feat(observability): add maintenance policy metrics

Prometheus: runs, duration, rows_deleted, config_changes; инкремент при CRUD и Execute.
Co-authored-by: Cursor <[email protected]>

internal/httpapi/routes_maintenance.go

@@ -6,6 +6,7 @@ import (
 	"strings"
 
 	"evobgp/internal/jobs"
+	"evobgp/internal/observability"
 	"evobgp/internal/store"
 )
 
@@ -107,6 +108,7 @@ func (s *Server) handleCreateMaintenancePolicy(w http.ResponseWriter, r *http.Re
 		return
 	}
 	_ = s.store.AppendMaintenancePolicyConfigAudit(actorPrefix(a), p.ID, "create", nil, maintenancePolicyJSON(p))
+	observability.IncMaintenanceConfigChange("create")
 	s.reloadMaintenanceConfig(r)
 	writeJSON(w, http.StatusCreated, maintenancePolicyJSON(p))
 }
@@ -133,6 +135,7 @@ func (s *Server) handlePatchMaintenancePolicy(w http.ResponseWriter, r *http.Req
 		return
 	}
 	_ = s.store.AppendMaintenancePolicyConfigAudit(actorPrefix(a), id, "update", maintenancePolicyJSON(before), maintenancePolicyJSON(updated))
+	observability.IncMaintenanceConfigChange("update")
 	s.reloadMaintenanceConfig(r)
 	writeJSON(w, http.StatusOK, maintenancePolicyJSON(updated))
 }
@@ -153,6 +156,7 @@ func (s *Server) handleDeleteMaintenancePolicy(w http.ResponseWriter, r *http.Re
 		return
 	}
 	_ = s.store.AppendMaintenancePolicyConfigAudit(actorPrefix(a), id, "delete", maintenancePolicyJSON(before), nil)
+	observability.IncMaintenanceConfigChange("delete")
 	s.reloadMaintenanceConfig(r)
 	w.WriteHeader(http.StatusNoContent)
 }

internal/maintenance/policy_executor.go

@@ -7,6 +7,7 @@ import (
 	"strings"
 	"time"
 
+	"evobgp/internal/observability"
 	"evobgp/internal/pgmonitor"
 	"evobgp/internal/store"
 
@@ -22,19 +23,29 @@ type PolicyExecutor struct {
 
 // Execute runs cleanup and/or vacuum steps for a policy.
 func (e *PolicyExecutor) Execute(ctx context.Context, policy *store.MaintenancePolicy, dryRun bool) (map[string]any, error) {
-	if e == nil || e.Pool == nil {
+	start := time.Now()
-		return nil, fmt.Errorf("maintenance: postgres not configured")
-	}
 	if policy == nil {
 		return nil, store.ErrInvalidInput
 	}
+	action := policyAction(policy)
+	record := func(status string, detail map[string]any) {
+		observability.RecordMaintenancePolicyRun(policy.ID, action, status, dryRun, time.Since(start), rowsDeletedFromDetail(detail))
+	}
+
+	if e == nil || e.Pool == nil {
+		record("failed", nil)
+		return nil, fmt.Errorf("maintenance: postgres not configured")
+	}
 	if err := ValidateTableName(policy.TableName); err != nil {
+		record("failed", nil)
 		return nil, err
 	}
 	if err := ValidateCondition(policy.Condition); err != nil {
+		record("failed", nil)
 		return nil, err
 	}
 	if !store.ValidVacuumStrategy(policy.VacuumStrategy) {
+		record("failed", nil)
 		return nil, store.ErrInvalidInput
 	}
 
