fix(health): разворачивать CNAME до IP в статусе таблицы

Пробы CNAME больше не ключуются hostname: цель разворачивается до origin/пула, а уже сохранённый статус по CNAME копируется на IP сервиса.

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

apps/api/src/services/service-config-service.ts

@@ -18,6 +18,7 @@ import {
   SYNC_PENDING_PUSH,
   SYNC_SYNCED,
   dnsRecordNamesMatch,
+  isIpLiteral,
   normalizeDnsRecordName,
 } from "@cfdm/shared";
 import type { CloudflareClient } from "../lib/cf-client.js";
@@ -344,6 +345,64 @@ async function buildView(db: Db, serviceId: number): Promise<ServiceView> {
   };
 }
 
+const HEALTH_RANK: Record<string, number> = {
+  down: 3,
+  degraded: 2,
+  unknown: 1,
+  up: 0,
+};
+
+function cnameLookupKeys(value: string, zoneName?: string | null): string[] {
+  const trimmed = value.trim();
+  if (!trimmed) return [];
+  const noDot = trimmed.replace(/\.+$/, "");
+  const lower = noDot.toLowerCase();
+  const keys = new Set([trimmed, noDot, lower]);
+  if (zoneName && !lower.includes(".")) {
+    keys.add(`${lower}.${zoneName.trim().toLowerCase().replace(/\.+$/, "")}`);
+  }
+  return [...keys];
+}
+
+type ServiceHealthRow = {
+  ip: string;
+  status: IpHealthState;
+  latency_ms: number | null;
+  last_checked_at: string | null;
+  last_error: string | null;
+  provider: ServiceView["ip_health"][number]["provider"];
+  colo: string | null;
+};
+
+/** Health rows keyed by CNAME hostname (legacy probes) applied to service IPs. */
+function fallbackCnameHealth(
+  rows: ServiceHealthRow[],
+  view: ServiceView,
+): ServiceHealthRow | undefined {
+  const cnameKeys = new Set<string>();
+  for (const domain of view.domains ?? []) {
+    const cname = domain.target_cname?.trim();
+    if (!cname) continue;
+    for (const key of cnameLookupKeys(cname, domain.zone_name)) {
+      cnameKeys.add(key);
+    }
+  }
+  const hostnameRows = rows.filter((row) => !isIpLiteral(row.ip));
+  if (hostnameRows.length === 0) return undefined;
+  const matched =
+    cnameKeys.size === 0
+      ? hostnameRows
+      : hostnameRows.filter((row) =>
+          cnameLookupKeys(row.ip).some((key) => cnameKeys.has(key)),
+        );
+  const candidates = matched.length > 0 ? matched : hostnameRows;
+  return candidates.reduce((worst, row) =>
+    (HEALTH_RANK[row.status] ?? 0) > (HEALTH_RANK[worst.status] ?? 0)
+      ? row
+      : worst,
+  );
+}
+
 function attachServiceHealth(
   db: Db,
   views: ServiceView[],
@@ -353,11 +412,16 @@ function attachServiceHealth(
   const ipHealthByService = repos.listIpHealthByServiceIds(db, ids);
   return views.map((view) => {
     const health = healthByService.get(view.id);
-    const byIp = new Map(
+    const rows = ipHealthByService.get(view.id) ?? [];
-      (ipHealthByService.get(view.id) ?? []).map((row) => [row.ip, row]),
+    const byIp = new Map(rows.map((row) => [row.ip, row]));
+    const cnameFallback = fallbackCnameHealth(rows, view);
+    const aRecordIps = new Set(
+      (view.domains ?? []).flatMap((domain) =>
+        domain.target_cname?.trim() ? [] : (domain.target_ips ?? []),
+      ),
     );
     const ip_health = (view.ips ?? []).map((ip) => {
-      const row = byIp.get(ip);
+      const row = byIp.get(ip) ?? (aRecordIps.has(ip) ? undefined : cnameFallback);
       return {
         ip,
         status: row?.status ?? ("unknown" as const),

