Reduce scheduler polling noise

server/cmd/scheduler/main.go

@@ -96,6 +96,7 @@ func main() {
 		WithCache(app.Cache)
 	monitoringRetentionWorker := internalscheduler.NewMonitoringRetentionWorker(app.MonitoringDB, app.Logger)
 	aiPointUsageCleanupWorker := internalscheduler.NewAIPointUsageCleanupWorker(app.DB, app.Logger)
+	schedulerRunRetentionWorker := internalscheduler.NewSchedulerRunRetentionWorker(app.DB, app.Logger)
 
 	controlPlane := internalscheduler.NewControlPlane(app.DB, app.Logger, []internalscheduler.ControlledJob{
 		{
@@ -152,6 +153,12 @@ func main() {
 				return aiPointUsageCleanupWorker.RunOnce(runCtx, jobCtx)
 			},
 		},
+		{
+			Key: "scheduler_run_retention",
+			Run: func(runCtx context.Context, jobCtx internalscheduler.JobRunContext) (map[string]any, error) {
+				return schedulerRunRetentionWorker.RunOnce(runCtx, jobCtx)
+			},
+		},
 	})
 	startSchedulerWorker(ctx, &workerWG, "control_plane", app.Logger.Sugar(), controlPlane.Run)
 

server/internal/ops/repository/scheduler.go

@@ -67,6 +67,20 @@ type SchedulerRunFinish struct {
 	ErrorMessage *string
 }
 
+type SchedulerRunRecord struct {
+	JobKey              string
+	TriggerID           *int64
+	SchedulerInstanceID string
+	TriggerType         string
+	ConfigVersion       int
+	ConfigSnapshot      map[string]any
+	Status              string
+	Stats               map[string]any
+	ErrorMessage        *string
+	StartedAt           time.Time
+	FinishedAt          time.Time
+}
+
 type SchedulerTriggerCreate struct {
 	JobKey         string
 	TriggerType    string
@@ -332,6 +346,35 @@ func (r *SchedulerRepository) FinishRun(ctx context.Context, in SchedulerRunFini
 	`, in.RunID, in.Status, statsJSON, in.ErrorMessage))
 }
 
+func (r *SchedulerRepository) RecordRun(ctx context.Context, in SchedulerRunRecord) (*domain.SchedulerRun, error) {
+	configJSON, err := marshalSchedulerJSONMap(in.ConfigSnapshot)
+	if err != nil {
+		return nil, err
+	}
+	statsJSON, err := marshalSchedulerJSONMap(in.Stats)
+	if err != nil {
+		return nil, err
+	}
+	duration := in.FinishedAt.Sub(in.StartedAt).Milliseconds()
+	if duration < 0 {
+		duration = 0
+	}
+	return scanSchedulerRun(r.pool.QueryRow(ctx, `
+		INSERT INTO ops.scheduler_job_runs (
+			job_key, trigger_id, scheduler_instance_id, trigger_type,
+			status, config_version, config_snapshot, stats, error_message,
+			started_at, finished_at, duration_ms
+		)
+		VALUES ($1, $2, $3, $4, $5, $6, COALESCE($7::jsonb, '{}'::jsonb),
+			COALESCE($8::jsonb, '{}'::jsonb), $9, $10, $11, $12)
+		RETURNING id, job_key, trigger_id, scheduler_instance_id, trigger_type, status,
+			config_version, config_snapshot, stats, error_message, started_at,
+			finished_at, duration_ms
+	`, in.JobKey, in.TriggerID, nullableString(in.SchedulerInstanceID), in.TriggerType,
+		in.Status, in.ConfigVersion, configJSON, statsJSON, in.ErrorMessage,
+		in.StartedAt, in.FinishedAt, duration))
+}
+
 func (r *SchedulerRepository) ListRuns(ctx context.Context, f SchedulerRunFilter) ([]domain.SchedulerRun, int64, error) {
 	args := []any{}
 	where := "1=1"

