strings.Repeat(",", 1<<20) + "\n", true},
+		{"quoted_extra", "64512,,\"" + strings.Repeat(",", 128) + "\nquoted \"\"text\"\"\"\n", false},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			s := &importBackend{}
+			_, err := ImportModuleEntriesCSV(context.Background(), s, "tenant", "module", strings.NewReader(tc.input))
+			if tc.oversized {
+				if !errors.Is(err, ErrCSVTooLarge) || s.created != 0 {
+					t.Fatalf("created=%d err=%v", s.created, err)
+				}
+			} else if err != nil || s.created != 1 {
+				t.Fatalf("created=%d err=%v", s.created, err)
+			}
+		})
+	}
+}

internal/jobs/deploy_confirmation_test.go

@@ -0,0 +1,231 @@
+package jobs
+
+import (
+	"context"
+	"encoding/json"
+	"io"
+	"net/http"
+	"strings"
+	"sync/atomic"
+	"testing"
+
+	"evobgp/internal/nodedispatch"
+	"evobgp/internal/store"
+
+	"github.com/google/uuid"
+)
+
+type deployRoundTrip func(*http.Request) (*http.Response, error)
+
+func (f deployRoundTrip) RoundTrip(r *http.Request) (*http.Response, error) { return f(r) }
+
+func TestDeployApplyConfirmation(t *testing.T) {
+	for _, tc := range []struct {
+		name, response, confirmation, status string
+		dispatch                             bool
+		stale                                bool
+		wantApplied                          bool
+	}{
+		{name: "dispatch disabled", confirmation: "unconfirmed", status: StatusSucceeded},
+		{name: "bare success", dispatch: true, response: `{"ok":true}`, confirmation: "unconfirmed", status: StatusSucceeded},
+		{name: "matching ACK", dispatch: true, response: "matching", confirmation: "agent_reported", status: StatusSucceeded, wantApplied: true},
+		{name: "mismatched ACK", dispatch: true, response: `{"ok":true,"applied_revision_id":"other"}`, confirmation: "unconfirmed", status: StatusFailed},
+		{name: "stale desired", dispatch: true, response: "matching", confirmation: "unconfirmed", status: StatusFailed, stale: true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Setenv("EVOBGP_BIRD_ACTIVE_DIR", "")
+			if tc.dispatch {
+				// A remote-only job must not touch the control plane BIRD tree.
+				t.Setenv("EVOBGP_BIRD_ACTIVE_DIR", t.TempDir())
+				t.Setenv("EVOBGP_NODE_DISPATCH_ENABLED", "1")
+			} else {
+				t.Setenv("EVOBGP_NODE_DISPATCH_ENABLED", "0")
+			}
+			m := store.NewMemory()
+			m.SeedDemo()
+			tenant, _, module, previous, speakerID := m.DemoIDs()
+			if err := m.SetLastAppliedRevision(tenant, speakerID, previous); err != nil {
+				t.Fatal(err)
+			}
+			revision := uuid.NewString()
+			if err := m.CreateRenderRevision(context.Background(), revision, tenant, module, &previous, "new-hash", map[string]string{"bird.conf": "# new\n"}, nil); err != nil {
+				t.Fatal(err)
+			}
+			newer := uuid.NewString()
+			if err := m.CreateRenderRevision(context.Background(), newer, tenant, module, &revision, "newer-hash", map[string]string{"bird.conf": "# newer\n"}, nil); err != nil {
+				t.Fatal(err)
+			}
+			meta := store.SpeakerMetaJSON(store.SpeakerMeta{AgentDomain: "agent.example.test", AgentSecret: "secret"})
+			if _, err := m.UpdateSpeaker(tenant, speakerID, &store.SpeakerPatch{MetaJSON: &meta}); err != nil {
+				t.Fatal(err)
+			}
+			client := &http.Client{Transport: deployRoundTrip(func(r *http.Request) (*http.Response, error) {
+				if r.Method != http.MethodPost || r.URL.Path != "/v1/agent/sync" || r.Header.Get("Authorization") != "Bearer secret" {
+					t.Errorf("unexpected sync request: %s %s", r.Method, r.URL)
+				}
+				var body map[string]string
+				if err := json.NewDecoder(r.Body).Decode(&body); err != nil || len(body) != 1 || body["revision_id"] != revision {
+					t.Errorf("legacy sync body: %#v, %v", body, err)
+				}
+				if tc.stale {
+					if err := m.PublishRevisionForSpeaker(speakerID, newer); err != nil {
+						t.Error(err)
+					}
+				}
+				response := tc.response
+				if response == "matching" {
+					response = `{"ok":true,"applied_revision_id":"` + revision + `"}`
+				}
+				return &http.Response{StatusCode: http.StatusOK, Body: io.NopCloser(strings.NewReader(response)), Header: make(http.Header)}, nil
+			})}
+			j := &Job{TenantID: tenant, Kind: KindDeployApply, Status: StatusQueued, Meta: map[string]any{"revision_id": revision, "speaker_id": speakerID}}
+			(&Worker{Store: m, NodeHTTPClient: client}).Process(j)
+			if j.Status != tc.status {
+				t.Fatalf("job status = %s, want %s; error=%v", j.Status, tc.status, j.Error)
+			}
+			sp, err := m.GetSpeaker(tenant, speakerID)
+			if err != nil {
+				t.Fatal(err)
+			}
+			want := previous
+			if tc.wantApplied {
+				want = revision
+			}
+			if sp.LastAppliedRevisionID == nil || *sp.LastAppliedRevisionID != want {
+				t.Fatalf("applied = %v, want %s", sp.LastAppliedRevisionID, want)
+			}
+			published, _, err := m.LatestPublishedRevision(speakerID)
+			if err != nil {
+				t.Fatal(err)
+			}
+			wantPublished := revision
+			if tc.stale {
+				wantPublished = newer
+			}
+			if published != wantPublished {
+				t.Fatalf("published = %s, want %s", published, wantPublished)
+			}
+			results := j.Meta["node_dispatch"].(map[string]any)["results"].([]nodedispatch.Result)
+			if len(results) != 1 || results[0].Confirmation != tc.confirmation {
+				t.Fatalf("results = %+v", results)
+			}
+		})
+	}
+}
+
+func TestDeployApplyPartialResult(t *testing.T) {
+	t.Setenv("EVOBGP_BIRD_ACTIVE_DIR", "")
+	t.Setenv("EVOBGP_NODE_DISPATCH_ENABLED", "1")
+	m := store.NewMemory()
+	m.SeedDemo()
+	tenant, _, module, previous, first := m.DemoIDs()
+	revision := uuid.NewString()
+	if err := m.CreateRenderRevision(context.Background(), revision, tenant, module, &previous, "new-hash", map[string]string{"bird.conf": "# new\n"}, nil); err != nil {
+		t.Fatal(err)
+	}
+	goodMeta := store.SpeakerMetaJSON(store.SpeakerMeta{AgentDomain: "good.example.test", AgentSecret: "secret"})
+	if _, err := m.UpdateSpeaker(tenant, first, &store.SpeakerPatch{MetaJSON: &goodMeta}); err != nil {
+		t.Fatal(err)
+	}
+	badMeta := store.SpeakerMetaJSON(store.SpeakerMeta{AgentDomain: "bad.example.test", AgentSecret: "secret"})
+	second, err := m.CreateSpeaker(tenant, &store.Speaker{Role: "replica", MetaJSON: badMeta})
+	if err != nil {
+		t.Fatal(err)
+	}
+	client := &http.Client{Transport: deployRoundTrip(func(r *http.Request) (*http.Response, error) {
+		response := `{"ok":true,"applied_revision_id":"` + revision + `"}`
+		if strings.HasPrefix(r.URL.Host, "bad.") {
+			response = `{"ok":false,"error":"apply failed"}`
+		}
+		return &http.Response{StatusCode: http.StatusOK, Body: io.NopCloser(strings.NewReader(response)), Header: make(http.Header)}, nil
+	})}
+	j := &Job{TenantID: tenant, Kind: KindDeployApply, Status: StatusQueued, Meta: map[string]any{"revision_id": revision}}
+	(&Worker{Store: m, NodeHTTPClient: client}).Process(j)
+	if j.Status != StatusFailed {
+		t.Fatalf("job status = %s, want failed", j.Status)
+	}
+	summary := j.Meta["apply_summary"].(map[string]any)
+	if summary["applied_count"] != 1 || summary["failed_count"] != 1 || summary["published_count"] != 2 {
+		t.Fatalf("partial summary = %#v", summary)
+	}
+	firstSp, _ := m.GetSpeaker(tenant, first)
+	secondSp, _ := m.GetSpeaker(tenant, second.ID)
+	if firstSp.LastAppliedRevisionID == nil || *firstSp.LastAppliedRevisionID != revision || secondSp.LastAppliedRevisionID != nil {
+		t.Fatalf("partial applied state: first=%v second=%v", firstSp.LastAppliedRevisionID, secondSp.LastAppliedRevisionID)
+	}
+	results := j.Meta["node_dispatch"].(map[string]any)["results"].([]nodedispatch.Result)
+	got := map[string]bool{}
+	for _, res := range results {
+		got[res.SpeakerID] = true
+	}
+	if len(results) != 2 || !got[first] || !got[second.ID] {
+		t.Fatalf("partial results=%+v", results)
+	}
+}
+
+func TestDeployApplyCancellationStopsFurtherPublication(t *testing.T) {
+	t.Setenv("EVOBGP_BIRD_ACTIVE_DIR", "")
+	t.Setenv("EVOBGP_NODE_DISPATCH_ENABLED", "1")
+	m := store.NewMemory()
+	m.SeedDemo()
+	tenant, _, module, previous, seeded := m.DemoIDs()
+	if err := m.DeleteSpeaker(tenant, seeded); err != nil {
+		t.Fatal(err)
+	}
+	revision := uuid.NewString()
+	if err := m.CreateRenderRevision(context.Background(), revision, tenant, module, &previous, "new-hash", map[string]string{"bird.conf": "# new\n"}, nil); err != nil {
+		t.Fatal(err)
+	}
+	meta := store.SpeakerMetaJSON(store.SpeakerMeta{AgentDomain: "agent.example.test", AgentSecret: "secret"})
+	speakerCount := speakerWakeLimit + 2
+	for range speakerCount {
+		if _, err := m.CreateSpeaker(tenant, &store.Speaker{Role: "replica", MetaJSON: meta}); err != nil {
+			t.Fatal(err)
+		}
+	}
+	speakers := m.ListSpeakersForTenant(tenant)
+	if len(speakers) != speakerCount {
+		t.Fatalf("speakers=%d, want %d", len(speakers), speakerCount)
+	}
+	var j *Job
+	var calls atomic.Int32
+	client := &http.Client{Transport: deployRoundTrip(func(r *http.Request) (*http.Response, error) {
+		// Cancel only once the in-flight wake limit is busy, so later speakers stay unpublished.
+		if calls.Add(1) == int32(speakerWakeLimit) {
+			j.RequestCancel()
+		}
+		<-r.Context().Done()
+		return nil, r.Context().Err()
+	})}
+	j = &Job{TenantID: tenant, Kind: KindDeployApply, Status: StatusQueued, Meta: map[string]any{"revision_id": revision}}
+	(&Worker{Store: m, NodeHTTPClient: client}).Process(j)
+	if j.Status != StatusCancelled || int(calls.Load()) != speakerWakeLimit {
+		t.Fatalf("cancelled job status=%s dispatch calls=%d", j.Status, calls.Load())
+	}
+	results := j.Meta["node_dispatch"].(map[string]any)["results"].([]nodedispatch.Result)
+	gotResult := map[string]bool{}
+	for _, res := range results {
+		gotResult[res.SpeakerID] = res.Status == "cancelled"
+	}
+	if len(results) != speakerWakeLimit {
+		t.Fatalf("cancelled results=%+v", results)
+	}
+	for i, sp := range speakers {
+		got, _, err := m.LatestPublishedRevision(sp.ID)
+		if i < speakerWakeLimit {
+			if err != nil || got != revision || !gotResult[sp.ID] {
+				t.Fatalf("speaker %s published=%s err=%v result=%v", sp.ID, got, err, gotResult[sp.ID])
+			}
+			if sp.LastAppliedRevisionID != nil {
+				t.Fatalf("cancelled speaker %s marked applied", sp.ID)
+			}
+			continue
+		}
+		if err == nil {
+			t.Fatalf("speaker %s published %s after cancellation", sp.ID, got)
+		}
+	}
+	if j.Meta["apply_summary"].(map[string]any)["published_count"] != speakerWakeLimit {
+		t.Fatalf("cancelled summary = %#v", j.Meta["apply_summary"])
+	}
+}

internal/jobs/deployment_reconcile.go

@@ -0,0 +1,126 @@
+package jobs
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"net/http"
+	"strconv"
+	"time"
+
+	"evobgp/internal/logging"
+	"evobgp/internal/nodedispatch"
+	"evobgp/internal/store"
+)
+
+func deploymentFenceForJob(j *Job) store.JobFence {
+	generation, _ := strconv.ParseInt(j.metaString("reconcile_generation"), 10, 64)
+	round, _ := strconv.Atoi(j.metaString("reconcile_round"))
+	return store.JobFence{JobID: j.ID, TenantID: j.TenantID, OwnerID: j.OwnerID, Token: j.FencingToken, ExpectedGeneration: generation, RetryRound: round}
+}
+
+func beginDeployment(ctx context.Context, st store.Deployments, f store.JobFence, speaker, revision string) (store.DeploymentSpeaker, error) {
+	for {
+		d, err := st.BeginDeploymentSpeaker(ctx, f, speaker, revision)
+		if !errors.Is(err, store.ErrDeploymentBusy) {
+			return d, err
+		}
+		timer := time.NewTimer(time.Second)
+		select {
+		case <-ctx.Done():
+			timer.Stop()
+			return store.DeploymentSpeaker{}, ctx.Err()
+		case <-timer.C:
+		}
+	}
+}
+
+// ReconcileDeployments schedules at most three recovery rounds for current desired
+// generations. Stable keys coalesce API replicas; the worker CAS checks generation
+// again before publishing, and preserves the original desired request sequence.
+func ReconcileDeployments(ctx context.Context, r *Registry, st store.Deployments) error {
+	rows, err := st.ListPendingDeployments(ctx, 100)
+	if err != nil {
+		return err
+	}
+	for _, d := range rows {
+		key := fmt.Sprintf("deployment_reconcile:%s:%s:%d", d.JobID, d.SpeakerID, d.RetryRound+1)
+		_, _, err = r.EnqueueContext(ctx, d.TenantID, KindDeployApply, &key, nil, map[string]any{"revision_id": d.RevisionID, "speaker_id": d.SpeakerID, "reconcile_generation": strconv.FormatInt(d.Generation, 10), "reconcile_round": strconv.Itoa(d.RetryRound + 1), "source_deployment_job_id": d.JobID})
+		if err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// StartDeploymentReconciliationLoop belongs to the API server lifecycle. Static
+// ingest/deploy/scheduler processes must not start this job-producing loop.
+func StartDeploymentReconciliationLoop(ctx context.Context, r *Registry, st store.Backend) {
+	if r == nil || r.queue == nil || st == nil {
+		return
+	}
+	r.durableMu.Lock()
+	if r.durableStop {
+		r.durableMu.Unlock()
+		return
+	}
+	r.durableLoopWG.Add(1)
+	r.durableMu.Unlock()
+	go func() {
+		defer r.durableLoopWG.Done()
+		ticker := time.NewTicker(10 * time.Second)
+		defer ticker.Stop()
+		for {
+			select {
+			case <-ctx.Done():
+				return
+			case <-ticker.C:
+			}
+			r.durableMu.Lock()
+			stopped := r.durableStop
+			r.durableMu.Unlock()
+			if stopped {
+				return
+			}
+			qctx, cancel := context.WithTimeout(ctx, 40*time.Second)
+			err := ObserveDeployments(qctx, st, nil)
+			if err == nil {
+				err = ReconcileDeployments(qctx, r, st)
+			}
+			cancel()
+			if err != nil && ctx.Err() == nil {
+				logging.With("component", "jobs.deployment_reconcile").Warn("reconciliation failed", "err", err)
+			}
+		}
+	}()
+}
+
+// ObserveDeployments samples compatible health before reconciling drift. The CP
+// receipt time and optional agent source time are stored separately; missing
+// legacy revision fields never manufacture an applied/verified confirmation.
+func ObserveDeployments(ctx context.Context, st store.Backend, client *http.Client) error {
+	rows, err := st.ListDeploymentObservations(ctx, 10)
+	if err != nil {
+		return err
+	}
+	for _, d := range rows {
+		sp, e := st.GetSpeaker(d.TenantID, d.SpeakerID)
+		if e != nil {
+			return e
+		}
+		observation := store.DeploymentObservation{Error: "remote health unavailable for this speaker"}
+		if nodedispatch.Enabled() && store.SpeakerNeedsRemoteDispatch(sp.Role, store.ParseSpeakerMeta(sp.MetaJSON)) {
+			hctx, cancel := context.WithTimeout(ctx, 3*time.Second)
+			h := nodedispatch.FetchAgentHealth(hctx, sp, nodedispatch.Options{HTTPClient: client, Timeout: 3 * time.Second})
+			cancel()
+			observation = store.DeploymentObservation{RevisionID: h.LastAppliedRevisionID, SourceAt: h.LastSyncAt, Error: h.Error}
+			if h.OK && observation.RevisionID == "" {
+				observation.Error = "legacy health did not report an applied revision"
+			}
+		}
+		if err = st.ObserveDeployment(ctx, d, observation); err != nil && !errors.Is(err, store.ErrJobLeaseLost) {
+			return err
+		}
+	}
+	return nil
+}

