From 6f33dc63e724770f16d2e7d212a043699befadc0 Mon Sep 17 00:00:00 2001
From: Lyndon-Li <lyonghui@vmware.com>
Date: Wed, 15 Jan 2025 16:18:14 +0800
Subject: [PATCH] get maintenance result only for failed jobs

Signed-off-by: Lyndon-Li <lyonghui@vmware.com>
---
 pkg/repository/maintenance/maintenance.go     | 24 ++++++++++++-----
 .../maintenance/maintenance_test.go           | 26 +++++++++++++++++--
 2 files changed, 41 insertions(+), 9 deletions(-)

diff --git a/pkg/repository/maintenance/maintenance.go b/pkg/repository/maintenance/maintenance.go
index b854046ba0..a01a2c464d 100644
--- a/pkg/repository/maintenance/maintenance.go
+++ b/pkg/repository/maintenance/maintenance.go
@@ -250,11 +250,16 @@ func WaitJobComplete(cli client.Client, ctx context.Context, jobName, ns string,
 		return velerov1api.BackupRepositoryMaintenanceStatus{}, errors.Wrap(err, "error to wait for maintenance job complete")
 	}
 
-	log.Info("Maintenance repo complete")
+	log.Infof("Maintenance repo complete, succeeded %v, failed %v", maintenanceJob.Status.Succeeded, maintenanceJob.Status.Failed)
 
-	result, err := getResultFromJob(cli, maintenanceJob)
-	if err != nil {
-		log.WithError(err).Warn("Failed to get maintenance job result")
+	result := ""
+	if maintenanceJob.Status.Failed > 0 {
+		if r, err := getResultFromJob(cli, maintenanceJob); err != nil {
+			log.WithError(err).Warn("Failed to get maintenance job result")
+			result = "Repo maintenance failed but result is not retrieveable"
+		} else {
+			result = r
+		}
 	}
 
 	return composeStatusFromJob(maintenanceJob, result), nil
@@ -303,9 +308,14 @@ func WaitAllJobsComplete(ctx context.Context, cli client.Client, repo *velerov1a
 			job = updated
 		}
 
-		message, err := getResultFromJob(cli, job)
-		if err != nil {
-			return nil, errors.Wrapf(err, "error getting maintenance job[%s] result", job.Name)
+		message := ""
+		if job.Status.Failed > 0 {
+			if msg, err := getResultFromJob(cli, job); err != nil {
+				log.WithError(err).Warnf("Failed to get result of maintenance job %s", job.Name)
+				message = "Repo maintenance failed but result is not retrieveable"
+			} else {
+				message = msg
+			}
 		}
 
 		history = append(history, composeStatusFromJob(job, message))
diff --git a/pkg/repository/maintenance/maintenance_test.go b/pkg/repository/maintenance/maintenance_test.go
index e4335a61fe..4a543426b3 100644
--- a/pkg/repository/maintenance/maintenance_test.go
+++ b/pkg/repository/maintenance/maintenance_test.go
@@ -602,13 +602,35 @@ func TestWaitAllJobsComplete(t *testing.T) {
 			expectedError: "error waiting maintenance job[job1] complete: context deadline exceeded",
 		},
 		{
-			name:          "get result error",
+			name:          "get result error on succeded job",
 			ctx:           context.TODO(),
 			runtimeScheme: scheme,
 			kubeClientObj: []runtime.Object{
 				jobSucceeded1,
 			},
-			expectedError: "error getting maintenance job[job1] result: no pod found for job job1",
+			expectedStatus: []velerov1api.BackupRepositoryMaintenanceStatus{
+				{
+					Result:            velerov1api.BackupRepositoryMaintenanceSucceeded,
+					StartTimestamp:    &metav1.Time{Time: now},
+					CompleteTimestamp: &metav1.Time{Time: now.Add(time.Hour)},
+				},
+			},
+		},
+		{
+			name:          "get result error on failed job",
+			ctx:           context.TODO(),
+			runtimeScheme: scheme,
+			kubeClientObj: []runtime.Object{
+				jobFailed1,
+			},
+			expectedStatus: []velerov1api.BackupRepositoryMaintenanceStatus{
+				{
+					Result:            velerov1api.BackupRepositoryMaintenanceFailed,
+					StartTimestamp:    &metav1.Time{Time: now.Add(time.Hour)},
+					CompleteTimestamp: &metav1.Time{Time: now.Add(time.Hour * 2)},
+					Message:           "Repo maintenance failed but result is not retrieveable",
+				},
+			},
 		},
 		{
 			name:          "less than limit",