server/internal/scheduler/control_plane.go

@@ -5,6 +5,7 @@ import (
 	"errors"
 	"fmt"
 	"hash/fnv"
+	"math"
 	"os"
 	"strconv"
 	"strings"
@@ -281,6 +282,12 @@ func (p *ControlPlane) runJobOnce(parent context.Context, job ControlledJob, tri
 	if dryRun {
 		mergedConfig["dry_run"] = true
 	}
+	// Empty-success suppression is only for high-frequency automatic polling jobs.
+	// Suppressed runs intentionally do not create running rows and do not advance
+	// LastAutoRunStartedAt, so process restarts may run them immediately once as
+	// overdue. Manual, dry-run, failed, and non-empty executions still get full
+	// audit rows below.
+	suppressEmptySuccessRun := shouldSuppressEmptySuccessRun(triggerType, dryRun, mergedConfig)
 
 	lockKey := advisoryLockKey(job.Key)
 	lockConn, locked := p.tryAdvisoryLock(parent, lockKey)
@@ -295,13 +302,17 @@ func (p *ControlPlane) runJobOnce(parent context.Context, job ControlledJob, tri
 	}
 	defer p.advisoryUnlock(context.Background(), lockConn, lockKey)
 
-	run, err := p.repo.StartRun(parent, opsrepo.SchedulerRunStart{
+	configSnapshot := schedulerConfigSnapshot(cfg, mergedConfig)
+	startedAt := time.Now()
+	var run *opsdomain.SchedulerRun
+	if !suppressEmptySuccessRun {
+		run, err = p.repo.StartRun(parent, opsrepo.SchedulerRunStart{
 			JobKey:              job.Key,
 			TriggerID:           triggerID,
 			SchedulerInstanceID: p.instanceID,
 			TriggerType:         triggerType,
 			ConfigVersion:       cfg.Version,
-		ConfigSnapshot:      schedulerConfigSnapshot(cfg, mergedConfig),
+			ConfigSnapshot:      configSnapshot,
 		})
 		if err != nil {
 			if p.logger != nil {
@@ -309,6 +320,8 @@ func (p *ControlPlane) runJobOnce(parent context.Context, job ControlledJob, tri
 			}
 			return
 		}
+		startedAt = run.StartedAt
+	}
 
 	timeout := time.Duration(cfg.TimeoutSeconds) * time.Second
 	if timeout <= 0 {
@@ -335,25 +348,61 @@ func (p *ControlPlane) runJobOnce(parent context.Context, job ControlledJob, tri
 		stats = map[string]any{}
 	}
 	stats["dry_run"] = dryRun
+	finishedAt := time.Now()
+	if suppressEmptySuccessRun && runErr == nil && isEmptySuccessRun(stats, mergedConfig) {
+		if p.logger != nil {
+			p.logger.Info("scheduler empty success run suppressed", zap.String("job_key", job.Key))
+		}
+		return
+	}
 
 	finishCtx, finishCancel := context.WithTimeout(context.Background(), schedulerFinishTimeout)
 	defer finishCancel()
-	finished, finishErr := p.repo.FinishRun(finishCtx, opsrepo.SchedulerRunFinish{
+	var finished *opsdomain.SchedulerRun
+	var finishErr error
+	if run == nil {
+		finished, finishErr = p.repo.RecordRun(finishCtx, opsrepo.SchedulerRunRecord{
+			JobKey:              job.Key,
+			TriggerID:           triggerID,
+			SchedulerInstanceID: p.instanceID,
+			TriggerType:         triggerType,
+			ConfigVersion:       cfg.Version,
+			ConfigSnapshot:      configSnapshot,
+			Status:              status,
+			Stats:               stats,
+			ErrorMessage:        errMsg,
+			StartedAt:           startedAt,
+			FinishedAt:          finishedAt,
+		})
+	} else {
+		finished, finishErr = p.repo.FinishRun(finishCtx, opsrepo.SchedulerRunFinish{
 			RunID:        run.ID,
 			Status:       status,
 			Stats:        stats,
 			ErrorMessage: errMsg,
 		})
+	}
 	if finishErr != nil && p.logger != nil {
-		p.logger.Warn("scheduler run finish failed", zap.String("job_key", job.Key), zap.Int64("run_id", run.ID), zap.Error(finishErr))
+		fields := []zap.Field{zap.String("job_key", job.Key), zap.Error(finishErr)}
+		if run != nil {
+			fields = append(fields, zap.Int64("run_id", run.ID))
+		}
+		if finished != nil {
+			fields = append(fields, zap.Int64("finished_run_id", finished.ID))
+		}
+		p.logger.Warn("scheduler run finish failed", fields...)
 	}
 	if triggerID != nil {
 		triggerStatus := status
 		if finishErr != nil {
 			triggerStatus = opsdomain.SchedulerRunFailed
 		}
-		runID := run.ID
-		_ = p.repo.FinishTrigger(finishCtx, *triggerID, &runID, triggerStatus, errMsg)
+		var runID *int64
+		if finished != nil {
+			value := finished.ID
+			runID = &value
+		}
+		_ = p.repo.FinishTrigger(finishCtx, *triggerID, runID, triggerStatus, errMsg)
 	}
 	if runErr != nil {
 		if p.logger != nil {
@@ -477,6 +526,101 @@ func boolFromMap(in map[string]any, key string) bool {
 	}
 }
 