internal/jobs/deployment_reconcile_test.go

@@ -0,0 +1,140 @@
+package jobs
+
+import (
+	"context"
+	"io"
+	"net/http"
+	"strings"
+	"testing"
+	"time"
+
+	"evobgp/internal/store"
+)
+
+func TestDeploymentReconcileDedupAndDesiredOrdering(t *testing.T) {
+	t.Setenv("EVOBGP_NODE_DISPATCH_ENABLED", "0")
+	m := store.NewMemory()
+	m.SeedDemo()
+	tenant, _, _, revision, speaker := m.DemoIDs()
+	ctx := context.Background()
+	r := NewRegistry(nil)
+	r.UseDurableQueue(m)
+	initial, _, err := r.EnqueueContext(ctx, tenant, KindDeployApply, nil, nil, map[string]any{"revision_id": revision, "speaker_id": speaker})
+	if err != nil {
+		t.Fatal(err)
+	}
+	claim := func(owner string) *Job {
+		t.Helper()
+		rows, e := m.ClaimJobs(ctx, owner, 1, time.Minute)
+		if e != nil || len(rows) != 1 {
+			t.Fatalf("claim: %+v %v", rows, e)
+		}
+		return jobFromRecord(rows[0])
+	}
+	first := claim("initial")
+	if first.ID != initial.ID {
+		t.Fatal("wrong initial job")
+	}
+	(&Worker{Store: m}).Process(first)
+	if err = m.FinishJob(ctx, recordFromJob(first)); err != nil {
+		t.Fatal(err)
+	}
+	// Enqueue an operator request before the synthetic recovery request. The latter
+	// must retain the earlier desired order despite its own larger queue sequence.
+	operator, _, err := r.EnqueueContext(ctx, tenant, KindDeployApply, nil, nil, map[string]any{"revision_id": revision, "speaker_id": speaker})
+	if err != nil {
+		t.Fatal(err)
+	}
+	for range 2 {
+		if err = ReconcileDeployments(ctx, r, m); err != nil {
+			t.Fatal(err)
+		}
+	}
+	all, _, _, err := m.ListJobs(ctx, tenant, "", "", "", 100)
+	if err != nil || len(all) != 3 {
+		t.Fatalf("reconcile dedupe: %d %v", len(all), err)
+	}
+	newer := claim("operator")
+	if newer.ID != operator.ID {
+		t.Fatal("wrong operator job")
+	}
+	recovery := claim("recovery")
+	f := deploymentFenceForJob(recovery)
+	d, err := m.BeginDeploymentSpeaker(ctx, f, speaker, revision)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if d.RequestSequence != first.RequestSequence {
+		t.Fatal("recovery lost original desired order")
+	}
+	if err = m.RecordDeploymentSpeaker(ctx, f, d, "unconfirmed", "unconfirmed", ""); err != nil {
+		t.Fatal(err)
+	}
+	newD, err := m.BeginDeploymentSpeaker(ctx, deploymentFenceForJob(newer), speaker, revision)
+	if err != nil {
+		t.Fatalf("newer operator should supersede recovery despite smaller queue sequence: %v", err)
+	}
+	if newD.RequestSequence != newer.RequestSequence {
+		t.Fatal("operator desired sequence changed")
+	}
+	if _, err = m.BeginDeploymentSpeaker(ctx, f, speaker, revision); err == nil {
+		t.Fatal("stale recovery generation reused")
+	}
+}
+
+func TestDeploymentHealthDetectsDriftAndKeepsLegacyQuality(t *testing.T) {
+	t.Setenv("EVOBGP_NODE_DISPATCH_ENABLED", "0")
+	m := store.NewMemory()
+	m.SeedDemo()
+	tenant, _, _, revision, speaker := m.DemoIDs()
+	ctx := context.Background()
+	r := NewRegistry(nil)
+	r.UseDurableQueue(m)
+	_, _, err := r.EnqueueContext(ctx, tenant, KindDeployApply, nil, nil, map[string]any{"revision_id": revision, "speaker_id": speaker})
+	if err != nil {
+		t.Fatal(err)
+	}
+	rows, err := m.ClaimJobs(ctx, "owner", 1, time.Minute)
+	if err != nil || len(rows) != 1 {
+		t.Fatal(err)
+	}
+	j := jobFromRecord(rows[0])
+	(&Worker{Store: m}).Process(j)
+	if err = m.FinishJob(ctx, recordFromJob(j)); err != nil {
+		t.Fatal(err)
+	}
+	meta := store.SpeakerMetaJSON(store.SpeakerMeta{AgentDomain: "agent.example.test", AgentSecret: "secret"})
+	if _, err = m.UpdateSpeaker(tenant, speaker, &store.SpeakerPatch{MetaJSON: &meta}); err != nil {
+		t.Fatal(err)
+	}
+	t.Setenv("EVOBGP_NODE_DISPATCH_ENABLED", "1")
+	response := `{"ok":true,"last_applied_revision_id":"` + revision + `"}`
+	client := &http.Client{Transport: deployRoundTrip(func(req *http.Request) (*http.Response, error) {
+		if req.Header.Get("Authorization") != "Bearer secret" || req.Method != "GET" {
+			t.Error("health must be authenticated GET")
+		}
+		return &http.Response{StatusCode: 200, Body: io.NopCloser(strings.NewReader(response)), Header: make(http.Header)}, nil
+	})}
+	if err = ObserveDeployments(ctx, m, client); err != nil {
+		t.Fatal(err)
+	}
+	progress, err := m.ListDeploymentSpeakers(ctx, tenant, j.ID)
+	if err != nil || len(progress) != 1 || progress[0].State != "applied" || progress[0].Confirmation != "agent_reported" || progress[0].HealthReceivedAt == nil || progress[0].HealthSourceAt != "" {
+		t.Fatalf("legacy health confirmation: %+v %v", progress, err)
+	}
+	response = `{"ok":true,"last_applied_revision_id":"older-late-operation"}`
+	if err = ObserveDeployments(ctx, m, client); err != nil {
+		t.Fatal(err)
+	}
+	progress, _ = m.ListDeploymentSpeakers(ctx, tenant, j.ID)
+	if progress[0].State != "unknown" || progress[0].Confirmation != "unconfirmed" {
+		t.Fatalf("late drift hidden: %+v", progress)
+	}
+	if err = ReconcileDeployments(ctx, r, m); err != nil {
+		t.Fatal(err)
+	}
+	jobs, _, _, _ := m.ListJobs(ctx, tenant, "queued", "", "", 100)
+	if len(jobs) != 1 {
+		t.Fatalf("drift recovery jobs: %d", len(jobs))
+	}
+}

internal/jobs/durable.go

@@ -2,136 +2,216 @@ package jobs
 
 import (
 	"context"
-	"os"
+	"errors"
-	"strconv"
-	"strings"
-	"time"
-
 	"evobgp/internal/logging"
 	"evobgp/internal/observability"
-	"evobgp/internal/repository"
+	"evobgp/internal/store"
+	"sync/atomic"
+	"time"
 )
 
-// Adopt registers a durable job_audit claim into the in-process registry and starts the worker.
+const durableLease = time.Minute
-// Used after SKIP LOCKED claim so two API processes can share the queue without losing work on restart.
+
-func (r *Registry) Adopt(j *Job) bool {
+// StartDurableQueueLoop claims work only after reserving local capacity.
-	if r == nil || j == nil || j.ID == "" {
+func StartDurableQueueLoop(ctx context.Context, r *Registry) {
-		return false
+	if r == nil || r.queue == nil {
+		return
 	}
+	log := logging.With("component", "jobs.durable")
+	r.durableMu.Lock()
+	if r.durableStop {
+		r.durableMu.Unlock()
+		return
+	}
+	r.durableLoopWG.Add(2)
+	r.durableMu.Unlock()
+	go r.reapExpiredJobsLoop(ctx)
+	go func() {
+		defer r.durableLoopWG.Done()
+		ticker := time.NewTicker(time.Second)
+		defer ticker.Stop()
+		for {
+			if ctx.Err() != nil {
+				return
+			}
+			r.durableMu.Lock()
+			stopped := r.durableStop
+			r.durableMu.Unlock()
+			if stopped {
+				return
+			}
+			metricCtx, metricCancel := context.WithTimeout(ctx, 5*time.Second)
+			if counts, err := r.queue.CountJobsByStatus(metricCtx); err == nil {
+				observability.RecordJobAuditDepth(counts)
+			}
+			metricCancel()
+			free := 0
+		reserve:
+			for free < cap(r.workerSem) {
+				select {
+				case r.workerSem <- struct{}{}:
+					free++
+				default:
+					break reserve
+				}
+			}
+			if free > 0 {
+				qctx, cancel := context.WithTimeout(ctx, 10*time.Second)
+				claimed, err := r.queue.ClaimJobs(qctx, r.ownerID, free, durableLease)
+				cancel()
+				if err != nil {
+					log.Warn("claim failed", "err", err)
+				}
+				for i := len(claimed); i < free; i++ {
+					<-r.workerSem
+				}
+				observability.RecordJobQueueDepth(len(r.workerSem), cap(r.workerSem))
+				for _, c := range claimed {
+					r.durableMu.Lock()
+					stopped = r.durableStop
+					if stopped || ctx.Err() != nil {
+						r.durableMu.Unlock()
+						releaseCtx, releaseCancel := context.WithTimeout(context.Background(), 10*time.Second)
+						if err := r.queue.ReleaseJob(releaseCtx, c.ID, c.OwnerID, c.FencingToken); err != nil {
+							log.Warn("release during shutdown failed", "job_id", c.ID, "err", err)
+						}
+						releaseCancel()
+						<-r.workerSem
+						continue
+					}
+					j := jobFromRecord(c)
 					r.mu.Lock()
-	if _, exists := r.byID[j.ID]; exists {
-		r.mu.Unlock()
-		return false
-	}
-	if j.Status == "" {
-		j.Status = StatusQueued
-	}
-	if j.Meta == nil {
-		j.Meta = map[string]any{}
-	}
 					r.byID[j.ID] = j
-	if j.IdempotencyKey != nil && *j.IdempotencyKey != "" {
-		r.byIdempo[idempoKey{tenant: j.TenantID, key: *j.IdempotencyKey}] = j
-	}
 					if isRefreshKind(j.Kind) {
 						r.inflightRefresh[j.TenantID]++
 					}
-	workerStart := r.workerStart
 					r.mu.Unlock()
-
+					r.durableWG.Add(1)
-	if workerStart != nil {
+					go r.runClaimed(j)
-		go func() {
+					r.durableMu.Unlock()
-			r.workerSem <- struct{}{}
-			active := len(r.workerSem)
-			capacity := cap(r.workerSem)
-			observability.RecordJobQueueDepth(active, capacity)
-			defer func() {
-				<-r.workerSem
-				observability.RecordJobQueueDepth(len(r.workerSem), capacity)
-			}()
-			workerStart(j)
-		}()
-	}
-	return true
-}
-
-// StartDurableQueueLoop periodically reclaims stale running rows and claims queued job_audit work (PG SKIP LOCKED).
-// No-op when audit is nil. Interval from EVOBGP_JOB_RECLAIM_INTERVAL (default 20s); stale from EVOBGP_JOB_STALE_AFTER (default 15m).
-func StartDurableQueueLoop(ctx context.Context, reg *Registry, audit *repository.JobAuditWriter) {
-	if ctx == nil || reg == nil || audit == nil {
-		return
-	}
-	interval := 20 * time.Second
-	if s := strings.TrimSpace(os.Getenv("EVOBGP_JOB_RECLAIM_INTERVAL")); s != "" {
-		if d, err := time.ParseDuration(s); err == nil && d > 0 {
-			interval = d
 				}
 			}
-	staleAfter := 15 * time.Minute
-	if s := strings.TrimSpace(os.Getenv("EVOBGP_JOB_STALE_AFTER")); s != "" {
-		if d, err := time.ParseDuration(s); err == nil && d > 0 {
-			staleAfter = d
-		}
-	}
-	limit := 8
-	if n, err := strconv.Atoi(strings.TrimSpace(os.Getenv("EVOBGP_JOB_CLAIM_LIMIT"))); err == nil && n > 0 {
-		limit = n
-	}
-	grace := 30 * time.Second
-	if s := strings.TrimSpace(os.Getenv("EVOBGP_JOB_CLAIM_GRACE")); s != "" {
-		if d, err := time.ParseDuration(s); err == nil && d >= 0 {
-			grace = d
-		}
-	}
-	log := logging.With("component", "jobs.durable")
-	run := func() {
-		cctx, cancel := context.WithTimeout(ctx, 15*time.Second)
-		defer cancel()
-		n, err := audit.ReclaimStaleRunning(cctx, staleAfter)
-		if err != nil {
-			log.Warn("reclaim stale running failed", "err", err)
-		} else if n > 0 {
-			log.Info("reclaimed stale running jobs", "count", n)
-		}
-		claimed, err := audit.ClaimQueued(cctx, limit, grace)
-		if err != nil {
-			log.Warn("claim queued failed", "err", err)
-			return
-		}
-		for i := range claimed {
-			c := claimed[i]
-			j := &Job{
-				ID:             c.ID,
-				TenantID:       c.TenantID,
-				Kind:           c.Kind,
-				Status:         StatusQueued,
-				IdempotencyKey: c.IdempotencyKey,
-				ModuleID:       c.ModuleID,
-				CreatedAt:      c.CreatedAt,
-				Meta:           c.Meta,
-			}
-			if !reg.Adopt(j) {
-				// Already local — leave DB running; local worker owns it.
-				continue
-			}
-			log.Info("adopted durable job", "job_id", j.ID, "kind", j.Kind, "tenant_id", j.TenantID)
-		}
-		counts, err := audit.CountByStatus(cctx)
-		if err == nil {
-			observability.RecordJobAuditDepth(counts)
-		}
-	}
-	go func() {
-		run()
-		t := time.NewTicker(interval)
-		defer t.Stop()
-		for {
 			select {
 			case <-ctx.Done():
 				return
-			case <-t.C:
+			case <-ticker.C:
-				run()
 			}
 		}
 	}()
 }
+
+// reapExpiredJobsLoop runs the job_audit reaper every 15s, not on each claim.
+func (r *Registry) reapExpiredJobsLoop(ctx context.Context) {
+	defer r.durableLoopWG.Done()
+	r.reapExpiredJobs(ctx)
+	reap := time.NewTicker(15 * time.Second)
+	stop := time.NewTicker(time.Second)
+	defer reap.Stop()
+	defer stop.Stop()
+	for {
+		if r.durableStopped() || ctx.Err() != nil {
+			return
+		}
+		select {
+		case <-ctx.Done():
+			return
+		case <-stop.C:
+			if r.durableStopped() {
+				return
+			}
+		case <-reap.C:
+			r.reapExpiredJobs(ctx)
+		}
+	}
+}
+
+func (r *Registry) durableStopped() bool {
+	if r == nil {
+		return true
+	}
+	r.durableMu.Lock()
+	defer r.durableMu.Unlock()
+	return r.durableStop
+}
+
+func (r *Registry) reapExpiredJobs(ctx context.Context) {
+	if r == nil || r.queue == nil {
+		return
+	}
+	cctx, cancel := context.WithTimeout(ctx, 10*time.Second)
+	defer cancel()
+	if err := r.queue.ReapExpiredJobs(cctx); err != nil && ctx.Err() == nil {
+		logging.With("component", "jobs.durable").Warn("reap expired jobs", "err", err)
+	}
+}
+
+func (r *Registry) runClaimed(j *Job) {
+	log := logging.With("component", "jobs.durable")
+	var leaseHealthy atomic.Bool
+	leaseHealthy.Store(true)
+	defer func() {
+		r.mu.Lock()
+		delete(r.byID, j.ID)
+		r.mu.Unlock()
+		<-r.workerSem
+		observability.RecordJobQueueDepth(len(r.workerSem), cap(r.workerSem))
+		r.durableWG.Done()
+	}()
+	stop := make(chan struct{})
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		t := time.NewTicker(durableLease / 4)
+		defer t.Stop()
+		for {
+			select {
+			case <-stop:
+				return
+			case <-t.C:
+				ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+				requested, err := r.queue.HeartbeatJob(ctx, j.ID, j.OwnerID, j.FencingToken, durableLease)
+				if err == nil {
+					err = r.queue.CheckpointJob(ctx, recordFromJob(j))
+				}
+				cancel()
+				if err != nil {
+					leaseHealthy.Store(false)
+					j.RequestCancel()
+					log.Warn("heartbeat failed", "job_id", j.ID, "err", err)
+					return
+				}
+				if requested {
+					j.RequestCancel()
+				}
+			}
+		}
+	}()
+	r.workerStart(j)
+	close(stop)
+	<-done
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	r.durableMu.Lock()
+	stopped := r.durableStop
+	r.durableMu.Unlock()
+	c := recordFromJob(j)
+	if stopped && c.Status != StatusSucceeded && c.Status != StatusFailed {
+		if err := r.queue.ReleaseJob(ctx, j.ID, j.OwnerID, j.FencingToken); err != nil && !errors.Is(err, store.ErrJobLeaseLost) {
+			log.Warn("release failed", "err", err)
+		}
+		return
+	}
+	if !leaseHealthy.Load() {
+		// Leave an uncertain lease for expiry/reclaim. This consumes an attempt,
+		// whereas a deliberate graceful shutdown releases without using the budget.
+		return
+	}
+	if c.Status != StatusSucceeded && c.Status != StatusFailed && c.Status != StatusCancelled {
+		c.Status = StatusFailed
+		msg := "worker returned without terminal status"
+		c.Error = &msg
+	}
+	if err := r.queue.FinishJob(ctx, c); err != nil {
+		log.Warn("finish failed; lease will retry", "job_id", j.ID, "err", err)
+	}
+}