apps/api/test/health-check.test.ts

@@ -235,4 +235,95 @@ describe("health-check state derivation via runAllChecks", () => {
     expect(targets).toHaveLength(1);
     expect(targets[0]?.verify_tls).toBe(true);
   });
+
+  it("unwraps CNAME target to origin A record IPs", async () => {
+    const { createMemoryDb, repos, runMigrations } = await import("@cfdm/db");
+    const { db, sqlite } = createMemoryDb();
+    runMigrations(sqlite);
+
+    const domain = repos.createDomain(db, null, "rkns.top", "zone-id");
+    repos.insertDnsRecord(
+      db,
+      domain.id,
+      "A",
+      "ihome",
+      "2.59.161.102",
+      1,
+      false,
+      null,
+      "synced",
+      "cf",
+      null,
+    );
+    const service = repos.createService(db, "RW Sub", "rw-sub");
+    const binding = repos.insertBinding(db, domain.id, service.id, "s", null);
+    repos.setBindingCnameTarget(db, binding.id, "ihome.rkns.top");
+    repos.updateBindingLbConfig(db, binding.id, {
+      health_check_enabled: true,
+      health_check_type: "tcp",
+      health_check_port: 443,
+    });
+
+    const targets = repos.listHealthCheckTargets(db);
+    expect(targets).toHaveLength(1);
+    expect(targets[0]?.ip).toBe("2.59.161.102");
+    expect(targets[0]?.hostname).toBe("s.rkns.top");
+  });
+
+  it("unwraps CNAME target to service IP pool when origin DNS is empty", async () => {
+    const { createMemoryDb, repos, runMigrations } = await import("@cfdm/db");
+    const { db, sqlite } = createMemoryDb();
+    runMigrations(sqlite);
+
+    const domain = repos.createDomain(db, null, "rkns.top", "zone-id");
+    const service = repos.createService(db, "RW Sub", "rw-sub");
+    repos.replaceServiceIps(db, service.id, ["2.59.161.102"]);
+    const binding = repos.insertBinding(db, domain.id, service.id, "s", null);
+    repos.setBindingCnameTarget(db, binding.id, "ihome.rkns.top");
+    repos.updateBindingLbConfig(db, binding.id, {
+      health_check_enabled: true,
+      health_check_type: "tcp",
+      health_check_port: 443,
+    });
+
+    const targets = repos.listHealthCheckTargets(db);
+    expect(targets).toHaveLength(1);
+    expect(targets[0]?.ip).toBe("2.59.161.102");
+    expect(targets[0]?.hostname).toBe("s.rkns.top");
+  });
+});
+
+describe("CNAME health mapped onto service IPs", () => {
+  it("getView copies CNAME-keyed health onto the service IP row", async () => {
+    const { createMemoryDb, repos, runMigrations } = await import("@cfdm/db");
+    const { getView } = await import("../src/services/service-config-service.js");
+    const { db, sqlite } = createMemoryDb();
+    runMigrations(sqlite);
+
+    const domain = repos.createDomain(db, null, "rkns.top", "zone-id");
+    const service = repos.createService(db, "RW Sub", "rw-sub");
+    repos.replaceServiceIps(db, service.id, ["2.59.161.102"]);
+    const binding = repos.insertBinding(db, domain.id, service.id, "s", null);
+    repos.setBindingCnameTarget(db, binding.id, "ihome.rkns.top");
+    repos.upsertIpHealthStatus(
+      db,
+      "binding",
+      binding.id,
+      "ihome.rkns.top",
+      "up",
+      12,
+      0,
+      null,
+    );
+
+    const view = await getView(db, service.id);
+    expect(view.health_status).toBe("up");
+    expect(view.ip_health).toEqual([
+      expect.objectContaining({
+        ip: "2.59.161.102",
+        status: "up",
+        latency_ms: 12,
+      }),
+    ]);
+  });
 });

