Merge pull request #177 from sgotti/runservice_use_trylock_on_fetching

runservice: use etcd mutex TryLock on fetching

internal/services/runservice/common/common.go

@@ -44,10 +44,12 @@ var (
 
 	EtcdPingKey = path.Join(EtcdSchedulerBaseDir, "ping")
 
-	EtcdCompactChangeGroupsLockKey = path.Join(EtcdSchedulerBaseDir, "compactchangegroupslock")
-	EtcdCacheCleanerLockKey        = path.Join(EtcdSchedulerBaseDir, "locks", "cachecleaner")
-	EtcdWorkspaceCleanerLockKey    = path.Join(EtcdSchedulerBaseDir, "locks", "workspacecleaner")
-	EtcdTaskUpdaterLockKey         = path.Join(EtcdSchedulerBaseDir, "locks", "taskupdater")
+	EtcdLocksDir = path.Join(EtcdSchedulerBaseDir, "locks")
+
+	EtcdCompactChangeGroupsLockKey = path.Join(EtcdLocksDir, "compactchangegroups")
+	EtcdCacheCleanerLockKey        = path.Join(EtcdLocksDir, "cachecleaner")
+	EtcdWorkspaceCleanerLockKey    = path.Join(EtcdLocksDir, "workspacecleaner")
+	EtcdTaskUpdaterLockKey         = path.Join(EtcdLocksDir, "taskupdater")
 
 	EtcdMaintenanceKey = "maintenance"
 )
@@ -55,6 +57,9 @@ var (
 func EtcdRunKey(runID string) string       { return path.Join(EtcdRunsDir, runID) }
 func EtcdExecutorKey(taskID string) string { return path.Join(EtcdExecutorsDir, taskID) }
 func EtcdTaskKey(taskID string) string     { return path.Join(EtcdTasksDir, taskID) }
+func EtcdTaskFetcherLockKey(taskID string) string {
+	return path.Join(EtcdLocksDir, "taskfetcher", taskID)
+}
 
 const (
 	EtcdChangeGroupMinRevisionRange = 100

internal/services/runservice/scheduler.go

@@ -1011,11 +1011,12 @@ func (s *Runservice) fetchTaskLogs(ctx context.Context, runID string, rt *types.
 	if rt.SetupStep.LogPhase == types.RunTaskFetchPhaseNotStarted {
 		if err := s.fetchLog(ctx, rt, true, 0); err != nil {
 			log.Errorf("err: %+v", err)
-		}
+		} else {
 			if err := s.finishSetupLogPhase(ctx, runID, rt.ID); err != nil {
 				log.Errorf("err: %+v", err)
 			}
 		}
+	}
 
 	// fetch steps logs
 	for i, rts := range rt.Steps {
@@ -1133,7 +1134,35 @@ func (s *Runservice) fetcher(ctx context.Context) error {
 		log.Debugf("r: %s", util.Dump(r))
 		for _, rt := range r.Tasks {
 			log.Debugf("rt: %s", util.Dump(rt))
-			if rt.Status.IsFinished() {
+			if err := s.taskFetcher(ctx, r, rt); err != nil {
+				return err
+			}
+		}
+	}
+
+	return nil
+}
+
+func (s *Runservice) taskFetcher(ctx context.Context, r *types.Run, rt *types.RunTask) error {
+	if !rt.Status.IsFinished() {
+		return nil
+	}
+	session, err := concurrency.NewSession(s.e.Client(), concurrency.WithTTL(5), concurrency.WithContext(ctx))
+	if err != nil {
+		return err
+	}
+	defer session.Close()
+
+	m := etcd.NewMutex(session, common.EtcdTaskFetcherLockKey(rt.ID))
+
+	if err := m.TryLock(ctx); err != nil {
+		if errors.Is(err, etcd.ErrLocked) {
+			return nil
+		}
+		return err
+	}
+	defer func() { _ = m.Unlock(ctx) }()
+
 	// write related logs runID
 	runIDPath := store.OSTRunTaskLogsRunPath(rt.ID, r.ID)
 	exists, err := s.OSTFileExists(runIDPath)
@@ -1167,11 +1196,8 @@ func (s *Runservice) fetcher(ctx context.Context) error {
 			return err
 		}
 	}
-			}
-		}
-	}
-	return nil
 
+	return nil
 }
 
 func (s *Runservice) runsSchedulerLoop(ctx context.Context) {