+func shouldSuppressEmptySuccessRun(triggerType string, dryRun bool, config map[string]any) bool {
+	if dryRun || triggerType != opsdomain.SchedulerTriggerAuto {
+		return false
+	}
+	return boolFromMap(config, "suppress_empty_success_runs")
+}
+
+func isEmptySuccessRun(stats map[string]any, config map[string]any) bool {
+	if stats == nil || !boolFromMap(config, "suppress_empty_success_runs") {
+		return false
+	}
+	keys := emptySuccessMetricKeys(config)
+	if len(keys) == 0 {
+		return false
+	}
+	for _, key := range keys {
+		value, ok := stats[key]
+		if !ok {
+			return false
+		}
+		if numericValue(value) != 0 {
+			return false
+		}
+	}
+	return true
+}
+
+func emptySuccessMetricKeys(config map[string]any) []string {
+	value, ok := config["empty_success_metric_keys"]
+	if !ok || value == nil {
+		return nil
+	}
+	switch typed := value.(type) {
+	case []string:
+		return typed
+	case []any:
+		out := make([]string, 0, len(typed))
+		for _, item := range typed {
+			if key := strings.TrimSpace(fmt.Sprint(item)); key != "" {
+				out = append(out, key)
+			}
+		}
+		return out
+	case string:
+		parts := strings.Split(typed, ",")
+		out := make([]string, 0, len(parts))
+		for _, part := range parts {
+			if key := strings.TrimSpace(part); key != "" {
+				out = append(out, key)
+			}
+		}
+		return out
+	default:
+		return nil
+	}
+}
+
+func numericValue(value any) float64 {
+	switch typed := value.(type) {
+	case int:
+		return float64(typed)
+	case int8:
+		return float64(typed)
+	case int16:
+		return float64(typed)
+	case int32:
+		return float64(typed)
+	case int64:
+		return float64(typed)
+	case uint:
+		return float64(typed)
+	case uint8:
+		return float64(typed)
+	case uint16:
+		return float64(typed)
+	case uint32:
+		return float64(typed)
+	case uint64:
+		return float64(typed)
+	case float32:
+		return float64(typed)
+	case float64:
+		if math.IsNaN(typed) {
+			return 0
+		}
+		return typed
+	case string:
+		parsed, err := strconv.ParseFloat(strings.TrimSpace(typed), 64)
+		if err == nil {
+			return parsed
+		}
+	}
+	return 0
+}
+
 func parseRunAtLocal(config map[string]any) string {
 	value, ok := config["run_at_local"]
 	if !ok || value == nil {

server/internal/scheduler/control_plane_test.go

@@ -45,3 +45,47 @@ func TestNextIntervalAutoAtSchedulesAfterCurrentCompletion(t *testing.T) {
 		t.Fatalf("nextIntervalAutoAt() = %s, want %s", got, want)
 	}
 }
+
+func TestIsEmptySuccessRunRequiresConfiguredZeroMetrics(t *testing.T) {
+	stats := map[string]any{
+		"hydrated_count": 0,
+		"claimed_count":  0,
+	}
+	config := map[string]any{
+		"suppress_empty_success_runs": true,
+		"empty_success_metric_keys":   []any{"hydrated_count", "claimed_count"},
+	}
+
+	if !isEmptySuccessRun(stats, config) {
+		t.Fatal("expected empty success run to be suppressed")
+	}
+}
+
+func TestIsEmptySuccessRunKeepsRunsWithOutput(t *testing.T) {
+	stats := map[string]any{
+		"recoverable_count": 1,
+		"republished_count": 0,
+	}
+	config := map[string]any{
+		"suppress_empty_success_runs": true,
+		"empty_success_metric_keys":   []any{"recoverable_count", "republished_count"},
+	}
+
+	if isEmptySuccessRun(stats, config) {
+		t.Fatal("expected non-empty success run to be retained")
+	}
+}
+
+func TestShouldSuppressEmptySuccessRunOnlyForAutoNonDryRun(t *testing.T) {
+	config := map[string]any{"suppress_empty_success_runs": true}
+
+	if !shouldSuppressEmptySuccessRun("auto", false, config) {
+		t.Fatal("expected auto non-dry-run to allow suppression")
+	}
+	if shouldSuppressEmptySuccessRun("manual", false, config) {
+		t.Fatal("manual runs must not be suppressed")
+	}
+	if shouldSuppressEmptySuccessRun("auto", true, config) {
+		t.Fatal("dry runs must not be suppressed")
+	}
+}