packages/db/dist/index.js

@@ -2217,6 +2217,40 @@ function pruneStaleIpHealthStatus(db, activeTargets) {
   }
   return deleted;
 }
+function normalizeCnameHost(target, zoneName) {
+  const trimmed = target.trim().toLowerCase().replace(/\.+$/, "");
+  if (!trimmed) return "";
+  if (trimmed.includes(".")) return trimmed;
+  const zone = zoneName.trim().toLowerCase().replace(/\.+$/, "");
+  return zone ? `${trimmed}.${zone}` : trimmed;
+}
+function resolveCnameProbeIps(db, cnameTarget, zoneName, serviceId) {
+  const fqdn = normalizeCnameHost(cnameTarget, zoneName);
+  if (fqdn) {
+    const fromDns = listOriginIpsForFqdn(db, fqdn).filter(isIpLiteral);
+    if (fromDns.length > 0) return [...new Set(fromDns)];
+  }
+  if (serviceId > 0) {
+    return [...new Set(listServiceIps(db, serviceId).filter(isIpLiteral))];
+  }
+  return [];
+}
+function expandCnameHealthTargets(db, rows) {
+  const expanded = [];
+  for (const row of rows) {
+    const ips = resolveCnameProbeIps(
+      db,
+      row.ip,
+      row.zone_name ?? "",
+      row.service_id ?? 0
+    );
+    const resolved = ips.length > 0 ? ips : [row.ip];
+    for (const ip of resolved) {
+      expanded.push({ ...row, ip });
+    }
+  }
+  return expanded;
+}
 function listHealthCheckTargets(db) {
   const fqdnExpr = sql2`CASE WHEN sb.hostname = '@' OR sb.hostname IS NULL THEN d.zone_name ELSE sb.hostname || '.' || d.zone_name END`;
   const bindingTargets = db.all(sql2`
@@ -2235,7 +2269,7 @@ function listHealthCheckTargets(db) {
     JOIN service_bindings sb ON sb.id = sbi.binding_id
     JOIN domains d ON d.id = sb.domain_id
     WHERE sb.health_check_enabled = 1
-  `);
+  `).filter((t) => isIpLiteral(t.ip));
   const groupTargets = db.all(sql2`
     SELECT DISTINCT 'group' AS scope, sg.id AS ref_id, sbi.ip,
       sg.domain AS hostname,
@@ -2258,7 +2292,7 @@ function listHealthCheckTargets(db) {
       AND s.enabled = 1
       AND sg.enabled = 1
       AND (sb.cname_target IS NULL OR sb.cname_target = '')
-  `);
+  `).filter((t) => isIpLiteral(t.ip));
   const groupInheritedBindingTargets = db.all(sql2`
     SELECT 'binding' AS scope, sb.id AS ref_id, sbi.ip,
       ${fqdnExpr} AS hostname,
@@ -2281,8 +2315,10 @@ function listHealthCheckTargets(db) {
       AND s.enabled = 1
       AND sg.enabled = 1
       AND sb.health_check_enabled = 0
-  `);
+  `).filter((t) => isIpLiteral(t.ip));
-  const cnameBindingTargets = db.all(sql2`
+  const cnameBindingTargets = expandCnameHealthTargets(
+    db,
+    db.all(sql2`
     SELECT 'binding' AS scope, sb.id AS ref_id, sb.cname_target AS ip,
       ${fqdnExpr} AS hostname,
       sb.health_check_type AS type,
@@ -2293,7 +2329,9 @@ function listHealthCheckTargets(db) {
       sb.health_check_verify_tls AS verify_tls,
       sb.health_check_providers AS providers_json,
       COALESCE(sb.health_check_aggregate, 'majority') AS aggregate,
-      COALESCE(sb.health_check_provider, 'local') AS provider
+      COALESCE(sb.health_check_provider, 'local') AS provider,
+      d.zone_name AS zone_name,
+      sb.service_id AS service_id
     FROM service_bindings sb
     JOIN domains d ON d.id = sb.domain_id
     JOIN services s ON s.id = sb.service_id
@@ -2301,8 +2339,11 @@ function listHealthCheckTargets(db) {
       AND sb.cname_target IS NOT NULL
       AND sb.cname_target <> ''
       AND s.enabled = 1
-  `);
+  `)
-  const groupInheritedCnameBindingTargets = db.all(sql2`
+  );
+  const groupInheritedCnameBindingTargets = expandCnameHealthTargets(
+    db,
+    db.all(sql2`
     SELECT 'binding' AS scope, sb.id AS ref_id, sb.cname_target AS ip,
       ${fqdnExpr} AS hostname,
       sg.health_check_type AS type,
@@ -2313,7 +2354,9 @@ function listHealthCheckTargets(db) {
       sg.health_check_verify_tls AS verify_tls,
       sg.health_check_providers AS providers_json,
       COALESCE(sg.health_check_aggregate, 'majority') AS aggregate,
-      COALESCE(sg.health_check_provider, 'local') AS provider
+      COALESCE(sg.health_check_provider, 'local') AS provider,
+      d.zone_name AS zone_name,
+      sb.service_id AS service_id
     FROM service_bindings sb
     JOIN domains d ON d.id = sb.domain_id
     JOIN services s ON s.id = sb.service_id
@@ -2325,7 +2368,8 @@ function listHealthCheckTargets(db) {
       AND sb.health_check_enabled = 0
       AND sb.cname_target IS NOT NULL
       AND sb.cname_target <> ''
-  `);
+  `)
+  );
   return [
     ...bindingTargets,
     ...groupTargets,