internal/jobs/durable_queue_test.go

@@ -0,0 +1,132 @@
+package jobs
+
+import (
+	"context"
+	"errors"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"evobgp/internal/store"
+)
+
+type blockingClaimQueue struct {
+	store.JobQueue
+	entered chan int
+	resume  chan struct{}
+}
+
+func (q *blockingClaimQueue) ClaimJobs(ctx context.Context, owner string, limit int, lease time.Duration) ([]store.JobRecord, error) {
+	q.entered <- limit
+	<-q.resume
+	return q.JobQueue.ClaimJobs(context.Background(), owner, limit, lease)
+}
+
+func TestDurableShutdownWaitsInflightClaimAndReleases(t *testing.T) {
+	t.Setenv("EVOBGP_JOB_MAX_CONCURRENT", "1")
+	m := store.NewMemory()
+	queued, _, err := m.EnqueueJob(context.Background(), store.JobEnqueue{TenantID: "tenant", Kind: "test"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	q := &blockingClaimQueue{JobQueue: m, entered: make(chan int, 1), resume: make(chan struct{})}
+	var runs atomic.Int32
+	r := NewRegistry(func(j *Job) { runs.Add(1); j.Succeed() })
+	r.UseDurableQueue(q)
+	StartDurableQueueLoop(context.Background(), r)
+	select {
+	case limit := <-q.entered:
+		if limit != cap(r.workerSem) {
+			t.Fatalf("claim capacity=%d reserved=%d", limit, cap(r.workerSem))
+		}
+	case <-time.After(3 * time.Second):
+		t.Fatal("no claim")
+	}
+	r.RequestCancelAll()
+	drained := make(chan error, 1)
+	go func() { drained <- r.Drain(context.Background()) }()
+	select {
+	case <-drained:
+		t.Fatal("drained before in-flight claim returned")
+	case <-time.After(20 * time.Millisecond):
+	}
+	close(q.resume)
+	select {
+	case err := <-drained:
+		if err != nil {
+			t.Fatal(err)
+		}
+	case <-time.After(3 * time.Second):
+		t.Fatal("shutdown did not drain")
+	}
+	if runs.Load() != 0 {
+		t.Fatal("started worker after stop")
+	}
+	got, err := m.GetJob(context.Background(), "tenant", queued.ID)
+	if err != nil || got.Status != "queued" || got.OwnerID != "" || got.Attempt != 0 {
+		t.Fatalf("not released: %+v %v", got, err)
+	}
+}
+
+func TestDurableSharedTerminalAndRequestReplay(t *testing.T) {
+	ctx := context.Background()
+	m := store.NewMemory()
+	key := "request"
+	a := NewRegistry(nil)
+	a.UseDurableQueue(m)
+	j, _, err := a.EnqueueContext(ctx, "tenant", "test", &key, nil, map[string]any{"nested": map[string]any{"input": "same"}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	claimed, err := m.ClaimJobs(ctx, "worker", 1, time.Minute)
+	if err != nil || len(claimed) != 1 {
+		t.Fatalf("claim %+v %v", claimed, err)
+	}
+	claimed[0].Status = StatusSucceeded
+	claimed[0].Meta = map[string]any{"result": "ok"}
+	if err = m.FinishJob(ctx, claimed[0]); err != nil {
+		t.Fatal(err)
+	}
+	b := NewRegistry(nil)
+	b.UseDurableQueue(m)
+	got, err := b.GetContext(ctx, "tenant", j.ID)
+	if err != nil || got.Status != StatusSucceeded {
+		t.Fatalf("shared get %+v %v", got, err)
+	}
+	list, _, _, err := b.ListContext(ctx, "tenant", StatusSucceeded, "", "", 10)
+	if err != nil || len(list) != 1 {
+		t.Fatalf("shared list %+v %v", list, err)
+	}
+	replay, created, err := b.EnqueueContext(ctx, "tenant", "test", &key, nil, map[string]any{"nested": map[string]any{"input": "same"}})
+	if err != nil || created || replay.ID != j.ID || replay.Status != StatusSucceeded {
+		t.Fatalf("terminal replay %+v %v %v", replay, created, err)
+	}
+	if _, _, err = b.EnqueueContext(ctx, "tenant", "test", &key, nil, map[string]any{"input": "different"}); !errors.Is(err, store.ErrJobConflict) {
+		t.Fatalf("changed payload accepted: %v", err)
+	}
+}
+
+func TestParentRetryRecoversCommittedDeployChild(t *testing.T) {
+	ctx := context.Background()
+	m := store.NewMemory()
+	r := NewRegistry(nil)
+	r.UseDurableQueue(m)
+	parent, _, err := r.EnqueueContext(ctx, "tenant", KindRevisionRollback, nil, nil, map[string]any{"revision_id": "original"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	key := "deploy_from:" + parent.ID
+	child, _, err := r.EnqueueContext(ctx, "tenant", KindDeployApply, &key, nil, map[string]any{"revision_id": "already-rendered"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	worker := &Worker{Store: m, Registry: r}
+	worker.Process(parent)
+	if parent.statusLocked() != StatusSucceeded || parent.metaString("revision_id") != "already-rendered" || parent.metaString("deploy_apply_job_id") != child.ID {
+		t.Fatalf("did not recover child: %+v", parent.Snapshot())
+	}
+	items, _, _, err := m.ListJobs(ctx, "tenant", "", KindDeployApply, "", 10)
+	if err != nil || len(items) != 1 {
+		t.Fatalf("duplicate child: %+v %v", items, err)
+	}
+}

internal/jobs/job.go

@@ -11,6 +11,7 @@ import (
 	"time"
 
 	"evobgp/internal/observability"
+	"evobgp/internal/store"
 
 	"github.com/google/uuid"
 )
@@ -38,6 +39,11 @@ type Job struct {
 	Error           *string
 	ProgressPct     *int16
 	Meta            map[string]any
+	OwnerID         string
+	FencingToken    int64
+	RequestSequence int64
+	Attempt         int
+	requestHash     string
 	cancelRequested bool
 	mu              sync.Mutex
 }
@@ -225,6 +231,30 @@ type Registry struct {
 	// have been enqueued but not yet finalized in finishModuleRefreshSuccess. Used for deterministic
 	// deploy coalescing under tenantRefreshMu (instead of polling job statuses).
 	inflightRefresh map[string]int
+	queue           store.JobQueue
+	ownerID         string
+	durableMu       sync.Mutex
+	durableWG       sync.WaitGroup
+	durableLoopWG   sync.WaitGroup
+	durableStop     bool
+}
+
+// UseDurableQueue makes the shared store authoritative for new work and job reads.
+func (r *Registry) UseDurableQueue(q store.JobQueue) {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	r.queue = q
+	r.ownerID = uuid.NewString()
+}
+
+func jobFromRecord(c store.JobRecord) *Job {
+	return &Job{ID: c.ID, TenantID: c.TenantID, Kind: c.Kind, Status: c.Status, IdempotencyKey: c.IdempotencyKey, ModuleID: c.ModuleID, Meta: c.Meta, CreatedAt: c.CreatedAt, StartedAt: c.StartedAt, FinishedAt: c.FinishedAt, Error: c.Error, OwnerID: c.OwnerID, FencingToken: c.FencingToken, RequestSequence: c.RequestSequence, Attempt: c.Attempt, cancelRequested: c.CancelRequested}
+}
+
+func recordFromJob(j *Job) store.JobRecord {
+	j.mu.Lock()
+	defer j.mu.Unlock()
+	return store.JobRecord{ID: j.ID, TenantID: j.TenantID, Kind: j.Kind, Status: j.Status, IdempotencyKey: j.IdempotencyKey, ModuleID: j.ModuleID, Meta: cloneMeta(j.Meta), CreatedAt: j.CreatedAt, StartedAt: j.StartedAt, FinishedAt: j.FinishedAt, Error: j.Error, OwnerID: j.OwnerID, FencingToken: j.FencingToken, RequestSequence: j.RequestSequence, Attempt: j.Attempt, CancelRequested: j.cancelRequested}
 }
 
 type idempoKey struct {
@@ -310,6 +340,24 @@ func registryMaxConcurrentJobs() int {
 	return 8
 }
 
+func jobRegistryPendingLimit() int {
+	if n, err := strconv.Atoi(strings.TrimSpace(os.Getenv("EVOBGP_JOB_MAX_PENDING"))); err == nil && n > 0 {
+		return n
+	}
+	return 10000
+}
+
+func (r *Registry) ActiveCountUnsafe() int {
+	n := 0
+	for _, j := range r.byID {
+		st := j.statusLocked()
+		if st == StatusQueued || st == StatusRunning {
+			n++
+		}
+	}
+	return n
+}
+
 // pruneTerminalIfOver удаляет самые старые завершённые джобы (succeeded/failed/cancelled), пока len(byID) > maxJobs.
 func (r *Registry) pruneTerminalIfOver(maxJobs int) {
 	if r == nil || maxJobs <= 0 || len(r.byID) <= maxJobs {
@@ -352,6 +400,33 @@ func (r *Registry) pruneTerminalIfOver(maxJobs int) {
 
 // Enqueue creates a job or returns an existing one for the same idempotency key.
 func (r *Registry) Enqueue(tenantID, kind string, idempotencyKey *string, moduleID *string, meta map[string]any) (*Job, bool, error) {
+	return r.EnqueueContext(context.Background(), tenantID, kind, idempotencyKey, moduleID, meta)
+}
+
+// EnqueueContext commits a durable job before returning it to an HTTP caller.
+func (r *Registry) EnqueueContext(ctx context.Context, tenantID, kind string, idempotencyKey *string, moduleID *string, meta map[string]any) (*Job, bool, error) {
+	return r.EnqueueRequestContext(ctx, tenantID, kind, idempotencyKey, moduleID, meta, nil)
+}
+
+// EnqueueRequestContext preserves idempotency of dynamically resolved selectors.
+func (r *Registry) EnqueueRequestContext(ctx context.Context, tenantID, kind string, idempotencyKey *string, moduleID *string, meta, request map[string]any) (*Job, bool, error) {
+	if r.queue != nil {
+		r.durableMu.Lock()
+		stopped := r.durableStop
+		r.durableMu.Unlock()
+		if stopped {
+			return nil, false, fmt.Errorf("jobs: shutting down")
+		}
+		c, created, err := r.queue.EnqueueJob(ctx, store.JobEnqueue{TenantID: tenantID, Kind: kind, IdempotencyKey: idempotencyKey, ModuleID: moduleID, Meta: meta, Request: request})
+		if err != nil {
+			return nil, false, err
+		}
+		return jobFromRecord(c), created, nil
+	}
+	hash, err := store.JobRequestHash(store.JobEnqueue{TenantID: tenantID, Kind: kind, IdempotencyKey: idempotencyKey, ModuleID: moduleID, Meta: meta, Request: request})
+	if err != nil {
+		return nil, false, err
+	}
 	r.mu.Lock()
 	maxJobs := registryMaxJobsFromEnv()
 	r.pruneTerminalIfOver(maxJobs)
@@ -359,13 +434,16 @@ func (r *Registry) Enqueue(tenantID, kind string, idempotencyKey *string, module
 	if idempotencyKey != nil && *idempotencyKey != "" {
 		k := idempoKey{tenant: tenantID, key: *idempotencyKey}
 		if existing, ok := r.byIdempo[k]; ok {
-			st := existing.statusLocked()
-			if st == StatusQueued || st == StatusRunning {
 			r.mu.Unlock()
+			if existing.requestHash != hash {
+				return nil, false, store.ErrJobConflict
+			}
 			return existing, false, nil
 		}
-			delete(r.byIdempo, k)
 	}
+	if r.ActiveCountUnsafe() >= jobRegistryPendingLimit() {
+		r.mu.Unlock()
+		return nil, false, store.ErrJobQueueFull
 	}
 
 	j := &Job{
@@ -377,6 +455,7 @@ func (r *Registry) Enqueue(tenantID, kind string, idempotencyKey *string, module
 		ModuleID:       moduleID,
 		CreatedAt:      time.Now().UTC(),
 		Meta:           cloneMeta(meta),
+		requestHash:    hash,
 	}
 	if idempotencyKey != nil && *idempotencyKey != "" {
 		r.byIdempo[idempoKey{tenant: tenantID, key: *idempotencyKey}] = j
@@ -571,6 +650,11 @@ func (r *Registry) RequestCancelAll() int {
 	if r == nil {
 		return 0
 	}
+	if r.queue != nil {
+		r.durableMu.Lock()
+		r.durableStop = true
+		r.durableMu.Unlock()
+	}
 	r.mu.RLock()
 	defer r.mu.RUnlock()
 	n := 0
@@ -585,6 +669,9 @@ func (r *Registry) RequestCancelAll() int {
 		j.RequestCancel()
 		n++
 	}
+	if r.queue != nil && len(r.workerSem) > n {
+		n = len(r.workerSem)
+	}
 	return n
 }
 
@@ -614,6 +701,16 @@ func (r *Registry) Drain(ctx context.Context) error {
 	if r == nil {
 		return nil
 	}
+	if r.queue != nil {
+		done := make(chan struct{})
+		go func() { r.durableLoopWG.Wait(); r.durableWG.Wait(); close(done) }()
+		select {
+		case <-done:
+			return nil
+		case <-ctx.Done():
+			return ctx.Err()
+		}
+	}
 	ticker := time.NewTicker(50 * time.Millisecond)
 	defer ticker.Stop()
 	for {

internal/jobs/registry_shared.go

@@ -0,0 +1,115 @@
+package jobs
+
+import (
+	"context"
+	"errors"
+
+	"evobgp/internal/store"
+)
+
+// FindIdempotentContext looks up a retained logical operation without inserting.
+func (r *Registry) FindIdempotentContext(ctx context.Context, tenant, key string) (*Job, error) {
+	if r.queue != nil {
+		c, err := r.queue.FindJobByIdempotencyKey(ctx, tenant, key)
+		if err != nil {
+			return nil, err
+		}
+		j := jobFromRecord(c)
+		j.requestHash = c.RequestHash
+		return j, nil
+	}
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	j := r.byIdempo[idempoKey{tenant: tenant, key: key}]
+	if j == nil {
+		return nil, store.ErrJobNotFound
+	}
+	return j, nil
+}
+
+// ReplayContext validates an immutable request before expensive live resolution.
+func (r *Registry) ReplayContext(ctx context.Context, in store.JobEnqueue) (*Job, bool, error) {
+	if in.IdempotencyKey == nil || *in.IdempotencyKey == "" {
+		return nil, false, nil
+	}
+	j, err := r.FindIdempotentContext(ctx, in.TenantID, *in.IdempotencyKey)
+	if errors.Is(err, store.ErrJobNotFound) {
+		return nil, false, nil
+	}
+	if err != nil {
+		return nil, false, err
+	}
+	h, err := store.JobRequestHash(in)
+	if err != nil {
+		return nil, false, err
+	}
+	if j.requestHash != h {
+		return nil, false, store.ErrJobConflict
+	}
+	return j, true, nil
+}
+
+// GetContext reads a job from the shared store when durability is enabled.
+func (r *Registry) GetContext(ctx context.Context, tenant, id string) (*Job, error) {
+	if r.queue == nil {
+		return r.Get(tenant, id)
+	}
+	c, err := r.queue.GetJob(ctx, tenant, id)
+	if errors.Is(err, store.ErrJobNotFound) {
+		return nil, ErrNotFound
+	}
+	if err != nil {
+		return nil, err
+	}
+	return jobFromRecord(c), nil
+}
+
+// ListContext returns a bounded tenant page from the authoritative queue.
+func (r *Registry) ListContext(ctx context.Context, tenant, status, kind, cursor string, limit int) ([]*Job, string, bool, error) {
+	if r.queue == nil {
+		j, n, m := r.List(tenant, status, kind, cursor, limit)
+		return j, n, m, nil
+	}
+	c, next, more, err := r.queue.ListJobs(ctx, tenant, status, kind, cursor, limit)
+	if err != nil {
+		return nil, "", false, err
+	}
+	out := make([]*Job, 0, len(c))
+	for _, v := range c {
+		out = append(out, jobFromRecord(v))
+	}
+	return out, next, more, nil
+}
+
+// RequestCancelContext persists a cancel request visible to any worker replica.
+func (r *Registry) RequestCancelContext(ctx context.Context, tenant, id string) (*Job, error) {
+	if r.queue == nil {
+		return r.RequestCancel(tenant, id)
+	}
+	c, err := r.queue.CancelJob(ctx, tenant, id)
+	if errors.Is(err, store.ErrJobNotFound) {
+		return nil, ErrNotFound
+	}
+	if err != nil {
+		return nil, err
+	}
+	r.mu.RLock()
+	local := r.byID[id]
+	r.mu.RUnlock()
+	if local != nil {
+		local.RequestCancel()
+	}
+	return jobFromRecord(c), nil
+}
+
+// RetryContext schedules an explicit retry of a failed shared operation.
+func (r *Registry) RetryContext(ctx context.Context, tenant, id string) (*Job, error) {
+	if r.queue == nil {
+		return nil, store.ErrJobConflict
+	}
+	c, err := r.queue.RetryJob(ctx, tenant, id)
+	if err != nil {
+		return nil, err
+	}
+	return jobFromRecord(c), nil
+}

internal/jobs/worker.go

@@ -2,6 +2,7 @@ package jobs
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"net/http"
 	"net/netip"
@@ -69,10 +70,15 @@ type Worker struct {
 	Store          store.Backend
 	PgPool         *pgxpool.Pool
 	HTTPClient     *http.Client // optional; CDN refresh uses this (default 45s timeout).
+	NodeHTTPClient *http.Client // optional; agent dispatch only, never shared with CDN transport.
 	// Registry is set after BootstrapWorkers creates the job queue; used to chain deploy_apply after refresh/rollback.
 	Registry *Registry
-	// refreshGate serializes deploy_apply gating after module_refresh per tenant (see finishModuleRefreshSuccess).
+	// refreshGate decides which module refresh renders for a tenant.
-	refreshGate sync.Map // map[string]*sync.Mutex
+	// The mutex covers only that decision; the render itself runs outside the lock.
+	refreshGate sync.Map // map[string]*tenantRefreshGate
+	// dispatchGate serializes publish and deployment state for one speaker in this process.
+	// Agent wake runs outside the mutex.
+	dispatchGate sync.Map // map[string]*sync.Mutex
 }
 
 type revisionLogEntry struct {
@@ -116,6 +122,23 @@ func (w *Worker) Process(j *Job) {
 		j.MarkCancelled()
 		return
 	}
+	// A committed child is the durable checkpoint for a parent that crashed
+	// before its own terminal write. Reuse it before rendering another revision.
+	if w.Registry != nil && (isRefreshKind(j.Kind) || j.Kind == KindRevisionRollback || j.Kind == KindPeerReconcile) {
+		child, err := w.Registry.FindIdempotentContext(context.Background(), j.TenantID, "deploy_from:"+j.ID)
+		if err == nil {
+			j.mergeMeta(map[string]any{"revision_id": child.metaString("revision_id"), "deploy_apply_job_id": child.ID, "recovered_child": true})
+			if isRefreshKind(j.Kind) {
+				w.Registry.finalizeRefreshCoalesce(j.TenantID)
+			}
+			j.Succeed()
+			return
+		}
+		if !errors.Is(err, store.ErrJobNotFound) {
+			j.Fail(err.Error())
+			return
+		}
+	}
 
 	switch j.Kind {
 	case KindModuleRefresh:
@@ -185,17 +208,17 @@ func (w *Worker) runPeerReconcile(j *Job) {
 	const peerJobTitle = "Обновление BGP пиров"
 	j.mergeMeta(map[string]any{"job_title": peerJobTitle})
 
-	var revID string
 	latest, _, _ := w.Store.ListRevisions(j.TenantID, "", "", 1)
 	triggerModuleID, err := w.peerTriggerModuleID(j.TenantID, latest)
 	if err != nil {
 		j.Fail(err.Error())
 		return
 	}
-	if len(latest) == 0 {
+	// Peer edits bump render generation. Render from module snapshots so the
+	// new revision includes current prefixes without paging the previous one.
 	ctx, cancel := j.workContext()
 	defer cancel()
-		rid, err := pipeline.RenderTenantRevision(ctx, w.Store, w.httpClient(), j.TenantID, triggerModuleID)
+	revID, err := pipeline.RenderTenantRevision(ctx, w.Store, w.httpClient(), j.TenantID, triggerModuleID)
 	if err != nil {
 		if ctx.Err() != nil {
 			j.MarkCancelled()
@@ -204,32 +227,6 @@ func (w *Worker) runPeerReconcile(j *Job) {
 		j.Fail(err.Error())
 		return
 	}
-		revID = rid
-	} else {
-		baseRevID := latest[0].ID
-		rows := make([]store.PrefixRow, 0, 1024)
-		cursor := ""
-		for {
-			page, next, more := w.Store.ListRevisionPrefixes(j.TenantID, baseRevID, cursor, 2000)
-			rows = append(rows, page...)
-			if !more || strings.TrimSpace(next) == "" {
-				break
-			}
-			cursor = next
-		}
-		ctx, cancel := j.workContext()
-		defer cancel()
-		rid, err := pipeline.RenderTenantRevisionFromPrefixes(ctx, w.Store, w.httpClient(), j.TenantID, triggerModuleID, rows)
-		if err != nil {
-			if ctx.Err() != nil {
-				j.MarkCancelled()
-				return
-			}
-			j.Fail(err.Error())
-			return
-		}
-		revID = rid
-	}
 	j.mergeMeta(map[string]any{"revision_id": revID})
 	if entries, total, err := w.buildRevisionLogEntries(j.TenantID, revID); err == nil {
 		j.mergeMeta(map[string]any{
@@ -240,7 +237,10 @@ func (w *Worker) runPeerReconcile(j *Job) {
 	} else {
 		j.mergeMeta(map[string]any{"log_build_error": err.Error()})
 	}
-	w.enqueueDeployAllSpeakers(j, j.TenantID, revID)
+	if err := w.enqueueDeployAllSpeakers(j, j.TenantID, revID); err != nil {
+		j.Fail(err.Error())
+		return
+	}
 	j.Succeed()
 }
 