server/internal/scheduler/scheduler_run_retention_worker.go

@@ -0,0 +1,248 @@
+package scheduler
+
+import (
+	"context"
+	"fmt"
+	"strings"
+	"time"
+
+	"github.com/jackc/pgx/v5/pgxpool"
+	"go.uber.org/zap"
+)
+
+const (
+	defaultSchedulerRunRetentionDays    = 15
+	defaultSchedulerRunRetentionBatch   = 5000
+	defaultSchedulerRunRetentionBatches = 100
+)
+
+type SchedulerRunRetentionWorker struct {
+	pool   *pgxpool.Pool
+	logger *zap.Logger
+}
+
+type schedulerRunRetentionTarget struct {
+	Name      string
+	DeleteSQL string
+	ProbeSQL  string
+}
+
+func NewSchedulerRunRetentionWorker(pool *pgxpool.Pool, logger *zap.Logger) *SchedulerRunRetentionWorker {
+	return &SchedulerRunRetentionWorker{pool: pool, logger: logger}
+}
+
+func (w *SchedulerRunRetentionWorker) RunOnce(ctx context.Context, run JobRunContext) (map[string]any, error) {
+	if w == nil || w.pool == nil {
+		return map[string]any{"skipped": true, "reason": "worker_not_ready"}, nil
+	}
+
+	startedAt := time.Now()
+	requestedRetentionDays := AsInt(run.Config, "retention_days", defaultSchedulerRunRetentionDays)
+	retentionDays := requestedRetentionDays
+	retentionDaysClamped := false
+	if requestedRetentionDays < defaultSchedulerRunRetentionDays {
+		retentionDays = defaultSchedulerRunRetentionDays
+		retentionDaysClamped = true
+		if w.logger != nil {
+			w.logger.Warn("scheduler run retention_days below hard floor",
+				zap.Int("requested_retention_days", requestedRetentionDays),
+				zap.Int("effective_retention_days", retentionDays),
+			)
+		}
+	}
+	batchSize := AsInt(run.Config, "batch_size", defaultSchedulerRunRetentionBatch)
+	if run.Job != nil && run.Job.BatchSize != nil {
+		batchSize = *run.Job.BatchSize
+	}
+	if batchSize <= 0 {
+		batchSize = defaultSchedulerRunRetentionBatch
+	}
+	maxBatches := AsInt(run.Config, "max_batches_per_run", defaultSchedulerRunRetentionBatches)
+	if maxBatches <= 0 {
+		maxBatches = defaultSchedulerRunRetentionBatches
+	}
+	statementTimeout := AsDuration(run.Config, "statement_timeout", 5*time.Second)
+	cutoff := schedulerRunRetentionCutoff(time.Now(), retentionDays)
+	dryRun := run.DryRun
+
+	stats := map[string]any{
+		"retention_days":           retentionDays,
+		"requested_retention_days": requestedRetentionDays,
+		"retention_days_clamped":   retentionDaysClamped,
+		"cutoff_at":                cutoff,
+		"batch_size":               batchSize,
+		"max_batches_per_run":      maxBatches,
+		"dry_run":                  dryRun,
+	}
+
+	conn, err := w.pool.Acquire(ctx)
+	if err != nil {
+		return stats, fmt.Errorf("acquire ops connection: %w", err)
+	}
+	defer conn.Release()
+
+	if statementTimeout > 0 {
+		if _, err := conn.Exec(ctx, `SELECT set_config('statement_timeout', $1, false)`, fmt.Sprintf("%dms", statementTimeout.Milliseconds())); err != nil {
+			return stats, fmt.Errorf("set statement timeout: %w", err)
+		}
+		defer conn.Exec(context.Background(), `RESET statement_timeout`)
+	}
+
+	totalDeleted := int64(0)
+	totalCandidateProbe := int64(0)
+	perTable := map[string]any{}
+	for _, target := range schedulerRunRetentionTargets() {
+		tableStats := map[string]any{}
+		if dryRun {
+			candidateProbe, err := w.probeTarget(ctx, conn, target, cutoff, batchSize)
+			if err != nil {
+				return stats, fmt.Errorf("probe %s: %w", target.Name, err)
+			}
+			tableStats["candidate_probe_count"] = candidateProbe
+			tableStats["candidate_probe_limited"] = candidateProbe >= int64(batchSize)
+			totalCandidateProbe += candidateProbe
+		} else {
+			deleted, batches, err := w.deleteTarget(ctx, conn, target, cutoff, batchSize, maxBatches)
+			if err != nil {
+				return stats, fmt.Errorf("delete %s: %w", target.Name, err)
+			}
+			tableStats["deleted_count"] = deleted
+			tableStats["batch_count"] = batches
+			totalDeleted += deleted
+		}
+		perTable[target.Name] = tableStats
+	}
+
+	if dryRun {
+		stats["candidate_probe_count"] = totalCandidateProbe
+	}