@@ -50,6 +61,7 @@ func (e *PolicyExecutor) Execute(ctx context.Context, policy *store.MaintenanceP
 			detail[k] = v
 		}
 		if err != nil {
+			record("failed", detail)
 			return detail, err
 		}
 	}
@@ -61,6 +73,7 @@ func (e *PolicyExecutor) Execute(ctx context.Context, policy *store.MaintenanceP
 			detail["vacuum"] = vacDetail
 		}
 		if err != nil {
+			record("failed", detail)
 			return detail, err
 		}
 	}
@@ -68,10 +81,42 @@ func (e *PolicyExecutor) Execute(ctx context.Context, policy *store.MaintenanceP
 	if !dryRun && e.Store != nil {
 		_ = e.Store.TouchMaintenancePolicyRun(policy.ID, "succeeded", "")
 	}
-
+	record("succeeded", detail)
 	return detail, nil
 }
 
+func policyAction(p *store.MaintenancePolicy) string {
+	if p == nil {
+		return "run"
+	}
+	if p.RetentionPeriodSec != nil || p.MaxRows != nil {
+		if p.VacuumStrategy != store.VacuumStrategyNone {
+			return "cleanup_vacuum"
+		}
+		return "cleanup"
+	}
+	if p.VacuumStrategy != store.VacuumStrategyNone {
+		return p.VacuumStrategy
+	}
+	return "run"
+}
+
+func rowsDeletedFromDetail(detail map[string]any) int64 {
+	if detail == nil {
+		return 0
+	}
+	switch v := detail["deleted"].(type) {
+	case int64:
+		return v
+	case int:
+		return int64(v)
+	case float64:
+		return int64(v)
+	default:
+		return 0
+	}
+}
+
 func (e *PolicyExecutor) runCleanup(ctx context.Context, policy *store.MaintenancePolicy, dryRun bool) (map[string]any, error) {
 	detail := map[string]any{"cleanup": true}
 	limit := NormalizeBatchLimit(policy.MaxRows)

internal/observability/maintenance_metrics.go

@@ -0,0 +1,72 @@
+package observability
+
+import (
+	"strconv"
+	"time"
+
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
+)
+
+var (
+	maintenancePolicyRuns = promauto.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: namespace,
+			Name:      "maintenance_policy_runs_total",
+			Help:      "Maintenance policy executions by outcome.",
+		},
+		[]string{"policy_id", "action", "status", "dry_run"},
+	)
+
+	maintenancePolicyDuration = promauto.NewHistogramVec(
+		prometheus.HistogramOpts{
+			Namespace: namespace,
+			Name:      "maintenance_policy_duration_seconds",
+			Help:      "Maintenance policy execution duration.",
+			Buckets:   prometheus.ExponentialBuckets(0.05, 2, 12),
+		},
+		[]string{"policy_id", "action"},
+	)
+
+	maintenanceRowsDeleted = promauto.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: namespace,
+			Name:      "maintenance_policy_rows_deleted_total",
+			Help:      "Rows deleted by maintenance cleanup policies.",
+		},
+		[]string{"policy_id"},
+	)
+
+	maintenanceConfigChanges = promauto.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: namespace,
+			Name:      "maintenance_config_changes_total",
+			Help:      "Maintenance policy configuration changes from UI/API.",
+		},
+		[]string{"action"},
+	)
+)
+
+// RecordMaintenancePolicyRun updates run counters and histograms.
+func RecordMaintenancePolicyRun(policyID, action, status string, dryRun bool, duration time.Duration, rowsDeleted int64) {
+	if policyID == "" {
+		policyID = "unknown"
+	}
+	if action == "" {
+		action = "run"
+	}
+	dry := strconv.FormatBool(dryRun)
+	maintenancePolicyRuns.WithLabelValues(policyID, action, status, dry).Inc()
+	maintenancePolicyDuration.WithLabelValues(policyID, action).Observe(duration.Seconds())
+	if rowsDeleted > 0 && !dryRun {
+		maintenanceRowsDeleted.WithLabelValues(policyID).Add(float64(rowsDeleted))
+	}
+}
+
+// IncMaintenanceConfigChange increments config audit metric.
+func IncMaintenanceConfigChange(action string) {
+	if action == "" {
+		action = "unknown"
+	}
+	maintenanceConfigChanges.WithLabelValues(action).Inc()
+}