@@ -269,8 +269,23 @@ func (w *Worker) peerTriggerModuleID(tenantID string, latest []*store.Revision)
 	return "", fmt.Errorf("missing module_id for peer reconcile")
 }
 
+type tenantRefreshGate struct {
+	mu        sync.Mutex
+	rendering bool
+	dirty     bool
+}
+
+func (w *Worker) tenantRefreshGate(tenantID string) *tenantRefreshGate {
+	v, _ := w.refreshGate.LoadOrStore(tenantID, &tenantRefreshGate{})
+	return v.(*tenantRefreshGate)
+}
+
 func (w *Worker) tenantRefreshMu(tenantID string) *sync.Mutex {
-	v, _ := w.refreshGate.LoadOrStore(tenantID, &sync.Mutex{})
+	return &w.tenantRefreshGate(tenantID).mu
+}
+
+func (w *Worker) speakerDispatchMu(speakerID string) *sync.Mutex {
+	v, _ := w.dispatchGate.LoadOrStore(speakerID, &sync.Mutex{})
 	return v.(*sync.Mutex)
 }
 
@@ -327,7 +342,8 @@ func (w *Worker) runTenantRefresh(j *Job) {
 
 	moduleIDs := moduleIDsFromJobMeta(j.metaCopy())
 	if len(moduleIDs) == 0 {
-		j.Fail("missing module_ids in job meta")
+		j.mergeMeta(map[string]any{"modules_refreshed": 0})
+		j.Succeed()
 		return
 	}
 	ctx, cancel := j.workContext()
@@ -382,20 +398,19 @@ func (w *Worker) finishModuleRefreshSuccess(j *Job, triggerModuleID string) {
 		j.Succeed()
 		return
 	}
-	mu := w.tenantRefreshMu(j.TenantID)
+	gate := w.tenantRefreshGate(j.TenantID)
-	mu.Lock()
+	gate.mu.Lock()
-	defer mu.Unlock()
+	// The counter is incremented in Enqueue and decremented here, so the "last one"
-	ctx, cancel := j.workContext()
+	// decision does not depend on job status polling. Render runs after the unlock.
-	defer cancel()
-	// Determine whether this is the last outstanding refresh for the tenant. The counter is
-	// incremented in Enqueue (under r.mu) and decremented here, so the "last one" decision is
-	// deterministic regardless of ingest timing — unlike the previous status-polling approach
-	// (CountOtherActiveRefresh) which could race under -race.
 	isLastRefresh := true
 	if w.Registry != nil {
 		isLastRefresh = w.Registry.finalizeRefreshCoalesce(j.TenantID)
 	}
-	if !isLastRefresh {
+	if !isLastRefresh || gate.rendering {
+		if gate.rendering {
+			gate.dirty = true
+		}
+		gate.mu.Unlock()
 		j.mergeMeta(map[string]any{
 			"deploy_apply_deferred":     true,
 			"deploy_apply_defer_reason": "parallel_module_refresh",
@@ -403,16 +418,33 @@ func (w *Worker) finishModuleRefreshSuccess(j *Job, triggerModuleID string) {
 		j.Succeed()
 		return
 	}
+	gate.rendering = true
+	gate.mu.Unlock()
+	defer func() {
+		gate.mu.Lock()
+		gate.rendering = false
+		gate.mu.Unlock()
+	}()
 
-	rev, err := pipeline.RenderTenantRevision(ctx, w.Store, w.httpClient(), j.TenantID, triggerModuleID)
+	ctx, cancel := j.workContext()
+	defer cancel()
+	rev, err := w.renderTenantUntilQuiet(ctx, j, triggerModuleID)
 	if err != nil {
-		if ctx.Err() != nil {
+		if errors.Is(err, errRenderSuperseded) {
+			j.Fail(err.Error())
+			return
+		}
+		if ctx.Err() != nil || j.IsCancelRequested() {
 			j.MarkCancelled()
 			return
 		}
 		j.Fail(err.Error())
 		return
 	}
+	if ctx.Err() != nil || j.IsCancelRequested() {
+		j.MarkCancelled()
+		return
+	}
 	j.mergeMeta(map[string]any{"revision_id": rev})
 	if entries, total, err := w.buildRevisionLogEntries(j.TenantID, rev); err == nil {
 		j.mergeMeta(map[string]any{
@@ -423,50 +455,240 @@ func (w *Worker) finishModuleRefreshSuccess(j *Job, triggerModuleID string) {
 	} else {
 		j.mergeMeta(map[string]any{"log_build_error": err.Error()})
 	}
-	w.enqueueDeployAllSpeakers(j, j.TenantID, rev)
+	if err := w.enqueueDeployAllSpeakers(j, j.TenantID, rev); err != nil {
+		j.Fail(err.Error())
+		return
+	}
 	j.Succeed()
 }
 
+const (
+	// renderQuietPassLimit stops a render loop that never observes a quiet input set.
+	// Hitting the cap fails the job; the last revision must not be published.
+	renderQuietPassLimit = 32
+	// speakerWakeLimit is the maximum number of concurrent agent wakes in one deploy.
+	speakerWakeLimit = 4
+)
+
+// errRenderSuperseded marks a render that must not be deployed. A newer snapshot
+// landed, the job was cancelled while inputs were still changing, or the quiet
+// pass cap was exhausted.
+var errRenderSuperseded = errors.New("newer snapshot still pending")
+
+// renderTenantUntilQuiet builds a revision outside the tenant gate.
+// It keeps rendering while another ingest marks the gate dirty or the snapshot
+// generation moves, and returns a revision only after a fresh check shows the
+// inputs are still the ones just rendered. Cancellation or the safety cap while
+// inputs are still changing fails the render instead of publishing a stale revision.
+func (w *Worker) renderTenantUntilQuiet(ctx context.Context, j *Job, triggerModuleID string) (string, error) {
+	gate := w.tenantRefreshGate(j.TenantID)
+	var rev string
+	for pass := 0; pass < renderQuietPassLimit; pass++ {
+		if abort := renderJobAborted(ctx, j); abort != nil {
+			if takeRefreshDirty(gate) {
+				return "", errRenderWhileDirty(abort)
+			}
+			return "", abort
+		}
+		before, err := w.captureRenderInputs(ctx, j.TenantID)
+		if err != nil {
+			if abort := renderJobAborted(ctx, j); abort != nil && takeRefreshDirty(gate) {
+				return "", errRenderWhileDirty(abort)
+			}
+			return "", err
+		}
+		rev, err = pipeline.RenderTenantRevision(ctx, w.Store, w.httpClient(), j.TenantID, triggerModuleID)
+		if err != nil {
+			dirty := takeRefreshDirty(gate)
+			if abort := renderJobAborted(ctx, j); abort != nil && dirty {
+				return "", errRenderWhileDirty(abort)
+			}
+			if abort := renderJobAborted(ctx, j); abort != nil {
+				return "", abort
+			}
+			if dirty {
+				continue
+			}
+			return "", err
+		}
+		after, err := w.captureRenderInputs(ctx, j.TenantID)
+		if err != nil {
+			if abort := renderJobAborted(ctx, j); abort != nil && takeRefreshDirty(gate) {
+				return "", errRenderWhileDirty(abort)
+			}
+			return "", err
+		}
+		dirty := takeRefreshDirty(gate)
+		if dirty || renderInputsChanged(before, after) {
+			if abort := renderJobAborted(ctx, j); abort != nil {
+				return "", errRenderWhileDirty(abort)
+			}
+			continue
+		}
+		confirm, err := w.captureRenderInputs(ctx, j.TenantID)
+		if err != nil {
+			if abort := renderJobAborted(ctx, j); abort != nil && takeRefreshDirty(gate) {
+				return "", errRenderWhileDirty(abort)
+			}
+			return "", err
+		}
+		if takeRefreshDirty(gate) || renderInputsChanged(after, confirm) {
+			if abort := renderJobAborted(ctx, j); abort != nil {
+				return "", errRenderWhileDirty(abort)
+			}
+			continue
+		}
+		if abort := renderJobAborted(ctx, j); abort != nil {
+			return "", abort
+		}
+		return rev, nil
+	}
+	if abort := renderJobAborted(ctx, j); abort != nil {
+		return "", errRenderWhileDirty(abort)
+	}
+	return "", errRenderWhileDirty(nil)
+}
+
+func errRenderWhileDirty(cause error) error {
+	if cause == nil {
+		return fmt.Errorf("jobs: tenant render did not quiesce after %d passes: %w", renderQuietPassLimit, errRenderSuperseded)
+	}
+	return fmt.Errorf("jobs: tenant render cancelled while a newer snapshot was still pending: %w", errRenderSuperseded)
+}
+
+func renderJobAborted(ctx context.Context, j *Job) error {
+	if err := ctx.Err(); err != nil {
+		return err
+	}
+	if j != nil && j.IsCancelRequested() {
+		return context.Canceled
+	}
+	return nil
+}
+
+func takeRefreshDirty(gate *tenantRefreshGate) bool {
+	if gate == nil {
+		return false
+	}
+	gate.mu.Lock()
+	defer gate.mu.Unlock()
+	dirty := gate.dirty
+	gate.dirty = false
+	return dirty
+}
+
+type renderInputView struct {
+	generation int64
+	snapshots  []store.ModuleSnapshotUse
+}
+
+func (w *Worker) captureRenderInputs(ctx context.Context, tenantID string) (renderInputView, error) {
+	if w == nil || w.Store == nil {
+		return renderInputView{}, fmt.Errorf("jobs: render store unavailable")
+	}
+	gen, err := w.Store.GetRenderGeneration(ctx, tenantID)
+	if err != nil {
+		if !errors.Is(err, store.ErrNotFound) {
+			return renderInputView{}, fmt.Errorf("jobs: read render generation: %w", err)
+		}
+		gen = 0
+	}
+	headers, err := w.Store.ListModuleSnapshotHeaders(ctx, tenantID)
+	if err != nil {
+		return renderInputView{}, fmt.Errorf("jobs: read snapshot headers: %w", err)
+	}
+	return renderInputView{generation: gen, snapshots: headers}, nil
+}
+
+func renderInputsChanged(before, after renderInputView) bool {
+	if before.generation != after.generation {
+		return true
+	}
+	return !sameSnapshotHeaders(before.snapshots, after.snapshots)
+}
+
+func sameSnapshotHeaders(a, b []store.ModuleSnapshotUse) bool {
+	if len(a) != len(b) {
+		return false
+	}
+	aa := append([]store.ModuleSnapshotUse(nil), a...)
+	bb := append([]store.ModuleSnapshotUse(nil), b...)
+	sort.Slice(aa, func(i, j int) bool { return aa[i].ModuleID < aa[j].ModuleID })
+	sort.Slice(bb, func(i, j int) bool { return bb[i].ModuleID < bb[j].ModuleID })
+	for i := range aa {
+		if aa[i].ModuleID != bb[i].ModuleID || aa[i].SnapshotVersion != bb[i].SnapshotVersion || aa[i].InputHash != bb[i].InputHash {
+			return false
+		}
+	}
+	return true
+}
+
 // enqueueDeployAllSpeakers queues the same work as POST /v1/apply (all speakers, no speaker_id).
-func (w *Worker) enqueueDeployAllSpeakers(j *Job, tenantID, revID string) {
+func (w *Worker) enqueueDeployAllSpeakers(j *Job, tenantID, revID string) error {
 	if w == nil || w.Registry == nil {
-		return
+		return fmt.Errorf("jobs: deploy queue unavailable")
 	}
 	revID = strings.TrimSpace(revID)
 	if revID == "" {
-		return
+		return fmt.Errorf("jobs: missing revision for deploy")
 	}
 	meta := map[string]any{"revision_id": revID}
 	if tr := strings.TrimSpace(j.metaString("trigger")); tr != "" {
 		meta["trigger"] = tr
 	}
-	applyJob, _, err := w.Registry.Enqueue(tenantID, KindDeployApply, nil, nil, meta)
+	key := "deploy_from:" + j.ID
+	applyJob, _, err := w.Registry.Enqueue(tenantID, KindDeployApply, &key, nil, meta)
 	if err != nil {
 		j.mergeMeta(map[string]any{"deploy_apply_enqueue_error": err.Error()})
-		return
+		return fmt.Errorf("jobs: enqueue deploy: %w", err)
 	}
 	if applyJob != nil {
 		j.mergeMeta(map[string]any{"deploy_apply_job_id": applyJob.ID})
 	}
+	return nil
 }
 