+	stats["deleted_count"] = totalDeleted
+	stats["tables"] = perTable
+	stats["duration_ms"] = time.Since(startedAt).Milliseconds()
+	return stats, nil
+}
+
+func schedulerRunRetentionCutoff(now time.Time, retentionDays int) time.Time {
+	if retentionDays < defaultSchedulerRunRetentionDays {
+		retentionDays = defaultSchedulerRunRetentionDays
+	}
+	loc, err := time.LoadLocation("Asia/Shanghai")
+	if err != nil {
+		loc = time.FixedZone("UTC+8", 8*60*60)
+	}
+	localNow := now.In(loc)
+	today := time.Date(localNow.Year(), localNow.Month(), localNow.Day(), 0, 0, 0, 0, loc)
+	return today.AddDate(0, 0, -(retentionDays - 1)).UTC()
+}
+
+func (w *SchedulerRunRetentionWorker) probeTarget(ctx context.Context, conn *pgxpool.Conn, target schedulerRunRetentionTarget, cutoff time.Time, limit int) (int64, error) {
+	var count int64
+	if limit <= 0 {
+		limit = defaultSchedulerRunRetentionBatch
+	}
+	if err := conn.QueryRow(ctx, target.ProbeSQL, cutoff.UTC(), limit).Scan(&count); err != nil {
+		return 0, err
+	}
+	return count, nil
+}
+
+func (w *SchedulerRunRetentionWorker) deleteTarget(ctx context.Context, conn *pgxpool.Conn, target schedulerRunRetentionTarget, cutoff time.Time, batchSize, maxBatches int) (int64, int, error) {
+	total := int64(0)
+	batches := 0
+	for batches < maxBatches {
+		tag, err := conn.Exec(ctx, target.DeleteSQL, cutoff.UTC(), batchSize)
+		if err != nil {
+			return total, batches, err
+		}
+		affected := tag.RowsAffected()
+		if affected == 0 {
+			break
+		}
+		total += affected
+		batches++
+		if affected < int64(batchSize) {
+			break
+		}
+	}
+	return total, batches, nil
+}
+
+func schedulerRunRetentionTargets() []schedulerRunRetentionTarget {
+	return []schedulerRunRetentionTarget{
+		{
+			Name:      "scheduler_job_triggers",
+			ProbeSQL:  schedulerRunRetentionTriggerProbeSQL(),
+			DeleteSQL: schedulerRunRetentionTriggerDeleteSQL(),
+		},
+		{
+			Name:      "scheduler_job_runs",
+			ProbeSQL:  schedulerRunRetentionRunProbeSQL(),
+			DeleteSQL: schedulerRunRetentionRunDeleteSQL(),
+		},
+	}
+}
+
+func schedulerRunRetentionTriggerProbeSQL() string {
+	return schedulerRunRetentionProbeSQL(
+		"ops.scheduler_job_triggers",
+		"id",
+		"status IN ('success', 'failed', 'skipped') AND requested_at < $1",
+		"requested_at ASC, id ASC",
+	)
+}
+
+func schedulerRunRetentionTriggerDeleteSQL() string {
+	return schedulerRunRetentionDeleteSQL(
+		"ops.scheduler_job_triggers",
+		"id",
+		"status IN ('success', 'failed', 'skipped') AND requested_at < $1",
+		"requested_at ASC, id ASC",
+	)
+}
+
+func schedulerRunRetentionRunProbeSQL() string {
+	return schedulerRunRetentionProbeSQL(
+		"ops.scheduler_job_runs r",
+		"r.id",
+		"r.status IN ('success', 'failed', 'skipped') AND r.started_at < $1 AND NOT EXISTS (SELECT 1 FROM ops.scheduler_job_triggers t WHERE t.run_id = r.id)",
+		"r.started_at ASC, r.id ASC",
+	)
+}
+
+func schedulerRunRetentionRunDeleteSQL() string {
+	return schedulerRunRetentionDeleteSQL(
+		"ops.scheduler_job_runs r",
+		"r.id",
+		"r.status IN ('success', 'failed', 'skipped') AND r.started_at < $1 AND NOT EXISTS (SELECT 1 FROM ops.scheduler_job_triggers t WHERE t.run_id = r.id)",
+		"r.started_at ASC, r.id ASC",
+	)
+}
+
+func schedulerRunRetentionProbeSQL(table, idColumn, predicate, orderBy string) string {
+	return strings.Join([]string{
+		"SELECT COUNT(*)::bigint",
+		"FROM (",
+		"SELECT " + idColumn,
+		"FROM " + table,
+		"WHERE " + predicate,
+		"ORDER BY " + orderBy,
+		"LIMIT $2",
+		") candidates",
+	}, "\n")
+}
+
+func schedulerRunRetentionDeleteSQL(table, idColumn, predicate, orderBy string) string {
+	return strings.Join([]string{
+		"WITH doomed AS (",
+		"SELECT " + idColumn,
+		"FROM " + table,
+		"WHERE " + predicate,
+		"ORDER BY " + orderBy,
+		"LIMIT $2",
+		"FOR UPDATE SKIP LOCKED",
+		")",
+		"DELETE FROM " + strings.Fields(table)[0] + " target",
+		"USING doomed",
+		"WHERE target.id = doomed.id",
+	}, "\n")
+}