packages/db/src/repos.ts

@@ -2493,13 +2493,73 @@ export function pruneStaleIpHealthStatus(
 
 // --- Health Check Targets ---
 
+function normalizeCnameHost(target: string, zoneName: string): string {
+  const trimmed = target.trim().toLowerCase().replace(/\.+$/, "");
+  if (!trimmed) return "";
+  if (trimmed.includes(".")) return trimmed;
+  const zone = zoneName.trim().toLowerCase().replace(/\.+$/, "");
+  return zone ? `${trimmed}.${zone}` : trimmed;
+}
+
+/**
+ * Unwrap a CNAME health target to origin IPs:
+ * 1) A/AAAA from local dns_records (follows CNAME chain)
+ * 2) this service's IP pool
+ *
+ * Hostname is kept as fallback so probes still run when nothing resolves.
+ */
+function resolveCnameProbeIps(
+  db: Db,
+  cnameTarget: string,
+  zoneName: string,
+  serviceId: number,
+): string[] {
+  const fqdn = normalizeCnameHost(cnameTarget, zoneName);
+  if (fqdn) {
+    const fromDns = listOriginIpsForFqdn(db, fqdn).filter(isIpLiteral);
+    if (fromDns.length > 0) return [...new Set(fromDns)];
+  }
+  if (serviceId > 0) {
+    return [...new Set(listServiceIps(db, serviceId).filter(isIpLiteral))];
+  }
+  return [];
+}
+
+type RawHealthTarget = HealthCheckTarget & {
+  providers_json?: string | null;
+  aggregate?: string | null;
+  zone_name?: string | null;
+  service_id?: number | null;
+};
+
+function expandCnameHealthTargets(
+  db: Db,
+  rows: RawHealthTarget[],
+): RawHealthTarget[] {
+  const expanded: RawHealthTarget[] = [];
+  for (const row of rows) {
+    const ips = resolveCnameProbeIps(
+      db,
+      row.ip,
+      row.zone_name ?? "",
+      row.service_id ?? 0,
+    );
+    const resolved = ips.length > 0 ? ips : [row.ip];
+    for (const ip of resolved) {
+      expanded.push({ ...row, ip });
+    }
+  }
+  return expanded;
+}
+
 export function listHealthCheckTargets(db: Db): HealthCheckTarget[] {
   // FQDN for a binding: "@" => zone_name, else "<hostname>.<zone_name>".
   // Used as SNI / Host header for HTTP(S) probes — the raw `sb.hostname` is just the record name
   // (e.g. "de" or "@"), which would break TLS SNI (ssl alert 112 "unrecognized name").
   const fqdnExpr = sql`CASE WHEN sb.hostname = '@' OR sb.hostname IS NULL THEN d.zone_name ELSE sb.hostname || '.' || d.zone_name END`;
 
-  const bindingTargets = db.all<HealthCheckTarget>(sql`
+  const bindingTargets = db
+    .all<RawHealthTarget>(sql`
     SELECT 'binding' AS scope, sb.id AS ref_id, sbi.ip,
       ${fqdnExpr} AS hostname,
       sb.health_check_type AS type,
@@ -2515,12 +2575,14 @@ export function listHealthCheckTargets(db: Db): HealthCheckTarget[] {
     JOIN service_bindings sb ON sb.id = sbi.binding_id
     JOIN domains d ON d.id = sb.domain_id
     WHERE sb.health_check_enabled = 1
-  `);
+  `)
+    .filter((t) => isIpLiteral(t.ip));
 
   // Group FQDN probes the same A-record IPs that DNS publishes (binding IPs of
   // enabled services) — NOT the full service_ips pool. Pool-only dead IPs must
   // not mark the group Down while the published domain stays healthy.
-  const groupTargets = db.all<HealthCheckTarget>(sql`
+  const groupTargets = db
+    .all<RawHealthTarget>(sql`
     SELECT DISTINCT 'group' AS scope, sg.id AS ref_id, sbi.ip,
       sg.domain AS hostname,
       sg.health_check_type AS type,
@@ -2542,13 +2604,14 @@ export function listHealthCheckTargets(db: Db): HealthCheckTarget[] {
       AND s.enabled = 1
       AND sg.enabled = 1
       AND (sb.cname_target IS NULL OR sb.cname_target = '')
-  `);
+  `)
+    .filter((t) => isIpLiteral(t.ip));
 