 func (w *Worker) runDeployApply(j *Job) {
 	revID := strings.TrimSpace(j.metaString("revision_id"))
 	spk := strings.TrimSpace(j.metaString("speaker_id"))
-	hasSpeaker := spk != ""
 	if revID == "" {
 		j.Fail("missing revision_id in job meta")
 		return
 	}
+	if j.OwnerID == "" {
+		if _, developmentMemory := w.Store.(*store.Memory); !developmentMemory {
+			j.Fail("deployment requires a current durable job lease")
+			return
+		}
+	}
+
 	ctx, cancel := j.workContext()
 	defer cancel()
-	activeDir := strings.TrimSpace(os.Getenv("EVOBGP_BIRD_ACTIVE_DIR"))
+	cancelled := func() bool { return ctx.Err() != nil || j.IsCancelRequested() }
-	if activeDir != "" {
+	var speakers []*store.Speaker
-		revObj, err := w.Store.GetRevision(j.TenantID, revID)
+	if spk != "" {
+		sp, err := w.Store.GetSpeaker(j.TenantID, spk)
 		if err != nil {
 			j.Fail(err.Error())
 			return
 		}
+		speakers = []*store.Speaker{sp}
+	} else {
+		speakers = w.Store.ListSpeakersForTenant(j.TenantID)
+	}
+	activeDir := strings.TrimSpace(os.Getenv("EVOBGP_BIRD_ACTIVE_DIR"))
+	localAttempted := false
+	var localErr error
+	applyLocal := func() error {
+		if localAttempted {
+			return localErr
+		}
+		localAttempted = true
+		revObj, err := w.Store.GetRevision(j.TenantID, revID)
+		if err != nil {
+			localErr = err
+			return err
+		}
 		staging := strings.TrimSpace(os.Getenv("EVOBGP_BIRD_STAGING_DIR"))
 		if staging == "" {
 			staging = os.TempDir() + "/evobgp-bird-staging"
@@ -480,39 +702,292 @@ func (w *Worker) runDeployApply(j *Job) {
 		}
 		ctl := &birdfmt.BirdCtl{Bird: cfg.BirdBin, Birdc: cfg.BirdcBin, Socket: cfg.Socket}
 		if err := birddeploy.ApplyRevision(ctx, ctl, revObj, cfg); err != nil {
-			if ctx.Err() != nil {
+			localErr = err
+			return err
+		}
+		return nil
+	}
+	tally := &deployTally{applied: make([]string, 0, len(speakers))}
+	slots := make(chan struct{}, speakerWakeLimit)
+	var wakeGroup sync.WaitGroup
+	acquireWake := func() bool {
+		select {
+		case slots <- struct{}{}:
+		case <-ctx.Done():
+			return false
+		}
+		if cancelled() {
+			<-slots
+			return false
+		}
+		return true
+	}
+	for _, sp := range speakers {
+		if cancelled() {
+			break
+		}
+		if sp == nil {
+			continue
+		}
+		held := false
+		if speakerRemoteWake(sp) {
+			if !acquireWake() {
+				break
+			}
+			held = true
+		}
+		staged := w.stageSpeakerDeploy(ctx, j, sp, revID, activeDir, cancelled, applyLocal)
+		if staged.wake && !staged.stop {
+			wakeGroup.Add(1)
+			go func(sp *store.Speaker, deployment store.DeploymentSpeaker) {
+				defer wakeGroup.Done()
+				defer func() { <-slots }()
+				wakeCtx, cancelWake := context.WithTimeout(ctx, 35*time.Second)
+				res := nodedispatch.WakeSpeaker(wakeCtx, sp, nodedispatch.Options{RevisionID: revID, HTTPClient: w.NodeHTTPClient, MaxRetries: 1})
+				cancelWake()
+				out, include, failedN, unconfirmedN, appliedID := w.finishSpeakerWake(ctx, j, sp, revID, deployment, res, cancelled)
+				tally.add(out, include, 0, failedN, unconfirmedN, appliedID)
+			}(sp, staged.deployment)
+		} else if held {
+			<-slots
+		}
+		tally.add(staged.result, staged.hasResult, staged.published, staged.failed, staged.unconfirmed, staged.appliedID)
+		if staged.stop {
+			break
+		}
+	}
+	wakeGroup.Wait()
+	results, applied, published, failed, unconfirmed := tally.snapshot()
+	j.mergeMeta(map[string]any{
+		"apply_summary": map[string]any{
+			"revision_id":       revID,
+			"speakers_count":    len(speakers),
+			"published_count":   published,
+			"applied_count":     len(applied),
+			"unconfirmed_count": unconfirmed,
+			"failed_count":      failed,
+			"speaker_ids":       applied,
+			"message":           fmt.Sprintf("Ревизия %s: подтверждено %d, без подтверждения %d, ошибок %d", shortID(revID), len(applied), unconfirmed, failed),
+		},
+		"node_dispatch": map[string]any{"revision_id": revID, "results": results},
+	})
+	if cancelled() {
 		j.MarkCancelled()
 		return
 	}
-			j.Fail(err.Error())
+	mergeBirdPostApplyMeta(j)
+	if cancelled() {
+		j.MarkCancelled()
 		return
 	}
+	if failed > 0 {
+		j.Fail(fmt.Sprintf("deploy delivery failed for %d of %d speakers", failed, len(speakers)))
+		return
 	}
-	applied := make([]string, 0, 8)
+	j.Succeed()
-	var dispatchResults []nodedispatch.Result
-	applyOne := func(speakerID string) error {
-		if err := w.Store.SetLastAppliedRevision(j.TenantID, speakerID, revID); err != nil {
-			return err
 }
-		// Replica / node pulls use LatestPublishedRevision; keep pointer in sync with successful deploy.
+
-		if err := w.Store.PublishRevisionForSpeaker(speakerID, revID); err != nil {
+type speakerStage struct {
-			return err
+	stop        bool
+	wake        bool
+	deployment  store.DeploymentSpeaker
+	result      nodedispatch.Result
+	hasResult   bool
+	published   int
+	failed      int
+	unconfirmed int
+	appliedID   string
 }
-		applied = append(applied, speakerID)
+
+type deployTally struct {
+	mu          sync.Mutex
+	results     []nodedispatch.Result
+	applied     []string
+	published   int
+	failed      int
+	unconfirmed int
+}
+
+func (t *deployTally) add(res nodedispatch.Result, include bool, published, failed, unconfirmed int, appliedID string) {
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	if include {
+		t.results = append(t.results, res)
+	}
+	t.published += published
+	t.failed += failed
+	t.unconfirmed += unconfirmed
+	if appliedID != "" {
+		t.applied = append(t.applied, appliedID)
+	}
+}
+
+func (t *deployTally) snapshot() (results []nodedispatch.Result, applied []string, published, failed, unconfirmed int) {
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	results = append([]nodedispatch.Result(nil), t.results...)
+	applied = append([]string(nil), t.applied...)
+	if applied == nil {
+		applied = []string{}
+	}
+	sort.Slice(results, func(i, j int) bool { return results[i].SpeakerID < results[j].SpeakerID })
+	sort.Strings(applied)
+	return results, applied, t.published, t.failed, t.unconfirmed
+}
+
+func speakerRemoteWake(sp *store.Speaker) bool {
+	if sp == nil || !nodedispatch.Enabled() {
+		return false
+	}
+	return store.SpeakerNeedsRemoteDispatch(sp.Role, store.ParseSpeakerMeta(sp.MetaJSON))
+}
+
+func (w *Worker) bindDeploymentWriters(ctx context.Context, j *Job, sp *store.Speaker, revID string, deployment store.DeploymentSpeaker) (func(string, string, string) error, func(string) error) {
+	fence := deploymentFenceForJob(j)
+	record := func(state, confirmation, message string) error {
+		if j.OwnerID == "" {
 			return nil
 		}
-	dispatchSpeaker := func(sp *store.Speaker) {
+		return w.Store.RecordDeploymentSpeaker(ctx, fence, deployment, state, confirmation, message)
-		if !nodedispatch.Enabled() || sp == nil {
+	}
-			return
+	markApplied := func(confirmation string) error {
+		if j.OwnerID == "" {
+			return w.Store.SetLastAppliedRevision(j.TenantID, sp.ID, revID)
+		}
+		return record("applied", confirmation, "")
+	}
+	return record, markApplied
+}
+
+// stageSpeakerDeploy publishes one speaker and records terminal states.
+// Remote wakes are left to the caller so the speaker lock is not held across the network.
+func (w *Worker) stageSpeakerDeploy(ctx context.Context, j *Job, sp *store.Speaker, revID, activeDir string, cancelled func() bool, applyLocal func() error) speakerStage {
+	var staged speakerStage
+	mu := w.speakerDispatchMu(sp.ID)
+	mu.Lock()
+	defer mu.Unlock()
+	if cancelled() {
+		staged.stop = true
+		return staged
+	}
+	var deployment store.DeploymentSpeaker
+	var publishErr error
+	if j.OwnerID != "" {
+		deployment, publishErr = beginDeployment(ctx, w.Store, deploymentFenceForJob(j), sp.ID, revID)
+	} else {
+		publishErr = w.Store.PublishRevisionForSpeaker(sp.ID, revID)
+	}
+	staged.deployment = deployment
+	record, markApplied := w.bindDeploymentWriters(ctx, j, sp, revID, deployment)
+	if publishErr != nil {
+		staged.failed = 1
+		staged.hasResult = true
+		staged.result = nodedispatch.Result{SpeakerID: sp.ID, Status: "error", Error: publishErr.Error()}
+		return staged
+	}
+	staged.published = 1
+	if deployment.State == "applied" {
+		staged.appliedID = sp.ID
+		staged.hasResult = true
+		staged.result = nodedispatch.Result{SpeakerID: sp.ID, Status: "ok", Confirmation: deployment.Confirmation, AppliedRevisionID: revID}
+		return staged
+	}
+	if cancelled() {
+		staged.stop = true
+		return staged
+	}
+	if !speakerRemoteWake(sp) {
+		if activeDir != "" && strings.EqualFold(sp.Role, "master") {
+			err := applyLocal()
+			if cancelled() {
+				staged.stop = true
+				return staged
+			}
+			if err != nil {
+				_ = record("failed", "unconfirmed", err.Error())
+				staged.failed = 1
+				staged.hasResult = true
+				staged.result = nodedispatch.Result{SpeakerID: sp.ID, Status: "error", Error: err.Error()}
+				return staged
+			}
+			if err := markApplied("local_applied"); err != nil {
+				staged.failed = 1
+				staged.hasResult = true
+				staged.result = nodedispatch.Result{SpeakerID: sp.ID, Status: "error", Error: err.Error()}
+				return staged
+			}
+			staged.appliedID = sp.ID
+			staged.hasResult = true
+			staged.result = nodedispatch.Result{SpeakerID: sp.ID, Status: "ok", Confirmation: "local_applied", AppliedRevisionID: revID}
+			return staged
+		}
+		if err := record("unconfirmed", "unconfirmed", "dispatch disabled or speaker has no remote agent"); err != nil {
+			staged.failed = 1
+		}
+		staged.unconfirmed = 1
+		staged.hasResult = true
+		staged.result = nodedispatch.Result{SpeakerID: sp.ID, Status: "skipped", Confirmation: "unconfirmed"}
+		return staged
+	}
+	if err := record("dispatched", "unconfirmed", "awaiting agent acknowledgement"); err != nil {
+		staged.failed = 1
+		staged.hasResult = true
+		staged.result = nodedispatch.Result{SpeakerID: sp.ID, Status: "error", Error: err.Error()}
+		return staged
+	}
+	if deployment.DispatchAttempts >= 3 {
+		_ = record("unknown", "unconfirmed", "delivery attempt budget exhausted")
+		staged.failed = 1
+		staged.hasResult = true
+		staged.result = nodedispatch.Result{SpeakerID: sp.ID, Status: "error", Confirmation: "unconfirmed", Error: "delivery attempt budget exhausted"}
+		return staged
+	}
+	staged.wake = true
+	return staged
+}
+
+// finishSpeakerWake records the agent acknowledgement. The speaker lock covers the
+// state update and is not held while WakeSpeaker is in flight.
+func (w *Worker) finishSpeakerWake(ctx context.Context, j *Job, sp *store.Speaker, revID string, deployment store.DeploymentSpeaker, res nodedispatch.Result, cancelled func() bool) (nodedispatch.Result, bool, int, int, string) {
+	mu := w.speakerDispatchMu(sp.ID)
+	mu.Lock()
+	defer mu.Unlock()
+	if cancelled != nil && cancelled() {
+		return nodedispatch.Result{SpeakerID: sp.ID, Endpoint: res.Endpoint, Status: "cancelled", Confirmation: "unconfirmed"}, true, 0, 0, ""
+	}
+	record, markApplied := w.bindDeploymentWriters(ctx, j, sp, revID, deployment)
+	appliedID := ""
+	if res.Status == "ok" && res.AppliedRevisionID != "" {
+		if res.AppliedRevisionID != revID {
+			res.Status = "error"
+			res.Confirmation = "unconfirmed"
+			res.Error = "agent reported a different applied revision"
+		} else if latest, _, err := w.Store.LatestPublishedRevision(sp.ID); err != nil {
+			res.Status = "error"
+			res.Confirmation = "unconfirmed"
+			res.Error = err.Error()
+		} else if latest != revID {
+			res.Status = "error"
+			res.Confirmation = "unconfirmed"
+			res.Error = "stale agent acknowledgement: desired revision changed"
+		} else if err := markApplied("agent_reported"); err != nil {
+			res.Status = "error"
+			res.Confirmation = "unconfirmed"
+			res.Error = err.Error()
+		} else {
+			appliedID = sp.ID
+		}
+	}
+	if res.Status != "ok" || res.Confirmation != "agent_reported" {
+		state := "unconfirmed"
+		if res.Status != "ok" {
+			state = "unknown"
+		}
+		if err := record(state, "unconfirmed", res.Error); err != nil {
+			res.Status = "error"
+			res.Error = err.Error()
 		}
-		meta := store.ParseSpeakerMeta(sp.MetaJSON)
-		if !store.SpeakerNeedsRemoteDispatch(sp.Role, meta) {
-			return
 	}
-		ctx2, cancel := context.WithTimeout(ctx, 35*time.Second)
-		defer cancel()
-		res := nodedispatch.WakeSpeaker(ctx2, sp, nodedispatch.Options{RevisionID: revID})
-		dispatchResults = append(dispatchResults, res)
 	patch := store.SpeakerMeta{
 		LastDispatchAt:     time.Now().UTC().Format(time.RFC3339Nano),
 		LastDispatchStatus: res.Status,
@@ -520,57 +995,27 @@ func (w *Worker) runDeployApply(j *Job) {
 	if res.Error != "" {
 		patch.LastDispatchError = res.Error
 		patch.SyncStatus = "error"
-		} else if res.Status == "ok" {
+	} else if res.Status == "ok" && res.Confirmation == "agent_reported" {
 		patch.LastDispatchError = ""
 		patch.SyncStatus = "synced"
+	} else if res.Status == "ok" {
+		patch.SyncStatus = "unconfirmed"
 	}
 	merged := store.MergeSpeakerMetaJSON(sp.MetaJSON, patch)
-		_, _ = w.Store.UpdateSpeaker(j.TenantID, sp.ID, &store.SpeakerPatch{MetaJSON: &merged})
+	if j.OwnerID == "" {
-	}
+		if _, err := w.Store.UpdateSpeaker(j.TenantID, sp.ID, &store.SpeakerPatch{MetaJSON: &merged}); err != nil {
-	if hasSpeaker && spk != "" {
+			res.Status = "error"
-		if err := applyOne(spk); err != nil {
+			res.Error = "dispatch metadata update: " + err.Error()
-			j.Fail(err.Error())
-			return
-		}
-		if sp, err := w.Store.GetSpeaker(j.TenantID, spk); err == nil {
-			dispatchSpeaker(sp)
-		}
-		if len(dispatchResults) > 0 {
-			j.mergeMeta(map[string]any{"node_dispatch": map[string]any{
-				"revision_id": revID,
-				"results":     dispatchResults,
-			}})
-		}
-		mergeBirdPostApplyMeta(j)
-		j.Succeed()
-		return
-	}
-	speakers := w.Store.ListSpeakersForTenant(j.TenantID)
-	for _, sp := range speakers {
-		if err := applyOne(sp.ID); err != nil {
-			j.Fail(err.Error())
-			return
 		}
 	}
-	for _, sp := range speakers {
+	failed := 0
-		dispatchSpeaker(sp)
+	unconfirmed := 0
+	if res.Status == "ok" && res.Confirmation == "unconfirmed" {
+		unconfirmed = 1
+	} else if res.Status != "ok" {
+		failed = 1
 	}
-	j.mergeMeta(map[string]any{
+	return res, true, failed, unconfirmed, appliedID
-		"apply_summary": map[string]any{
-			"revision_id":    revID,
-			"speakers_count": len(applied),
-			"speaker_ids":    applied,
-			"message":        fmt.Sprintf("Ревизия %s применена на %d спикерах", shortID(revID), len(applied)),
-		},
-	})
-	if len(dispatchResults) > 0 {
-		j.mergeMeta(map[string]any{"node_dispatch": map[string]any{
-			"revision_id": revID,
-			"results":     dispatchResults,
-		}})
-	}
-	mergeBirdPostApplyMeta(j)
-	j.Succeed()
 }
 
 func (w *Worker) runRollback(j *Job) {
@@ -592,7 +1037,10 @@ func (w *Worker) runRollback(j *Job) {
 			"message":            fmt.Sprintf("Откат %s → %s", shortID(src), shortID(newID)),
 		},
 	})
-	w.enqueueDeployAllSpeakers(j, j.TenantID, newID)
+	if err := w.enqueueDeployAllSpeakers(j, j.TenantID, newID); err != nil {
+		j.Fail(err.Error())
+		return
+	}
 	j.Succeed()
 }
 

internal/jobs/worker_test.go

@@ -1,6 +1,9 @@
 package jobs
 
 import (
+	"context"
+	"fmt"
+	"strings"
 	"testing"
 	"time"
 
@@ -185,3 +188,187 @@ func TestPeerReconcile_RendersFromLatestRevisionAndQueuesDeploy(t *testing.T) {
 		t.Fatalf("expected deploy_apply_job_id in peer_reconcile meta, got %#v", meta)
 	}
 }
+
+type renderHookStore struct {
+	*store.Memory
+	onCreate func(prefixes []store.PrefixRow)
+}
+
+func (s *renderHookStore) CreateRenderRevision(ctx context.Context, revisionID, tenantID, moduleID string, parent *string, contentHash string, preview map[string]string, prefixes []store.PrefixRow) error {
+	if err := s.Memory.CreateRenderRevision(ctx, revisionID, tenantID, moduleID, parent, contentHash, preview, prefixes); err != nil {
+		return err
+	}
+	if s.onCreate != nil {
+		copied := append([]store.PrefixRow(nil), prefixes...)
+		s.onCreate(copied)
+	}
+	return nil
+}
+
+func newIPRenderFixture(t *testing.T) (tenant, module string, mem *store.Memory) {
+	t.Helper()
+	t.Setenv("EVOBGP_ASN_RESOLVE", "0")
+	t.Setenv("EVOBGP_BIRD_ACTIVE_DIR", "")
+	mem = store.NewMemory()
+	mem.SeedDemo()
+	var modIP string
+	tenant, _, modIP, _, _ = mem.DemoIDs()
+	module = modIP
+	for _, mod := range mem.ListModules(tenant) {
+		if mod == nil || mod.ID == module {
+			continue
+		}
+		disabled := false
+		if _, err := mem.UpdateModule(tenant, mod.ID, &store.ModulePatch{Enabled: &disabled}); err != nil {
+			t.Fatal(err)
+		}
+	}
+	return tenant, module, mem
+}
+
+func installModuleSnapshot(t *testing.T, mem *store.Memory, tenant, module, prefix string) {
+	t.Helper()
+	err := mem.SetModulePrefixSnapshot(context.Background(), tenant, module, "snap-"+prefix, []store.PrefixRow{{
+		Prefix: prefix,
+		Source: "ip_range",
+	}})
+	if err != nil {
+		t.Fatalf("snapshot %s: %v", prefix, err)
+	}
+}
+
+func revisionHasPrefix(t *testing.T, mem *store.Memory, tenant, revID, prefix string) bool {
+	t.Helper()
+	rows, _, _ := mem.ListRevisionPrefixes(tenant, revID, "", 100)
+	for _, row := range rows {
+		if row.Prefix == prefix {
+			return true
+		}
+	}
+	return false
+}
+
+func TestRenderTenantUntilQuietOverlappingIngests(t *testing.T) {
+	t.Run("latest snapshot", func(t *testing.T) {
+		tenant, module, mem := newIPRenderFixture(t)
+		prefixes := []string{"10.8.0.0/24", "10.9.0.0/24", "10.10.0.0/24", "10.11.0.0/24"}
+		installModuleSnapshot(t, mem, tenant, module, prefixes[0])
+		hook := &renderHookStore{Memory: mem}
+		var wk *Worker
+		step := 0
+		hook.onCreate = func([]store.PrefixRow) {
+			step++
+			if step >= len(prefixes) {
+				return
+			}
+			installModuleSnapshot(t, mem, tenant, module, prefixes[step])
+			gate := wk.tenantRefreshGate(tenant)
+			gate.mu.Lock()
+			gate.dirty = true
+			gate.mu.Unlock()
+		}
+		wk = &Worker{Store: hook, Registry: NewRegistry(nil)}
+		j := &Job{ID: "refresh-overlap", TenantID: tenant, Kind: KindModuleRefresh, Status: StatusRunning}
+		wk.finishModuleRefreshSuccess(j, module)
+		if j.Status != StatusSucceeded {
+			errText := ""
+			if j.Error != nil {
+				errText = *j.Error
+			}
+			t.Fatalf("status=%s err=%s", j.Status, errText)
+		}
+		if step != len(prefixes) {
+			t.Fatalf("render commits=%d, want %d so three overlapping ingests are included", step, len(prefixes))
+		}
+		rev := j.metaString("revision_id")
+		if !revisionHasPrefix(t, mem, tenant, rev, prefixes[len(prefixes)-1]) {
+			t.Fatalf("deployed revision %s missing latest prefix %s", rev, prefixes[len(prefixes)-1])
+		}
+		if revisionHasPrefix(t, mem, tenant, rev, prefixes[len(prefixes)-2]) {
+			t.Fatalf("deployed revision %s still has stale prefix %s", rev, prefixes[len(prefixes)-2])
+		}
+		deployID := j.metaString("deploy_apply_job_id")
+		deploy, err := wk.Registry.Get(tenant, deployID)
+		if err != nil {
+			t.Fatal(err)
+		}
+		if deploy.metaString("revision_id") != rev {
+			t.Fatalf("deploy revision=%s, want %s", deploy.metaString("revision_id"), rev)
+		}
+	})
+
+	t.Run("cancel does not deploy", func(t *testing.T) {
+		tenant, module, mem := newIPRenderFixture(t)
+		installModuleSnapshot(t, mem, tenant, module, "10.12.0.0/24")
+		hook := &renderHookStore{Memory: mem}
+		var wk *Worker
+		var j *Job
+		commits := 0
+		hook.onCreate = func([]store.PrefixRow) {
+			commits++
+			installModuleSnapshot(t, mem, tenant, module, "10.13.0.0/24")
+			gate := wk.tenantRefreshGate(tenant)
+			gate.mu.Lock()
+			gate.dirty = true
+			gate.mu.Unlock()
+			j.RequestCancel()
+		}
+		wk = &Worker{Store: hook, Registry: NewRegistry(nil)}
+		j = &Job{ID: "refresh-cancel", TenantID: tenant, Kind: KindModuleRefresh, Status: StatusRunning}
+		wk.finishModuleRefreshSuccess(j, module)
+		if j.Status != StatusFailed {
+			t.Fatalf("status=%s, want failed", j.Status)
+		}
+		if j.Error == nil || !strings.Contains(*j.Error, "newer snapshot") {
+			t.Fatalf("error=%v, want superseded render", j.Error)
+		}
+		if commits != 1 {
+			t.Fatalf("render commits=%d, want 1", commits)
+		}
+		if j.metaString("deploy_apply_job_id") != "" || j.metaString("revision_id") != "" {
+			t.Fatalf("cancelled dirty render published meta %#v", j.metaCopy())
+		}
+		deploys, _, _ := wk.Registry.List(tenant, "", KindDeployApply, "", 10)
+		if len(deploys) != 0 {
+			t.Fatalf("deploy jobs=%d, want 0", len(deploys))
+		}
+		revs, _, _ := mem.ListRevisions(tenant, "", "", 50)
+		for _, rev := range revs {
+			if rev != nil && revisionHasPrefix(t, mem, tenant, rev.ID, "10.13.0.0/24") {
+				t.Fatalf("revision %s contains the snapshot that landed during cancel", rev.ID)
+			}
+		}
+	})
+
+	t.Run("cap does not publish", func(t *testing.T) {
+		tenant, module, mem := newIPRenderFixture(t)
+		installModuleSnapshot(t, mem, tenant, module, "10.20.0.0/24")
+		hook := &renderHookStore{Memory: mem}
+		var wk *Worker
+		commits := 0
+		hook.onCreate = func([]store.PrefixRow) {
+			commits++
+			installModuleSnapshot(t, mem, tenant, module, fmt.Sprintf("10.21.%d.0/24", commits))
+			gate := wk.tenantRefreshGate(tenant)
+			gate.mu.Lock()
+			gate.dirty = true
+			gate.mu.Unlock()
+		}
+		wk = &Worker{Store: hook, Registry: NewRegistry(nil)}
+		j := &Job{ID: "refresh-cap", TenantID: tenant, Kind: KindModuleRefresh, Status: StatusRunning}
+		wk.finishModuleRefreshSuccess(j, module)
+		if j.Status != StatusFailed || j.Error == nil || !strings.Contains(*j.Error, "did not quiesce") {
+			errText := ""
+			if j.Error != nil {
+				errText = *j.Error
+			}
+			t.Fatalf("status=%s err=%s", j.Status, errText)
+		}
+		if commits != renderQuietPassLimit {
+			t.Fatalf("render commits=%d, want cap %d", commits, renderQuietPassLimit)
+		}
+		if j.metaString("deploy_apply_job_id") != "" {
+			t.Fatal("safety cap published a deploy")
+		}
+	})
+}