server/internal/scheduler/scheduler_run_retention_worker_test.go

@@ -0,0 +1,53 @@
+package scheduler
+
+import (
+	"strings"
+	"testing"
+	"time"
+)
+
+func TestSchedulerRunRetentionCutoffKeepsFifteenDays(t *testing.T) {
+	now := time.Date(2026, 5, 20, 22, 0, 0, 0, time.FixedZone("UTC+8", 8*60*60))
+
+	got := schedulerRunRetentionCutoff(now, defaultSchedulerRunRetentionDays)
+	want := time.Date(2026, 5, 6, 0, 0, 0, 0, time.FixedZone("UTC+8", 8*60*60)).UTC()
+	if !got.Equal(want) {
+		t.Fatalf("cutoff = %s, want %s", got.Format(time.RFC3339), want.Format(time.RFC3339))
+	}
+}
+
+func TestSchedulerRunRetentionCutoffHasHardFloor(t *testing.T) {
+	now := time.Date(2026, 5, 20, 22, 0, 0, 0, time.FixedZone("UTC+8", 8*60*60))
+
+	got := schedulerRunRetentionCutoff(now, 7)
+	want := schedulerRunRetentionCutoff(now, defaultSchedulerRunRetentionDays)
+	if !got.Equal(want) {
+		t.Fatalf("cutoff = %s, want hard floor cutoff %s", got.Format(time.RFC3339), want.Format(time.RFC3339))
+	}
+}
+
+func TestSchedulerRunRetentionSQLIsBoundedAndFKSafe(t *testing.T) {
+	triggerDeleteSQL := schedulerRunRetentionTriggerDeleteSQL()
+	for _, want := range []string{
+		"status IN ('success', 'failed', 'skipped')",
+		"LIMIT $2",
+		"FOR UPDATE SKIP LOCKED",
+		"DELETE FROM ops.scheduler_job_triggers target",
+	} {
+		if !strings.Contains(triggerDeleteSQL, want) {
+			t.Fatalf("trigger retention SQL missing %q in SQL:\n%s", want, triggerDeleteSQL)
+		}
+	}
+
+	runDeleteSQL := schedulerRunRetentionRunDeleteSQL()
+	for _, want := range []string{
+		"NOT EXISTS (SELECT 1 FROM ops.scheduler_job_triggers t WHERE t.run_id = r.id)",
+		"LIMIT $2",
+		"FOR UPDATE SKIP LOCKED",
+		"DELETE FROM ops.scheduler_job_runs target",
+	} {
+		if !strings.Contains(runDeleteSQL, want) {
+			t.Fatalf("run retention SQL missing %q in SQL:\n%s", want, runDeleteSQL)
+		}
+	}
+}

