fix(scheduler): убрать двойной SaveCheckRun (Checker персистит), SetDrift через CountDriftDomains, resolved после error

internal/scheduler/scheduler.go

@@ -28,14 +28,16 @@ const (
 )
 
 // SchedStore is the narrow store dependency the scheduler needs: due
-// schedules, their domains, and per-domain status bookkeeping.
+// schedules, their domains, and per-domain status bookkeeping. Persisting
+// the check result itself (check_runs) is the Checker's job — see Checker
+// below — not the scheduler's.
 type SchedStore interface {
 	ListDueSchedules(ctx context.Context, now time.Time) ([]store.Schedule, error)
 	TouchScheduleRun(ctx context.Context, projectID uuid.UUID, at time.Time) error
 	ListDomains(ctx context.Context, projectID uuid.UUID) ([]store.Domain, error)
 	GetDomainStatus(ctx context.Context, domainID uuid.UUID) (string, error)
 	SetDomainStatus(ctx context.Context, domainID uuid.UUID, status string) error
-	SaveCheckRun(ctx context.Context, domainID uuid.UUID, cs diff.Changeset) error
+	CountDriftDomains(ctx context.Context) (int, error)
 }
 
 // Checker computes the diff between a domain's desired template and its
@@ -92,8 +94,6 @@ func (s *Scheduler) RunOnce(ctx context.Context, now time.Time) error {
 		return fmt.Errorf("list due schedules: %w", err)
 	}
 
-	driftCount := 0
-
 	for _, sch := range due {
 		domains, err := s.store.ListDomains(ctx, sch.ProjectID)
 		if err != nil {
@@ -102,9 +102,7 @@ func (s *Scheduler) RunOnce(ctx context.Context, now time.Time) error {
 		}
 
 		for _, d := range domains {
-			if s.checkDomain(ctx, sch.ProjectID, d, now) == StatusDrift {
-				driftCount++
-			}
+			s.checkDomain(ctx, sch.ProjectID, d, now)
 		}
 
 		if err := s.store.TouchScheduleRun(ctx, sch.ProjectID, now); err != nil {
@@ -112,7 +110,15 @@ func (s *Scheduler) RunOnce(ctx context.Context, now time.Time) error {
 		}
 	}
 
-	s.metrics.SetDrift(driftCount)
+	// The real, system-wide count of drift domains — not a local
+	// accumulator scoped to this tick's due projects — so the gauge
+	// reflects reality even across ticks where different projects are due.
+	count, err := s.store.CountDriftDomains(ctx)
+	if err != nil {
+		log.Printf("scheduler: count drift domains failed: %v", err)
+	} else {
+		s.metrics.SetDrift(count)
+	}
 	return nil
 }
 
@@ -139,12 +145,10 @@ func (s *Scheduler) checkDomain(ctx context.Context, projectID uuid.UUID, d stor
 		prev = StatusUnknown
 	}
 
-	// A failed Check has no changeset worth recording; a successful one does.
-	if checkErr == nil {
-		if err := s.store.SaveCheckRun(ctx, d.ID, cs); err != nil {
-			log.Printf("scheduler: save check run for %s failed: %v", d.ID, err)
-		}
-	}
+	// Persisting the check_runs row is the Checker's job: DomainService.Check
+	// already calls Recorder.SaveCheckRun internally on every successful
+	// check (drift or in_sync). Calling it again here would double-write
+	// check_runs history for the same check.
 
 	if err := s.store.SetDomainStatus(ctx, d.ID, newStatus); err != nil {
 		log.Printf("scheduler: set domain status for %s failed: %v", d.ID, err)
@@ -170,15 +174,16 @@ func (s *Scheduler) checkDomain(ctx context.Context, projectID uuid.UUID, d stor
 // shouldNotify decides whether a prev -> new status transition is worth
 // alerting on:
 //   - entering drift or error from any other status is always notified;
-//   - recovering from drift back to in_sync ("resolved") is notified;
+//   - recovering from drift OR error back to in_sync ("resolved") is
+//     notified — including recovery after a provider/check failure;
 //   - the initial unknown -> in_sync transition (first successful check of a
-//     domain that never drifted) is NOT notified — it is not news, it is the
-//     expected steady state.
+//     domain that never drifted or errored) is NOT notified — it is not
+//     news, it is the expected steady state.
 func shouldNotify(prev, newStatus string) bool {
 	if (newStatus == StatusDrift || newStatus == StatusError) && newStatus != prev {
 		return true
 	}
-	if prev == StatusDrift && newStatus == StatusInSync {
+	if (prev == StatusDrift || prev == StatusError) && newStatus == StatusInSync {
 		return true
 	}
 	return false