internal/lookup/lookup.go

@@ -107,14 +107,14 @@ func LookupWithResolver(
 	if addr, err := netip.ParseAddr(raw); err == nil {
 		out.QueryKind = KindIP
 		out.Normalized = addr.String()
-		if err := lookupIP(st, tenantID, addr, out, commByID, ""); err != nil {
+		if err := lookupIP(ctx, st, tenantID, addr, out, commByID, ""); err != nil {
 			return nil, err
 		}
 	} else if pfx, err := netip.ParsePrefix(raw); err == nil {
 		masked := pfx.Masked()
 		out.QueryKind = KindCIDR
 		out.Normalized = masked.String()
-		if err := lookupCIDR(st, tenantID, masked, out, commByID); err != nil {
+		if err := lookupCIDR(ctx, st, tenantID, masked, out, commByID); err != nil {
 			return nil, err
 		}
 	} else {
@@ -124,7 +124,7 @@ func LookupWithResolver(
 		}
 		out.QueryKind = KindDomain
 		out.Normalized = fqdn
-		if err := lookupDomain(st, tenantID, fqdn, out, commByID); err != nil {
+		if err := lookupDomain(ctx, st, tenantID, fqdn, out, commByID); err != nil {
 			return nil, err
 		}
 		if resolve == nil {
@@ -178,7 +178,7 @@ func lookupResolvedIPs(
 	out.ResolvedIPs = make([]string, 0, len(ips))
 	for _, ip := range ips {
 		out.ResolvedIPs = append(out.ResolvedIPs, ip.String())
-		if err := lookupIP(st, tenantID, ip, out, commByID, ip.String()); err != nil {
+		if err := lookupIP(ctx, st, tenantID, ip, out, commByID, ip.String()); err != nil {
 			return err
 		}
 	}
@@ -186,6 +186,7 @@ func lookupResolvedIPs(
 }
 
 func lookupIP(
+	ctx context.Context,
 	st store.Backend,
 	tenantID string,
 	addr netip.Addr,
@@ -227,7 +228,7 @@ func lookupIP(
 			}
 		}
 
-		snap, ok, err := st.GetModulePrefixSnapshot(tenantID, mod.ID)
+		snap, ok, err := st.GetModulePrefixSnapshot(ctx, tenantID, mod.ID)
 		if err != nil {
 			return err
 		}
@@ -269,6 +270,7 @@ func prefixCoversQuery(listPfx, query netip.Prefix) bool {
 }
 
 func lookupCIDR(
+	ctx context.Context,
 	st store.Backend,
 	tenantID string,
 	query netip.Prefix,
@@ -308,7 +310,7 @@ func lookupCIDR(
 			}
 		}
 
-		snap, ok, err := st.GetModulePrefixSnapshot(tenantID, mod.ID)
+		snap, ok, err := st.GetModulePrefixSnapshot(ctx, tenantID, mod.ID)
 		if err != nil {
 			return err
 		}
@@ -338,7 +340,7 @@ func lookupCIDR(
 	return nil
 }
 
-func lookupDomain(st store.Backend, tenantID, fqdn string, out *Result, commByID map[string]*store.Community) error {
+func lookupDomain(ctx context.Context, st store.Backend, tenantID, fqdn string, out *Result, commByID map[string]*store.Community) error {
 	matchedModuleIDs := make(map[string]*store.Module)
 
 	for _, mod := range st.ListModules(tenantID) {
@@ -372,7 +374,7 @@ func lookupDomain(st store.Backend, tenantID, fqdn string, out *Result, commByID
 	}
 
 	for mid, mod := range matchedModuleIDs {
-		snap, ok, err := st.GetModulePrefixSnapshot(tenantID, mid)
+		snap, ok, err := st.GetModulePrefixSnapshot(ctx, tenantID, mid)
 		if err != nil {
 			return err
 		}

internal/lookup/lookup_test.go

@@ -35,7 +35,7 @@ func TestLookupIPEntryAndSnapshot(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	if err := m.SetModulePrefixSnapshot(tenant, modIP, "hash1", []store.PrefixRow{
+	if err := m.SetModulePrefixSnapshot(context.Background(), tenant, modIP, "hash1", []store.PrefixRow{
 		{Prefix: "203.0.113.0/24", CommunityID: &cid, Source: "ip_range"},
 	}); err != nil {
 		t.Fatal(err)
@@ -129,7 +129,7 @@ func TestLookupDomainEntryAndSnapshot(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	if err := m.SetModulePrefixSnapshot(tenant, mod.ID, "hash-d", []store.PrefixRow{
+	if err := m.SetModulePrefixSnapshot(context.Background(), tenant, mod.ID, "hash-d", []store.PrefixRow{
 		{Prefix: "198.51.100.1/32", CommunityID: &cid, Source: "domain"},
 		{Prefix: "203.0.113.9/32", CommunityID: &cid, Source: "other"},
 	}); err != nil {
@@ -183,7 +183,7 @@ func TestLookupDomainResolvedIPAgainstRanges(t *testing.T) {
 	}); err != nil {
 		t.Fatal(err)
 	}
-	if err := m.SetModulePrefixSnapshot(tenant, modIP, "hash-r", []store.PrefixRow{
+	if err := m.SetModulePrefixSnapshot(context.Background(), tenant, modIP, "hash-r", []store.PrefixRow{
 		{Prefix: "203.0.113.0/24", CommunityID: &cid, Source: "ip_range"},
 	}); err != nil {
 		t.Fatal(err)
@@ -234,7 +234,7 @@ func TestLookupCIDREntryAndSnapshot(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	if err := m.SetModulePrefixSnapshot(tenant, modIP, "hash-cidr", []store.PrefixRow{
+	if err := m.SetModulePrefixSnapshot(context.Background(), tenant, modIP, "hash-cidr", []store.PrefixRow{
 		{Prefix: "203.0.113.0/24", CommunityID: &cid, Source: "ip_range"},
 	}); err != nil {
 		t.Fatal(err)

internal/nodecli/commands.go

@@ -61,13 +61,17 @@ func nodeHTTPClient() *http.Client {
 }
 
 func fetchLatestRevision(base, token, speaker string) (string, error) {
+	return fetchLatestRevisionContext(context.Background(), base, token, speaker)
+}
+
+func fetchLatestRevisionContext(parent context.Context, base, token, speaker string) (string, error) {
 	u := strings.TrimRight(base, "/") + "/v1/speakers/" + speaker + "/revisions/latest"
 	req, err := http.NewRequest(http.MethodGet, u, nil)
 	if err != nil {
 		return "", err
 	}
 	req.Header.Set("Authorization", "Bearer "+token)
-	ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)
+	ctx, cancel := context.WithTimeout(parent, 60*time.Second)
 	defer cancel()
 	resp, err := httpclient.DoWithRetry(ctx, nodeHTTPClient(), req, 3)
 	if err != nil {
@@ -75,13 +79,17 @@ func fetchLatestRevision(base, token, speaker string) (string, error) {
 	}
 	defer func() { _ = resp.Body.Close() }()
 	if resp.StatusCode != http.StatusOK {
-		b, _ := io.ReadAll(resp.Body)
+		b, _ := io.ReadAll(io.LimitReader(resp.Body, 4<<10))
 		return "", fmt.Errorf("latest revision: %s: %s", resp.Status, strings.TrimSpace(string(b)))
 	}
 	var out struct {
 		RevisionID string `json:"revision_id"`
 	}
-	if err := json.NewDecoder(resp.Body).Decode(&out); err != nil {
+	body, err := readHTTPBytes(resp, 64<<10)
+	if err != nil {
+		return "", err
+	}
+	if err := json.Unmarshal(body, &out); err != nil {
 		return "", err
 	}
 	if out.RevisionID == "" {
@@ -91,13 +99,21 @@ func fetchLatestRevision(base, token, speaker string) (string, error) {
 }
 
 func fetchBundle(base, token, speaker, revision string) ([]byte, error) {
+	return fetchBundleContext(context.Background(), base, token, speaker, revision, bundle.Limits{})
+}
+
+func fetchBundleContext(parent context.Context, base, token, speaker, revision string, limits bundle.Limits) ([]byte, error) {
+	l, err := limits.Normalized()
+	if err != nil {
+		return nil, err
+	}
 	u := strings.TrimRight(base, "/") + "/v1/speakers/" + speaker + "/bundle/" + revision
 	req, err := http.NewRequest(http.MethodGet, u, nil)
 	if err != nil {
 		return nil, err
 	}
 	req.Header.Set("Authorization", "Bearer "+token)
-	ctx, cancel := context.WithTimeout(context.Background(), 120*time.Second)
+	ctx, cancel := context.WithTimeout(parent, 120*time.Second)
 	defer cancel()
 	resp, err := httpclient.DoWithRetry(ctx, nodeHTTPClient(), req, 3)
 	if err != nil {
@@ -105,10 +121,37 @@ func fetchBundle(base, token, speaker, revision string) ([]byte, error) {
 	}
 	defer func() { _ = resp.Body.Close() }()
 	if resp.StatusCode != http.StatusOK {
-		b, _ := io.ReadAll(resp.Body)
+		b, _ := io.ReadAll(io.LimitReader(resp.Body, 4<<10))
 		return nil, fmt.Errorf("bundle: %s: %s", resp.Status, strings.TrimSpace(string(b)))
 	}
-	return io.ReadAll(resp.Body)
+	return readHTTPBytes(resp, l.MaxCompressedBytes)
+}
+
+func readHTTPBytes(resp *http.Response, limit int64) ([]byte, error) {
+	if resp.ContentLength > limit {
+		return nil, bundle.ErrLimit
+	}
+	return bundle.ReadCompressed(resp.Body, bundle.Limits{MaxCompressedBytes: limit})
+}
+
+func readBundleFile(path string, limits bundle.Limits) ([]byte, error) {
+	l, err := limits.Normalized()
+	if err != nil {
+		return nil, err
+	}
+	f, err := os.Open(path)
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = f.Close() }()
+	info, err := f.Stat()
+	if err != nil {
+		return nil, err
+	}
+	if info.Size() > l.MaxCompressedBytes {
+		return nil, bundle.ErrLimit
+	}
+	return bundle.ReadCompressed(f, l)
 }
 
 func loadPubKey(pubB64, pubHex string) (ed25519.PublicKey, error) {
@@ -146,7 +189,7 @@ func CmdVerifyBundle(args []string) int {
 		fmt.Fprintln(os.Stderr, err)
 		return 2
 	}
-	raw, err := os.ReadFile(*path)
+	raw, err := readBundleFile(*path, bundle.Limits{})
 	if err != nil {
 		fmt.Fprintln(os.Stderr, err)
 		return 1
@@ -182,7 +225,7 @@ func CmdApplyBundle(args []string) int {
 		fmt.Fprintln(os.Stderr, err)
 		return 2
 	}
-	raw, err := os.ReadFile(*path)
+	raw, err := readBundleFile(*path, bundle.Limits{})
 	if err != nil {
 		fmt.Fprintln(os.Stderr, err)
 		return 1

internal/nodecli/limits_test.go

@@ -0,0 +1,81 @@
+package nodecli
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"evobgp/internal/bundle"
+)
+
+type countingBody struct {
+	io.Reader
+	read int
+}
+
+func (b *countingBody) Read(p []byte) (int, error) {
+	n, e := b.Reader.Read(p)
+	b.read += n
+	return n, e
+}
+func (b *countingBody) Close() error { return nil }
+
+func TestHTTPBundleBoundKnownAndChunked(t *testing.T) {
+	for _, length := range []int64{-1, 2048} {
+		body := &countingBody{Reader: bytes.NewReader(bytes.Repeat([]byte{'x'}, 2048))}
+		_, err := readHTTPBytes(&http.Response{Body: body, ContentLength: length}, 1024)
+		if !errors.Is(err, bundle.ErrLimit) {
+			t.Fatalf("length=%d limit=%v", length, err)
+		}
+		if body.read > 1025 || (length > 1024 && body.read != 0) {
+			t.Fatalf("unbounded read %d for length %d", body.read, length)
+		}
+	}
+}
+
+func TestLocalBundleSizeBound(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "bundle.tar.gz")
+	if err := os.WriteFile(path, bytes.Repeat([]byte{'x'}, 2048), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := readBundleFile(path, bundle.Limits{MaxCompressedBytes: 1024}); !errors.Is(err, bundle.ErrLimit) {
+		t.Fatalf("local input cap: %v", err)
+	}
+}
+
+func TestBundleHTTPMetadataAndDiagnosticsBound(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch {
+		case strings.Contains(r.URL.Path, "/revisions/latest"):
+			_, _ = io.WriteString(w, strings.Repeat(" ", 65<<10))
+		default:
+			w.WriteHeader(http.StatusBadRequest)
+			_, _ = io.WriteString(w, strings.Repeat("x", 64<<10))
+		}
+	}))
+	defer server.Close()
+	if _, err := fetchLatestRevision(server.URL, "key", "speaker"); !errors.Is(err, bundle.ErrLimit) {
+		t.Fatalf("metadata unbounded: %v", err)
+	}
+	if _, err := fetchBundle(server.URL, "key", "speaker", "revision"); err == nil || len(err.Error()) > 5<<10 {
+		t.Fatalf("diagnostic unbounded: %v", err)
+	}
+}
+
+func TestSyncFetchRespectsCallerCancellation(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+	if _, err := fetchBundleContext(ctx, "http://127.0.0.1:1", "key", "speaker", "revision", bundle.Limits{}); !errors.Is(err, context.Canceled) {
+		t.Fatalf("download ignored cancel: %v", err)
+	}
+	if _, err := fetchLatestRevisionContext(ctx, "http://127.0.0.1:1", "key", "speaker"); !errors.Is(err, context.Canceled) {
+		t.Fatalf("metadata ignored cancel: %v", err)
+	}
+}

internal/nodecli/sync.go

@@ -12,11 +12,11 @@ import (
 
 	"evobgp/internal/birdfmt"
 	"evobgp/internal/bundle"
-	"evobgp/internal/signing"
 )
 
 // SyncConfig drives pull → verify → apply on a replica node.
 type SyncConfig struct {
+	BundleLimits bundle.Limits // zero selects bounded large-config defaults
 	BaseURL      string
 	Token        string
 	SpeakerID    string
@@ -59,12 +59,12 @@ func SyncBundle(ctx context.Context, cfg SyncConfig) (SyncResult, error) {
 	rev := strings.TrimSpace(cfg.RevisionID)
 	if rev == "" {
 		var err error
-		rev, err = fetchLatestRevision(cfg.BaseURL, cfg.Token, cfg.SpeakerID)
+		rev, err = fetchLatestRevisionContext(ctx, cfg.BaseURL, cfg.Token, cfg.SpeakerID)
 		if err != nil {
 			return SyncResult{}, err
 		}
 	}
-	raw, err := fetchBundle(cfg.BaseURL, cfg.Token, cfg.SpeakerID, rev)
+	raw, err := fetchBundleContext(ctx, cfg.BaseURL, cfg.Token, cfg.SpeakerID, rev, cfg.BundleLimits)
 	if err != nil {
 		return SyncResult{}, err
 	}
@@ -75,7 +75,7 @@ func SyncBundle(ctx context.Context, cfg SyncConfig) (SyncResult, error) {
 	if err := os.WriteFile(bundlePath, raw, 0o644); err != nil {
 		return SyncResult{}, err
 	}
-	v, err := signing.VerifyGzippedTar(raw, pub)
+	v, err := bundle.VerifyGzippedTarWithLimits(raw, pub, cfg.BundleLimits)
 	if err != nil {
 		return SyncResult{}, err
 	}
@@ -83,7 +83,7 @@ func SyncBundle(ctx context.Context, cfg SyncConfig) (SyncResult, error) {
 	if err := os.MkdirAll(root, 0o755); err != nil {
 		return SyncResult{}, err
 	}
-	if err := bundle.WriteExtractedFiles(root, v); err != nil {
+	if err := bundle.WriteExtractedFilesWithLimits(root, v, cfg.BundleLimits); err != nil {
 		return SyncResult{}, err
 	}
 	mainRel := v.FindMainBirdConf()

internal/nodedispatch/bird.go

@@ -9,6 +9,7 @@ import (
 	"strings"
 
 	"evobgp/internal/birdfmt"
+	"evobgp/internal/httpclient"
 	"evobgp/internal/store"
 )
 
