Merge pull request #7752 from qiuming-best/maintenance-job-start-fix

Maintenance job should not be launched if the repo already has a runn…
vmware-tanzu · May 7, 2024 · 3cbf2eb · 3cbf2eb
2 parents dda6c1f + e91d9b9
commit 3cbf2eb
Show file tree

Hide file tree

Showing 3 changed files with 23 additions and 7 deletions.
diff --git a/pkg/repository/maintenance.go b/pkg/repository/maintenance.go
@@ -238,10 +238,15 @@ func getMaintenanceResultFromJob(cli client.Client, job *batchv1.Job) (string, e
 	return podList.Items[0].Status.ContainerStatuses[0].State.Terminated.Message, nil
 }
 
-func GetLatestMaintenanceJob(cli client.Client, repo string) (*batchv1.Job, error) {
+func getLatestMaintenanceJob(cli client.Client, ns string) (*batchv1.Job, error) {
 	// Get the maintenance job list by label
 	jobList := &batchv1.JobList{}
-	err := cli.List(context.TODO(), jobList, client.MatchingLabels(map[string]string{RepositoryNameLabel: repo}))
+	err := cli.List(context.TODO(), jobList, &client.ListOptions{
+		Namespace: ns,
+	},
+		&client.HasLabels{RepositoryNameLabel},
+	)
+
 	if err != nil {
 		return nil, err
 	}
@@ -254,5 +259,6 @@ func GetLatestMaintenanceJob(cli client.Client, repo string) (*batchv1.Job, erro
 	sort.Slice(jobList.Items, func(i, j int) bool {
 		return jobList.Items[i].CreationTimestamp.Time.After(jobList.Items[j].CreationTimestamp.Time)
 	})
+
 	return &jobList.Items[0], nil
 }
diff --git a/pkg/repository/maintenance_test.go b/pkg/repository/maintenance_test.go
@@ -257,7 +257,7 @@ func TestGetLatestMaintenanceJob(t *testing.T) {
 	cli := fake.NewClientBuilder().WithScheme(scheme).WithObjects(objs...).Build()
 
 	// Call the function
-	job, err := GetLatestMaintenanceJob(cli, repo)
+	job, err := getLatestMaintenanceJob(cli, "default")
 	assert.NoError(t, err)
 
 	// We expect the returned job to be the newer job

diff --git a/pkg/repository/manager.go b/pkg/repository/manager.go
@@ -191,17 +191,27 @@ func (m *manager) PruneRepo(repo *velerov1api.BackupRepository) error {
 		return errors.WithStack(err)
 	}
 
-	if err := prd.BoostRepoConnect(context.Background(), param); err != nil {
-		return errors.WithStack(err)
-	}
-
 	log := m.log.WithFields(logrus.Fields{
 		"BSL name":  param.BackupLocation.Name,
 		"repo type": param.BackupRepo.Spec.RepositoryType,
 		"repo name": param.BackupRepo.Name,
 		"repo UID":  param.BackupRepo.UID,
 	})
 
+	job, err := getLatestMaintenanceJob(m.client, m.namespace)
+	if err != nil {
+		return errors.WithStack(err)
+	}
+
+	if job != nil && job.Status.Succeeded == 0 && job.Status.Failed == 0 {
+		log.Debugf("There already has a unfinished maintenance job %s/%s for repository %s, please wait for it to complete", job.Namespace, job.Name, param.BackupRepo.Name)
+		return nil
+	}
+
+	if err := prd.BoostRepoConnect(context.Background(), param); err != nil {
+		return errors.WithStack(err)
+	}
+
 	log.Info("Start to maintence repo")
 
 	maintenanceJob, err := buildMaintenanceJob(m.maintenanceCfg, param, m.client, m.namespace)