Go: clear stale task cancel flag so first re-parse after cancel runs (#17708)

internal/service/ingestion_task_service.go

@@ -229,7 +229,16 @@ func (s *IngestionTaskService) StartRunning(ctx context.Context, taskID string)
 		}
 		return task, nil
 	case common.STOPPING:
-		return s.transition(ctx, taskID, common.STOPPED)
+		task, err = s.transition(ctx, taskID, common.STOPPED)
+		if err != nil {
+			return nil, err
+		}
+		// The stop is finalized here without a worker (e.g. MQ redelivery of
+		// a task that was nacked before execution), so the Redis cancel flag
+		// that RequestStop set would otherwise leak until TTL and cancel the
+		// next run of this task at the worker's pre-start check.
+		clearCancelFlag(ctx, taskID)
+		return task, nil
 	case common.RUNNING, common.COMPLETED, common.STOPPED, common.FAILED:
 		return task, nil
 	default:
@@ -387,6 +396,11 @@ func (s *IngestionTaskService) CreateAndEnqueue(ctx context.Context, task *entit
 			if err != nil {
 				return nil, err
 			}
+			// The previous run is terminal, so any leftover Redis cancel flag
+			// is stale: a genuine cancel of the new run can only come through
+			// RequestStop once the task is RUNNING again. Clear it so the
+			// re-queued task is not cancelled at the worker's pre-start check.
+			clearCancelFlag(ctx, existing.ID)
 			if err = s.enqueueTask(existing.ID); err != nil {
 				if rollbackErr := s.rollbackRetriedTask(ctx, existing.ID, originalStatus); rollbackErr != nil {
 					return nil, fmt.Errorf("enqueue task %s: %w (rollback failed: %v)", existing.ID, err, rollbackErr)
@@ -427,6 +441,15 @@ func (s *IngestionTaskService) rollbackCreatedTask(ctx context.Context, taskID s
 	return err
 }
 
+// clearCancelFlag removes the Redis cancel marker ({task_id}-cancel) that
+// RequestStop sets for a RUNNING task. No-op when Redis is unavailable —
+// the DB STOPPING status remains the fallback cancel signal.
+func clearCancelFlag(ctx context.Context, taskID string) {
+	if rc := redis2.Get(); rc != nil {
+		rc.Delete(ctx, fmt.Sprintf("%s-cancel", taskID))
+	}
+}
+
 func (s *IngestionTaskService) enqueueTask(taskID string) error {
 	taskMessage := common.TaskMessage{
 		TaskID:   taskID,

internal/service/ingestion_task_service_test.go

@@ -319,6 +319,30 @@ func TestStartRunningLeavesTerminalDocumentUntouched(t *testing.T) {
 	}
 }
 
+// TestStartRunningFinalizesStoppingTask locks in the redelivery path: a
+// STOPPING task (cancelled after being nacked, before any worker ran it) is
+// moved to STOPPED by StartRunning, so the task reaches a terminal state and
+// a later re-parse can transition it back to CREATED.
+func TestStartRunningFinalizesStoppingTask(t *testing.T) {
+	db := setupServiceTestDB(t)
+	pushServiceDB(t, db)
+	insertTestIngestionTask(t, "task-1", "user-1", "doc-1", "kb-1")
+	if err := db.Model(&entity.IngestionTask{}).Where("id = ?", "task-1").
+		Update("status", common.STOPPING).Error; err != nil {
+		t.Fatalf("set STOPPING: %v", err)
+	}
+
+	svc := NewIngestionTaskService()
+	ctx := t.Context()
+	task, err := svc.StartRunning(ctx, "task-1")
+	if err != nil {
+		t.Fatalf("StartRunning failed: %v", err)
+	}
+	if task.Status != common.STOPPED {
+		t.Fatalf("status = %q, want %q", task.Status, common.STOPPED)
+	}
+}
+
 func TestIngestionTaskServiceRequestStopTransitionsCreatedTaskToStopped(t *testing.T) {
 	db := setupServiceTestDB(t)
 	pushServiceDB(t, db)