@@ -52,11 +53,16 @@ func FetchBirdProtocols(ctx context.Context, sp *store.Speaker, opts Options) Bi
 		return res
 	}
 	defer func() { _ = resp.Body.Close() }()
-	b, _ := io.ReadAll(resp.Body)
 	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+		b, _ := io.ReadAll(io.LimitReader(resp.Body, 4<<10))
 		res.Error = fmt.Sprintf("HTTP %d: %s", resp.StatusCode, strings.TrimSpace(string(b)))
 		return res
 	}
+	b, err := httpclient.ReadResponseBody(resp, 8<<20)
+	if err != nil {
+		res.Error = err.Error()
+		return res
+	}
 	var out struct {
 		Sessions []birdfmt.BGPSession `json:"sessions"`
 	}

internal/nodedispatch/dispatch.go

@@ -20,6 +20,7 @@ type Result struct {
 	SpeakerID         string `json:"speaker_id"`
 	Endpoint          string `json:"endpoint,omitempty"`
 	Status            string `json:"status"`
+	Confirmation      string `json:"confirmation,omitempty"`
 	AppliedRevisionID string `json:"applied_revision_id,omitempty"`
 	Error             string `json:"error,omitempty"`
 }
@@ -62,12 +63,13 @@ func Enabled() bool {
 
 // WakeSpeaker POSTs /v1/agent/sync to a replica agent (HTTPS via Traefik).
 func WakeSpeaker(ctx context.Context, sp *store.Speaker, opts Options) Result {
-	res := Result{SpeakerID: sp.ID}
+	res := Result{}
 	if sp == nil {
 		res.Status = "error"
 		res.Error = "nil speaker"
 		return res
 	}
+	res.SpeakerID = sp.ID
 	meta := store.ParseSpeakerMeta(sp.MetaJSON)
 	url := store.AgentSyncURL(meta)
 	if url == "" {
@@ -117,13 +119,31 @@ func WakeSpeaker(ctx context.Context, sp *store.Speaker, opts Options) Result {
 		_ = resp.Body.Close()
 		if resp.StatusCode >= 200 && resp.StatusCode < 300 {
 			var out struct {
+				OK                *bool  `json:"ok"`
 				AppliedRevisionID string `json:"applied_revision_id"`
+				Error             string `json:"error"`
+			}
+			if len(bytes.TrimSpace(b)) > 0 {
+				if err := json.Unmarshal(b, &out); err != nil {
+					res.Status = "error"
+					res.Error = "invalid agent sync response"
+					return res
+				}
+			}
+			if (out.OK != nil && !*out.OK) || strings.TrimSpace(out.Error) != "" {
+				res.Status = "error"
+				res.Error = strings.TrimSpace(out.Error)
+				if res.Error == "" {
+					res.Error = "agent sync reported failure"
+				}
+				return res
 			}
-			_ = json.Unmarshal(b, &out)
 			res.Status = "ok"
 			res.AppliedRevisionID = strings.TrimSpace(out.AppliedRevisionID)
-			if res.AppliedRevisionID == "" {
+			if res.AppliedRevisionID != "" {
-				res.AppliedRevisionID = strings.TrimSpace(opts.RevisionID)
+				res.Confirmation = "agent_reported"
+			} else {
+				res.Confirmation = "unconfirmed"
 			}
 			return res
 		}

internal/nodedispatch/health.go

@@ -8,6 +8,7 @@ import (
 	"net/http"
 	"strings"
 
+	"evobgp/internal/httpclient"
 	"evobgp/internal/store"
 )
 
@@ -43,12 +44,16 @@ func FetchAgentHealth(ctx context.Context, sp *store.Speaker, opts Options) Agen
 		return AgentHealthResult{Error: err.Error()}
 	}
 	defer func() { _ = resp.Body.Close() }()
-	b, _ := io.ReadAll(resp.Body)
 	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+		b, _ := io.ReadAll(io.LimitReader(resp.Body, 4<<10))
 		return AgentHealthResult{
 			Error: fmt.Sprintf("HTTP %d: %s", resp.StatusCode, strings.TrimSpace(string(b))),
 		}
 	}
+	b, err := httpclient.ReadResponseBody(resp, 64<<10)
+	if err != nil {
+		return AgentHealthResult{Error: err.Error()}
+	}
 	var out struct {
 		OK                    bool   `json:"ok"`
 		LastAppliedRevisionID string `json:"last_applied_revision_id"`

internal/nodedispatch/legacy_agent_binary_test.go

@@ -0,0 +1,223 @@
+package nodedispatch_test
+
+import (
+	"context"
+	"crypto/ed25519"
+	"crypto/rand"
+	"encoding/base64"
+	"fmt"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"evobgp/internal/bundle"
+	"evobgp/internal/nodedispatch"
+	"evobgp/internal/store"
+)
+
+func TestMain(m *testing.M) {
+	if os.Getenv("EVOBGP_TEST_LEGACY_BIRD_HELPER") != "1" {
+		os.Exit(m.Run())
+	}
+	args := os.Args[1:]
+	log, err := os.OpenFile(os.Getenv("EVOBGP_TEST_BIRD_LOG"), os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0o600)
+	if err != nil {
+		os.Exit(2)
+	}
+	_, _ = fmt.Fprintln(log, strings.Join(args, " "))
+	_ = log.Close()
+	switch {
+	case len(args) == 3 && args[0] == "-c" && args[2] == "-p":
+		if _, err := os.Stat(args[1]); err == nil {
+			os.Exit(0)
+		}
+	case len(args) == 3 && args[0] == "-s" && args[2] == "configure":
+		os.Exit(0)
+	case len(args) == 5 && args[0] == "-s" && strings.Join(args[2:], " ") == "show protocols all":
+		fmt.Println("peer_test BGP master up 2026-01-01 Established")
+		os.Exit(0)
+	}
+	os.Exit(2)
+}
+
+// TestLegacyAgentBinaryCurrentDispatch runs the historical agent as a separate
+// process. All network traffic stays on loopback and BIRD executables are fakes.
+func TestLegacyAgentBinaryCurrentDispatch(t *testing.T) {
+	binary := os.Getenv("EVOBGP_TEST_LEGACY_AGENT_BIN")
+	if binary == "" {
+		t.Skip("EVOBGP_TEST_LEGACY_AGENT_BIN not set")
+	}
+	binary, err := filepath.Abs(binary)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := os.Stat(binary); err != nil {
+		t.Fatal(err)
+	}
+	const speakerID = "legacy-speaker"
+	const revisionID = "legacy-revision"
+	const nodeToken = "local-node-token"
+	const agentSecret = "local-agent-secret"
+	pub, priv, err := ed25519.GenerateKey(rand.Reader)
+	if err != nil {
+		t.Fatal(err)
+	}
+	archive, err := bundle.BuildGzippedTar(revisionID, speakerID, map[string]string{
+		"bird.conf": "router id 192.0.2.1;\n",
+	}, priv)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var bundleGets atomic.Int32
+	controlPlane := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.Header.Get("Authorization") != "Bearer "+nodeToken {
+			http.Error(w, "unauthorized", http.StatusUnauthorized)
+			return
+		}
+		switch r.URL.Path {
+		case "/v1/speakers/" + speakerID + "/bundle/" + revisionID:
+			bundleGets.Add(1)
+			w.Header().Set("Content-Type", "application/gzip")
+			_, _ = w.Write(archive)
+		default:
+			http.NotFound(w, r)
+		}
+	}))
+	defer controlPlane.Close()
+
+	root := t.TempDir()
+	logPath := filepath.Join(root, "bird-invocations")
+	helper, err := os.Executable()
+	if err != nil {
+		t.Fatal(err)
+	}
+	listener, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatal(err)
+	}
+	agentAddr := listener.Addr().String()
+	_ = listener.Close()
+	outputPath := filepath.Join(root, "agent.log")
+	output, err := os.Create(outputPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() { _ = output.Close() }()
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	cmd := exec.CommandContext(ctx, binary, "-listen", agentAddr, "serve")
+	cmd.Stdout, cmd.Stderr = output, output
+	cmd.Env = append(os.Environ(),
+		"EVOBGP_AGENT_SECRET="+agentSecret,
+		"EVOBGP_CONTROL_PLANE_URL="+controlPlane.URL,
+		"EVOBGP_NODE_TOKEN="+nodeToken,
+		"EVOBGP_SPEAKER_ID="+speakerID,
+		"EVOBGP_BUNDLE_PUBKEY_BASE64="+base64.StdEncoding.EncodeToString(pub),
+		"EVOBGP_BIRD_EXTRACT_DIR="+filepath.Join(root, "extract"),
+		"EVOBGP_BIRD_BIN="+helper,
+		"EVOBGP_BIRDC_BIN="+helper,
+		"EVOBGP_BIRDC_SOCKET="+filepath.Join(root, "fake.ctl"),
+		"EVOBGP_TEST_LEGACY_BIRD_HELPER=1",
+		"EVOBGP_TEST_BIRD_LOG="+logPath,
+		"TMPDIR="+root,
+		"TMP="+root,
+		"TEMP="+root,
+	)
+	if err := cmd.Start(); err != nil {
+		t.Fatal(err)
+	}
+	defer func() {
+		cancel()
+		_ = cmd.Wait()
+	}()
+
+	transport := http.DefaultTransport.(*http.Transport).Clone()
+	// Under -race each helper process can spend a second flushing race output at
+	// exit. Sync executes parse-check and configure before returning its ACK.
+	client := &http.Client{Timeout: 10 * time.Second, Transport: legacyAgentTransport{addr: agentAddr, base: transport}}
+	defer transport.CloseIdleConnections()
+	sp := &store.Speaker{ID: speakerID, Role: "replica", MetaJSON: store.SpeakerMetaJSON(store.SpeakerMeta{
+		AgentDomain: "legacy-agent.example.test", AgentSecret: agentSecret,
+	})}
+	deadline := time.Now().Add(10 * time.Second)
+	for {
+		health := nodedispatch.FetchAgentHealth(context.Background(), sp, nodedispatch.Options{HTTPClient: client})
+		if health.OK {
+			if health.LastAppliedRevisionID != "" {
+				t.Fatalf("initial health has applied revision: %+v", health)
+			}
+			break
+		}
+		if time.Now().After(deadline) {
+			log, _ := os.ReadFile(outputPath)
+			t.Fatalf("legacy agent did not become healthy: %+v; output: %s", health, log)
+		}
+		time.Sleep(50 * time.Millisecond)
+	}
+
+	// The historical agent must enforce the bearer token before reaching sync.
+	badReq, err := http.NewRequest(http.MethodPost, "https://legacy-agent.example.test/v1/agent/sync", strings.NewReader(`{"revision_id":"legacy-revision"}`))
+	if err != nil {
+		t.Fatal(err)
+	}
+	badReq.Header.Set("Authorization", "Bearer wrong-secret")
+	badResp, err := client.Do(badReq)
+	if err != nil {
+		t.Fatal(err)
+	}
+	_ = badResp.Body.Close()
+	if badResp.StatusCode != http.StatusUnauthorized || bundleGets.Load() != 0 {
+		t.Fatalf("unauthorized sync: status=%d bundle_gets=%d", badResp.StatusCode, bundleGets.Load())
+	}
+
+	result := nodedispatch.WakeSpeaker(context.Background(), sp, nodedispatch.Options{
+		HTTPClient: client, MaxRetries: 1, RevisionID: revisionID,
+	})
+	if result.Status != "ok" || result.Confirmation != "agent_reported" || result.AppliedRevisionID != revisionID {
+		log, _ := os.ReadFile(outputPath)
+		t.Fatalf("current dispatch to legacy agent: %+v; output: %s", result, log)
+	}
+	if bundleGets.Load() != 1 {
+		t.Fatalf("signed bundle gets = %d, want 1", bundleGets.Load())
+	}
+	health := nodedispatch.FetchAgentHealth(context.Background(), sp, nodedispatch.Options{HTTPClient: client})
+	if !health.OK || health.LastAppliedRevisionID != revisionID || health.LastSyncAt == "" {
+		t.Fatalf("post-sync legacy health: %+v", health)
+	}
+	protocols := nodedispatch.FetchBirdProtocols(context.Background(), sp, nodedispatch.Options{HTTPClient: client})
+	if protocols.Error != "" || len(protocols.Sessions) != 1 || protocols.Sessions[0].State != "Established" {
+		t.Fatalf("legacy bird protocols: %+v", protocols)
+	}
+	mainConfig, err := os.ReadFile(filepath.Join(root, "extract", "bird.conf"))
+	if err != nil || string(mainConfig) != "router id 192.0.2.1;\n" {
+		t.Fatalf("extracted signed bundle: content=%q err=%v", mainConfig, err)
+	}
+	invocations, err := os.ReadFile(logPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	wantParse := fmt.Sprintf("-c %s -p", filepath.Join(root, "extract", "bird.conf"))
+	if !strings.Contains(string(invocations), wantParse) || !strings.Contains(string(invocations), "-s "+filepath.Join(root, "fake.ctl")+" configure") {
+		t.Fatalf("legacy agent did not parse-check and configure via stand-ins: %s", invocations)
+	}
+}
+
+type legacyAgentTransport struct {
+	addr string
+	base http.RoundTripper
+}
+
+func (tr legacyAgentTransport) RoundTrip(req *http.Request) (*http.Response, error) {
+	copyReq := req.Clone(req.Context())
+	copyReq.URL.Scheme = "http"
+	copyReq.URL.Host = tr.addr
+	copyReq.Host = tr.addr
+	return tr.base.RoundTrip(copyReq)
+}

internal/nodedispatch/legacy_compatibility_test.go

@@ -0,0 +1,106 @@
+package nodedispatch_test
+
+import (
+	"context"
+	"encoding/json"
+	"io"
+	"net/http"
+	"reflect"
+	"strings"
+	"testing"
+
+	"evobgp/internal/nodedispatch"
+	"evobgp/internal/store"
+)
+
+type legacyRoundTripFunc func(*http.Request) (*http.Response, error)
+
+func (f legacyRoundTripFunc) RoundTrip(r *http.Request) (*http.Response, error) { return f(r) }
+
+func TestLegacyCompatibilityAgentSync(t *testing.T) {
+	for _, tc := range []struct {
+		name, response, wantApplied string
+		status                      int
+	}{
+		{"legacy bare success", `{"ok":true}`, "", http.StatusOK},
+		{"legacy empty no-content", "", "", http.StatusNoContent},
+		{"legacy explicit revision", `{"ok":true,"applied_revision_id":"revision-1"}`, "revision-1", http.StatusOK},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			var gotAuth, gotType string
+			var gotBody map[string]json.RawMessage
+			client := &http.Client{Transport: legacyRoundTripFunc(func(r *http.Request) (*http.Response, error) {
+				if r.Method != http.MethodPost || r.URL.Path != "/v1/agent/sync" {
+					t.Errorf("unexpected sync request: %s %s", r.Method, r.URL)
+				}
+				gotAuth, gotType = r.Header.Get("Authorization"), r.Header.Get("Content-Type")
+				body, _ := io.ReadAll(r.Body)
+				if err := json.Unmarshal(body, &gotBody); err != nil {
+					t.Errorf("sync JSON: %v", err)
+				}
+				return &http.Response{StatusCode: tc.status, Body: io.NopCloser(strings.NewReader(tc.response)), Header: make(http.Header)}, nil
+			})}
+			sp := &store.Speaker{ID: "speaker-1", Role: "replica", MetaJSON: store.SpeakerMetaJSON(store.SpeakerMeta{
+				AgentDomain: "agent.example.test", AgentSecret: "legacy-secret",
+			})}
+			got := nodedispatch.WakeSpeaker(context.Background(), sp, nodedispatch.Options{HTTPClient: client, MaxRetries: 1, RevisionID: "revision-1"})
+			if got.Status != "ok" || got.AppliedRevisionID != tc.wantApplied {
+				t.Fatalf("sync result: %+v", got)
+			}
+			wantConfirmation := "unconfirmed"
+			if tc.wantApplied != "" {
+				wantConfirmation = "agent_reported"
+			}
+			if got.Confirmation != wantConfirmation {
+				t.Fatalf("confirmation = %q, want %q", got.Confirmation, wantConfirmation)
+			}
+			if gotAuth != "Bearer legacy-secret" || gotType != "application/json" {
+				t.Fatalf("legacy headers: authorization=%q content-type=%q", gotAuth, gotType)
+			}
+			if !reflect.DeepEqual(gotBody, map[string]json.RawMessage{"revision_id": json.RawMessage(`"revision-1"`)}) {
+				t.Fatalf("unexpected fields sent to strict legacy decoder: %s", gotBody)
+			}
+		})
+	}
+}
+
+func TestAgentSyncRejectsMalformedSuccess(t *testing.T) {
+	for _, response := range []string{`{"ok":false}`, `{"ok":true`, `{"ok":false,"applied_revision_id":"revision-1"}`} {
+		t.Run(response, func(t *testing.T) {
+			client := &http.Client{Transport: legacyRoundTripFunc(func(*http.Request) (*http.Response, error) {
+				return &http.Response{StatusCode: http.StatusOK, Body: io.NopCloser(strings.NewReader(response)), Header: make(http.Header)}, nil
+			})}
+			sp := &store.Speaker{ID: "speaker-1", Role: "replica", MetaJSON: store.SpeakerMetaJSON(store.SpeakerMeta{AgentDomain: "agent.example.test", AgentSecret: "secret"})}
+			got := nodedispatch.WakeSpeaker(context.Background(), sp, nodedispatch.Options{HTTPClient: client, MaxRetries: 1, RevisionID: "revision-1"})
+			if got.Status != "error" || got.AppliedRevisionID != "" {
+				t.Fatalf("invalid 2xx response accepted: %+v", got)
+			}
+		})
+	}
+}
+
+func TestLegacyCompatibilityAgentHealthMissingNewFields(t *testing.T) {
+	for _, tc := range []struct {
+		name, body, wantRevision, wantSync string
+	}{
+		{"bare legacy health", `{"ok":true}`, "", ""},
+		{"revision without time", `{"ok":true,"last_applied_revision_id":"revision-1"}`, "revision-1", ""},
+		{"revision with time", `{"ok":true,"last_applied_revision_id":"revision-1","last_sync_at":"2026-10-26T00:00:00Z"}`, "revision-1", "2026-10-26T00:00:00Z"},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			client := &http.Client{Transport: legacyRoundTripFunc(func(r *http.Request) (*http.Response, error) {
+				if r.Method != http.MethodGet || r.URL.Path != "/v1/agent/health" || r.Header.Get("Authorization") != "Bearer legacy-secret" {
+					t.Errorf("unexpected health request: %s %s", r.Method, r.URL)
+				}
+				return &http.Response{StatusCode: http.StatusOK, Body: io.NopCloser(strings.NewReader(tc.body)), Header: make(http.Header)}, nil
+			})}
+			sp := &store.Speaker{ID: "speaker-1", MetaJSON: store.SpeakerMetaJSON(store.SpeakerMeta{
+				AgentDomain: "agent.example.test", AgentSecret: "legacy-secret",
+			})}
+			got := nodedispatch.FetchAgentHealth(context.Background(), sp, nodedispatch.Options{HTTPClient: client})
+			if !got.OK || got.Error != "" || got.LastAppliedRevisionID != tc.wantRevision || got.LastSyncAt != tc.wantSync {
+				t.Fatalf("legacy health: %+v", got)
+			}
+		})
+	}
+}