   // IPs of A-bindings of enabled services in a group whose group has health-check enabled.
   // These inherit the group's health-check config (scope='binding', ref_id=binding_id),
   // so per-domain badges reflect group rules. Skipped for bindings that already have
   // their own health_check_enabled=1 (covered by bindingTargets above).
-  const groupInheritedBindingTargets = db.all<HealthCheckTarget>(sql`
+  const groupInheritedBindingTargets = db.all<RawHealthTarget>(sql`
     SELECT 'binding' AS scope, sb.id AS ref_id, sbi.ip,
       ${fqdnExpr} AS hostname,
       sg.health_check_type AS type,
@@ -2570,10 +2633,13 @@ export function listHealthCheckTargets(db: Db): HealthCheckTarget[] {
       AND s.enabled = 1
       AND sg.enabled = 1
       AND sb.health_check_enabled = 0
-  `);
+  `)
+    .filter((t) => isIpLiteral(t.ip));
 
-  // CNAME-bindings with their own health_check_enabled: probe the CNAME target host.
+  // CNAME-bindings: unwrap to origin/service IPs so ip_health keys match the IP table.
-  const cnameBindingTargets = db.all<HealthCheckTarget>(sql`
+  const cnameBindingTargets = expandCnameHealthTargets(
+    db,
+    db.all<RawHealthTarget>(sql`
     SELECT 'binding' AS scope, sb.id AS ref_id, sb.cname_target AS ip,
       ${fqdnExpr} AS hostname,
       sb.health_check_type AS type,
@@ -2584,7 +2650,9 @@ export function listHealthCheckTargets(db: Db): HealthCheckTarget[] {
       sb.health_check_verify_tls AS verify_tls,
       sb.health_check_providers AS providers_json,
       COALESCE(sb.health_check_aggregate, 'majority') AS aggregate,
-      COALESCE(sb.health_check_provider, 'local') AS provider
+      COALESCE(sb.health_check_provider, 'local') AS provider,
+      d.zone_name AS zone_name,
+      sb.service_id AS service_id
     FROM service_bindings sb
     JOIN domains d ON d.id = sb.domain_id
     JOIN services s ON s.id = sb.service_id
@@ -2592,11 +2660,14 @@ export function listHealthCheckTargets(db: Db): HealthCheckTarget[] {
       AND sb.cname_target IS NOT NULL
       AND sb.cname_target <> ''
       AND s.enabled = 1
-  `);
+  `),
+  );
 
   // CNAME-bindings of services in a group with group health-check enabled
   // (inherit group config). Only for bindings without their own health_check_enabled.
-  const groupInheritedCnameBindingTargets = db.all<HealthCheckTarget>(sql`
+  const groupInheritedCnameBindingTargets = expandCnameHealthTargets(
+    db,
+    db.all<RawHealthTarget>(sql`
     SELECT 'binding' AS scope, sb.id AS ref_id, sb.cname_target AS ip,
       ${fqdnExpr} AS hostname,
       sg.health_check_type AS type,
@@ -2607,7 +2678,9 @@ export function listHealthCheckTargets(db: Db): HealthCheckTarget[] {
       sg.health_check_verify_tls AS verify_tls,
       sg.health_check_providers AS providers_json,
       COALESCE(sg.health_check_aggregate, 'majority') AS aggregate,
-      COALESCE(sg.health_check_provider, 'local') AS provider
+      COALESCE(sg.health_check_provider, 'local') AS provider,
+      d.zone_name AS zone_name,
+      sb.service_id AS service_id
     FROM service_bindings sb
     JOIN domains d ON d.id = sb.domain_id
     JOIN services s ON s.id = sb.service_id
@@ -2619,7 +2692,8 @@ export function listHealthCheckTargets(db: Db): HealthCheckTarget[] {
       AND sb.health_check_enabled = 0
       AND sb.cname_target IS NOT NULL
       AND sb.cname_target <> ''
-  `);
+  `),
+  );
 
   return [
     ...bindingTargets,