server/migrations_ops/20260520151000_add_scheduler_run_retention_job.down.sql

@@ -0,0 +1,11 @@
+DELETE FROM ops.scheduler_job_triggers
+WHERE job_key = 'scheduler_run_retention';
+
+DELETE FROM ops.scheduler_job_runs
+WHERE job_key = 'scheduler_run_retention';
+
+DELETE FROM ops.scheduler_jobs
+WHERE job_key = 'scheduler_run_retention';
+
+DROP INDEX IF EXISTS ops.idx_scheduler_job_triggers_finished_requested;
+DROP INDEX IF EXISTS ops.idx_scheduler_job_triggers_run_id;

server/migrations_ops/20260520151000_add_scheduler_run_retention_job.up.sql

@@ -0,0 +1,19 @@
+CREATE INDEX IF NOT EXISTS idx_scheduler_job_triggers_run_id
+    ON ops.scheduler_job_triggers (run_id)
+    WHERE run_id IS NOT NULL;
+
+CREATE INDEX IF NOT EXISTS idx_scheduler_job_triggers_finished_requested
+    ON ops.scheduler_job_triggers (requested_at ASC, id ASC)
+    WHERE status IN ('success', 'failed', 'skipped');
+
+INSERT INTO ops.scheduler_jobs
+    (job_key, display_name, category, description, enabled, schedule_type, interval_seconds, timezone, timeout_seconds, batch_size, config)
+VALUES
+    ('scheduler_run_retention', '调度运行记录留存清理', 'ops', '限批删除 15 天以前已结束的调度触发与运行记录，控制高频 scheduler 审计表体积。', true, 'interval', 86400, 'Asia/Shanghai', 300, 5000,
+     '{"retention_days":15,"batch_size":5000,"max_batches_per_run":100,"dry_run":false,"statement_timeout":"5s","run_at_local":"03:20"}'::jsonb)
+ON CONFLICT (job_key) DO UPDATE
+SET display_name = EXCLUDED.display_name,
+    category = EXCLUDED.category,
+    description = EXCLUDED.description,
+    config = ops.scheduler_jobs.config || EXCLUDED.config,
+    updated_at = NOW();

server/migrations_ops/20260521100000_tune_scheduler_polling_noise.down.sql

@@ -0,0 +1,10 @@
+UPDATE ops.scheduler_jobs
+SET interval_seconds = 60,
+    config = config - 'claim_ttl' - 'suppress_empty_success_runs' - 'empty_success_metric_keys',
+    updated_at = NOW()
+WHERE job_key = 'monitoring_result_recovery';
+
+UPDATE ops.scheduler_jobs
+SET config = config - 'suppress_empty_success_runs' - 'empty_success_metric_keys',
+    updated_at = NOW()
+WHERE job_key = 'schedule_dispatch';

server/migrations_ops/20260521100000_tune_scheduler_polling_noise.up.sql

@@ -0,0 +1,15 @@
+UPDATE ops.scheduler_jobs
+SET interval_seconds = 120,
+    config = config || '{"claim_ttl":"5m"}'::jsonb,
+    updated_at = NOW()
+WHERE job_key = 'monitoring_result_recovery';
+
+UPDATE ops.scheduler_jobs
+SET config = config || '{"suppress_empty_success_runs":true,"empty_success_metric_keys":["hydrated_count","claimed_count"]}'::jsonb,
+    updated_at = NOW()
+WHERE job_key = 'schedule_dispatch';
+
+UPDATE ops.scheduler_jobs
+SET config = config || '{"suppress_empty_success_runs":true,"empty_success_metric_keys":["recoverable_count","republished_count"]}'::jsonb,
+    updated_at = NOW()
+WHERE job_key = 'monitoring_result_recovery';