internal/nodedispatch/telemetry_limits_test.go

@@ -0,0 +1,80 @@
+package nodedispatch_test
+
+import (
+	"context"
+	"errors"
+	"io"
+	"net/http"
+	"strings"
+	"testing"
+
+	"evobgp/internal/nodedispatch"
+	"evobgp/internal/store"
+)
+
+type telemetryTestBody struct {
+	io.Reader
+	read   int
+	closed bool
+}
+
+func (b *telemetryTestBody) Read(p []byte) (int, error) {
+	n, err := b.Reader.Read(p)
+	b.read += n
+	return n, err
+}
+func (b *telemetryTestBody) Close() error { b.closed = true; return nil }
+
+type telemetryErrorReader struct{}
+
+func (telemetryErrorReader) Read([]byte) (int, error) { return 0, errors.New("injected body failure") }
+
+func TestAgentTelemetryCompleteBodyLimits(t *testing.T) {
+	sp := &store.Speaker{ID: "legacy", Role: "replica", MetaJSON: store.SpeakerMetaJSON(store.SpeakerMeta{AgentDomain: "agent.example.test", AgentSecret: "secret"})}
+	for _, tc := range []struct {
+		name, prefix string
+		limit        int
+		call         func(nodedispatch.Options) string
+	}{
+		{"health", `{"ok":true}`, 64 << 10, func(opts nodedispatch.Options) string {
+			return nodedispatch.FetchAgentHealth(context.Background(), sp, opts).Error
+		}},
+		{"protocols", `{"sessions":[]}`, 8 << 20, func(opts nodedispatch.Options) string {
+			return nodedispatch.FetchBirdProtocols(context.Background(), sp, opts).Error
+		}},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			for _, known := range []bool{true, false} {
+				for _, extra := range []int{0, 1} {
+					data := tc.prefix + strings.Repeat(" ", tc.limit-len(tc.prefix)+extra)
+					body := &telemetryTestBody{Reader: strings.NewReader(data)}
+					client := &http.Client{Transport: legacyRoundTripFunc(func(*http.Request) (*http.Response, error) {
+						length := int64(-1)
+						if known {
+							length = int64(len(data))
+						}
+						return &http.Response{StatusCode: 200, ContentLength: length, Body: body, Header: make(http.Header)}, nil
+					})}
+					err := tc.call(nodedispatch.Options{HTTPClient: client})
+					if (extra == 0 && err != "") || (extra == 1 && err == "") || body.read > tc.limit+1 || !body.closed {
+						t.Fatalf("known=%v extra=%d read=%d closed=%v error=%s", known, extra, body.read, body.closed, err)
+					}
+				}
+			}
+			for _, status := range []int{200, 400} {
+				reader := io.MultiReader(strings.NewReader(tc.prefix), telemetryErrorReader{})
+				if status == 400 {
+					reader = io.MultiReader(strings.NewReader(strings.Repeat("x", 16000)))
+				}
+				body := &telemetryTestBody{Reader: reader}
+				client := &http.Client{Transport: legacyRoundTripFunc(func(*http.Request) (*http.Response, error) {
+					return &http.Response{StatusCode: status, ContentLength: -1, Body: body, Header: make(http.Header)}, nil
+				})}
+				err := tc.call(nodedispatch.Options{HTTPClient: client})
+				if err == "" || !body.closed || (status == 400 && body.read > 4<<10) {
+					t.Fatalf("status=%d read=%d closed=%v error=%s", status, body.read, body.closed, err)
+				}
+			}
+		})
+	}
+}

internal/observability/metrics.go

@@ -20,10 +20,12 @@ import (
 const namespace = "evobgp"
 
 var (
-	metricsStore          atomic.Value // store.Backend
+	metricsStore          atomic.Value // storeBackendHolder
 	registerCollectorOnce sync.Once
 )
 
+type storeBackendHolder struct{ backend store.Backend }
+
 var (
 	httpRequests = promauto.NewCounterVec(
 		prometheus.CounterOpts{
@@ -117,7 +119,7 @@ var (
 	upstreamBreakerOpen = promauto.NewGaugeVec(prometheus.GaugeOpts{
 		Namespace: namespace,
 		Name:      "upstream_breaker_open",
-		Help:      "1 if per-host HTTP circuit breaker is open.",
+		Help:      "Last request-time observation of the per-host circuit breaker (1=open, 0=closed); idle host series expire with the bounded cache.",
 	}, []string{"host"})
 )
 
@@ -173,6 +175,9 @@ func RecordJobTerminal(kind, status string) {
 
 // RecordJobAuditDepth updates durable job_audit depth gauges.
 func RecordJobAuditDepth(counts map[string]int64) {
+	for _, status := range []string{"queued", "running", "succeeded", "failed", "cancelled"} {
+		jobAuditDepth.WithLabelValues(status).Set(float64(counts[status]))
+	}
 	for status, n := range counts {
 		if status == "" {
 			status = "unknown"
@@ -193,6 +198,15 @@ func SetUpstreamBreakerOpen(host string, open bool) {
 	upstreamBreakerOpen.WithLabelValues(host).Set(v)
 }
 
+// RemoveUpstreamBreaker removes an expired host series together with its cache entry.
+// Callers must retain active entries until their final metric update completes.
+func RemoveUpstreamBreaker(host string) {
+	if host == "" {
+		host = "_"
+	}
+	upstreamBreakerOpen.DeleteLabelValues(host)
+}
+
 // SetBuildInfo sets evobgp_build_info gauge (idempotent labels).
 func SetBuildInfo(version, gitSHA string) {
 	if version == "" {
@@ -209,30 +223,45 @@ type memoryStoreCollector struct {
 	prefixSumDesc         *prometheus.Desc
 	peersDesc             *prometheus.Desc
 	peerStateDesc         *prometheus.Desc
+	storedMaxDesc         *prometheus.Desc
+	storedSumDesc         *prometheus.Desc
+	inventoryDesc         *prometheus.Desc
+	storedStateDesc       *prometheus.Desc
+	collectionSuccessDesc *prometheus.Desc
 }
 
 func newMemoryStoreCollector() *memoryStoreCollector {
 	return &memoryStoreCollector{
 		prefixMaxDesc: prometheus.NewDesc(
 			prometheus.BuildFQName(namespace, "", "materialized_prefixes_max"),
-			"Maximum materialized_prefix_count among all revisions in the store.",
+			"Deprecated: maximum materialized_prefix_count among all stored revisions; use stored_revision_materialized_prefixes_max.",
 			nil, nil,
 		),
 		prefixSumDesc: prometheus.NewDesc(
 			prometheus.BuildFQName(namespace, "", "materialized_prefixes_sum"),
-			"Sum of materialized_prefix_count over revisions (development aggregate).",
+			"Deprecated: sum of materialized_prefix_count over all stored revisions; use stored_revision_materialized_prefixes_sum.",
 			nil, nil,
 		),
 		peersDesc: prometheus.NewDesc(
 			prometheus.BuildFQName(namespace, "", "bgp_peers_configured_total"),
-			"BGP peers configured in the control-plane store.",
+			"Deprecated: BGP peers configured in the control-plane store; use inventory_bgp_peers.",
 			nil, nil,
 		),
 		peerStateDesc: prometheus.NewDesc(
 			prometheus.BuildFQName(namespace, "", "bgp_peer_sessions"),
-			"Configured BGP peers in the store by session_state (intent / last known, not live BIRD).",
+			"Deprecated: configured BGP peers by persisted session_state, not live BIRD; use stored_bgp_peer_session_state.",
 			[]string{"state"}, nil,
 		),
+		storedMaxDesc: prometheus.NewDesc(prometheus.BuildFQName(namespace, "", "stored_revision_materialized_prefixes_max"),
+			"Maximum materialized prefix count across all stored revisions; historical, not current routes.", nil, nil),
+		storedSumDesc: prometheus.NewDesc(prometheus.BuildFQName(namespace, "", "stored_revision_materialized_prefixes_sum"),
+			"Sum of materialized prefix counts across all stored revisions; historical, not current routes.", nil, nil),
+		inventoryDesc: prometheus.NewDesc(prometheus.BuildFQName(namespace, "", "inventory_bgp_peers"),
+			"Configured BGP peer inventory in the control-plane store, regardless of live state.", nil, nil),
+		storedStateDesc: prometheus.NewDesc(prometheus.BuildFQName(namespace, "", "stored_bgp_peer_session_state"),
+			"BGP peer inventory grouped by persisted session_state; not a live observation.", []string{"state"}, nil),
+		collectionSuccessDesc: prometheus.NewDesc(prometheus.BuildFQName(namespace, "", "store_metrics_collection_success"),
+			"1 if the latest store metrics collection succeeded, 0 on database aggregation error; other store metrics are omitted on error.", nil, nil),
 	}
 }
 
@@ -241,6 +270,11 @@ func (c *memoryStoreCollector) Describe(ch chan<- *prometheus.Desc) {
 	ch <- c.prefixSumDesc
 	ch <- c.peersDesc
 	ch <- c.peerStateDesc
+	ch <- c.storedMaxDesc
+	ch <- c.storedSumDesc
+	ch <- c.inventoryDesc
+	ch <- c.storedStateDesc
+	ch <- c.collectionSuccessDesc
 }
 
 func (c *memoryStoreCollector) Collect(ch chan<- prometheus.Metric) {
@@ -248,19 +282,44 @@ func (c *memoryStoreCollector) Collect(ch chan<- prometheus.Metric) {
 	if v == nil {
 		return
 	}
-	b, ok := v.(store.Backend)
+	holder, ok := v.(storeBackendHolder)
-	if !ok || b == nil {
+	if !ok || holder.backend == nil {
 		return
 	}
-	maxN, sumN := b.MaterializedPrefixStats()
+	b := holder.backend
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	maxN, sumN, peers, states, err := b.ObserveStoreMetrics(ctx)
+	cancel()
+	if err != nil {
+		ch <- prometheus.MustNewConstMetric(c.collectionSuccessDesc, prometheus.GaugeValue, 0)
+		return
+	}
+	ch <- prometheus.MustNewConstMetric(c.collectionSuccessDesc, prometheus.GaugeValue, 1)
 	ch <- prometheus.MustNewConstMetric(c.prefixMaxDesc, prometheus.GaugeValue, float64(maxN))
 	ch <- prometheus.MustNewConstMetric(c.prefixSumDesc, prometheus.GaugeValue, float64(sumN))
-	ch <- prometheus.MustNewConstMetric(c.peersDesc, prometheus.GaugeValue, float64(b.PeerCount()))
+	ch <- prometheus.MustNewConstMetric(c.storedMaxDesc, prometheus.GaugeValue, float64(maxN))
-	for state, n := range b.PeerSessionCountsByState() {
+	ch <- prometheus.MustNewConstMetric(c.storedSumDesc, prometheus.GaugeValue, float64(sumN))
+	ch <- prometheus.MustNewConstMetric(c.peersDesc, prometheus.GaugeValue, float64(peers))
+	ch <- prometheus.MustNewConstMetric(c.inventoryDesc, prometheus.GaugeValue, float64(peers))
+	canonicalStates := make(map[string]int)
+	for state, n := range states {
 		if state == "" {
 			state = "unknown"
 		}
 		ch <- prometheus.MustNewConstMetric(c.peerStateDesc, prometheus.GaugeValue, float64(n), state)
+		canonicalStates[canonicalPeerState(state)] += n
+	}
+	for state, n := range canonicalStates {
+		ch <- prometheus.MustNewConstMetric(c.storedStateDesc, prometheus.GaugeValue, float64(n), state)
+	}
+}
+
+func canonicalPeerState(state string) string {
+	switch strings.ToLower(strings.TrimSpace(state)) {
+	case "idle", "connect", "active", "opensent", "openconfirm", "established", "unknown":
+		return strings.ToLower(strings.TrimSpace(state))
+	default:
+		return "other"
 	}
 }
 
@@ -269,7 +328,7 @@ func RegisterStoreBackend(b store.Backend) {
 	if b == nil {
 		return
 	}
-	metricsStore.Store(b)
+	metricsStore.Store(storeBackendHolder{backend: b})
 	registerCollectorOnce.Do(func() {
 		prometheus.DefaultRegisterer.MustRegister(newMemoryStoreCollector())
 	})

internal/observability/metrics_store_test.go

@@ -0,0 +1,116 @@
+package observability
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	"github.com/prometheus/client_golang/prometheus"
+
+	"evobgp/internal/store"
+)
+
+type failingMetricsBackend struct{ store.Backend }
+
+func (failingMetricsBackend) ObserveStoreMetrics(context.Context) (int, int, int, map[string]int, error) {
+	return 0, 0, 0, nil, errors.New("database unavailable")
+}
+
+type storedMetricsBackend struct{ store.Backend }
+
+func (storedMetricsBackend) ObserveStoreMetrics(context.Context) (int, int, int, map[string]int, error) {
+	return 12, 24, 1, map[string]int{"established": 1}, nil
+}
+
+type legacyStateBackend struct{ store.Backend }
+
+func (legacyStateBackend) ObserveStoreMetrics(context.Context) (int, int, int, map[string]int, error) {
+	return 0, 0, 2, map[string]int{" Established ": 1, "established": 1}, nil
+}
+
+func TestStoreMetricsCollectionErrorOmitsAggregates(t *testing.T) {
+	metricsStore.Store(storeBackendHolder{backend: failingMetricsBackend{}})
+	registry := prometheus.NewRegistry()
+	if err := registry.Register(newMemoryStoreCollector()); err != nil {
+		t.Fatal(err)
+	}
+	families, err := registry.Gather()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(families) != 1 || families[0].GetName() != "evobgp_store_metrics_collection_success" {
+		t.Fatalf("expected only collection status on error, got %v", families)
+	}
+	if got := families[0].Metric[0].Gauge.GetValue(); got != 0 {
+		t.Fatalf("collection status = %v, want 0", got)
+	}
+}
+
+func TestStoredRevisionMetricsDoNotClaimCurrentRoutes(t *testing.T) {
+	metricsStore.Store(storeBackendHolder{backend: storedMetricsBackend{}})
+	registry := prometheus.NewRegistry()
+	if err := registry.Register(newMemoryStoreCollector()); err != nil {
+		t.Fatal(err)
+	}
+	families, err := registry.Gather()
+	if err != nil {
+		t.Fatal(err)
+	}
+	values := make(map[string]float64)
+	for _, family := range families {
+		if len(family.Metric) > 0 && family.Metric[0].Gauge != nil {
+			values[family.GetName()] = family.Metric[0].Gauge.GetValue()
+		}
+	}
+	if values["evobgp_stored_revision_materialized_prefixes_sum"] != 24 || values["evobgp_materialized_prefixes_sum"] != 24 {
+		t.Fatalf("stored revision and legacy sums must match: %v", values)
+	}
+	if values["evobgp_store_metrics_collection_success"] != 1 {
+		t.Fatalf("successful collection not reported: %v", values)
+	}
+	if _, ok := values["evobgp_current_routes"]; ok {
+		t.Fatal("historical revision sum must not be published as current routes")
+	}
+}
+
+func TestCanonicalPeerStateBoundsNewLabelValues(t *testing.T) {
+	if got := canonicalPeerState("Established"); got != "established" {
+		t.Fatalf("state = %q", got)
+	}
+	if got := canonicalPeerState("unexpected peer text"); got != "other" {
+		t.Fatalf("state = %q", got)
+	}
+}
+
+func TestLegacyPeerStateLabelPreservedWhileNewLabelIsCanonical(t *testing.T) {
+	metricsStore.Store(storeBackendHolder{backend: legacyStateBackend{}})
+	registry := prometheus.NewRegistry()
+	if err := registry.Register(newMemoryStoreCollector()); err != nil {
+		t.Fatal(err)
+	}
+	families, err := registry.Gather()
+	if err != nil {
+		t.Fatal(err)
+	}
+	legacyRaw, newCanonical := false, false
+	for _, family := range families {
+		for _, metric := range family.Metric {
+			if len(metric.Label) != 1 {
+				continue
+			}
+			switch family.GetName() {
+			case "evobgp_bgp_peer_sessions":
+				if metric.Label[0].GetValue() == " Established " && metric.Gauge.GetValue() == 1 {
+					legacyRaw = true
+				}
+			case "evobgp_stored_bgp_peer_session_state":
+				if metric.Label[0].GetValue() == "established" && metric.Gauge.GetValue() == 2 {
+					newCanonical = true
+				}
+			}
+		}
+	}
+	if !legacyRaw || !newCanonical {
+		t.Fatalf("legacy raw state preserved=%v, new canonical state aggregated=%v", legacyRaw, newCanonical)
+	}
+}

internal/pipeline/aggregate.go

@@ -86,20 +86,26 @@ func aggregateTenantPrefixRowsAll(ctx context.Context, st store.Backend, hc *htt
 }
 
 func rowsForModule(ctx context.Context, st store.Backend, hc *http.Client, tenantID string, mod *store.Module) ([]store.PrefixRow, error) {
-	if rows, ok, err := moduleRowsFromSnapshot(st, tenantID, mod); err != nil {
+	snap, ok, err := st.GetModulePrefixSnapshot(ctx, tenantID, mod.ID)
-		return nil, err
-	} else if ok {
-		return rows, nil
-	}
-	var prior []store.PrefixRow
-	if snap, ok, _ := st.GetModulePrefixSnapshot(tenantID, mod.ID); ok && snap != nil {
-		prior = snap.Prefixes
-	}
-	rows, err := collectModulePrefixRows(ctx, st, hc, tenantID, mod, prior)
 	if err != nil {
 		return nil, err
 	}
-	if err := persistModuleSnapshot(st, tenantID, mod, rows); err != nil {
+	if ok && snap != nil {
+		if rows, matched, err := snapshotRowsIfCurrent(ctx, st, tenantID, mod, snap); err != nil {
+			return nil, err
+		} else if matched {
+			return rows, nil
+		}
+	}
+	var priorRows []store.PrefixRow
+	if ok && snap != nil {
+		priorRows = snap.Prefixes
+	}
+	rows, err := collectModulePrefixRows(ctx, st, hc, tenantID, mod, modulePrior{provided: true, rows: priorRows})
+	if err != nil {
+		return nil, err
+	}
+	if err := persistModuleSnapshot(ctx, st, tenantID, mod, rows); err != nil {
 		return nil, err
 	}
 	return rows, nil