chore: collect the log of the backup failed (#8208)

apecloud · Sep 26, 2024 · 7c93f30 · 7c93f30
1 parent f2fbe45
commit 7c93f30
Showing 1 changed file with 22 additions and 11 deletions.
diff --git a/controllers/dataprotection/log_collection_controller.go b/controllers/dataprotection/log_collection_controller.go
@@ -42,6 +42,7 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/predicate"
 
 	dpv1alpha1 "github.com/apecloud/kubeblocks/apis/dataprotection/v1alpha1"
+	"github.com/apecloud/kubeblocks/pkg/constant"
 	intctrlutil "github.com/apecloud/kubeblocks/pkg/controllerutil"
 	dprestore "github.com/apecloud/kubeblocks/pkg/dataprotection/restore"
 	dptypes "github.com/apecloud/kubeblocks/pkg/dataprotection/types"
@@ -171,18 +172,28 @@ func (r *LogCollectionReconciler) patchBackupStatus(reqCtx intctrlutil.RequestCt
 	if err := r.Client.Get(reqCtx.Ctx, client.ObjectKey{Name: backupName, Namespace: reqCtx.Req.Namespace}, backup); err != nil {
 		return err
 	}
-	if backup.Status.FailureReason != "" {
-		return nil
-	}
-	errorLogs, err := r.collectErrorLogs(reqCtx, job)
-	if err != nil {
-		return fmt.Errorf("collect error logs failed: %s", err.Error())
-	}
-	if errorLogs == "" {
-		return nil
+	for i := range backup.Status.Actions {
+		action := &backup.Status.Actions[i]
+		if action.ObjectRef == nil {
+			continue
+		}
+		if action.ObjectRef.Kind != constant.JobKind || action.ObjectRef.Name != job.Name {
+			continue
+		}
+		if strings.HasPrefix(action.FailureReason, dptypes.LogCollectorOutput) {
+			return nil
+		}
+		errorLogs, err := r.collectErrorLogs(reqCtx, job)
+		if err != nil {
+			return fmt.Errorf("collect error logs failed: %s", err.Error())
+		}
+		if errorLogs == "" {
+			return nil
+		}
+		action.FailureReason = fmt.Sprintf("%s: %s", dptypes.LogCollectorOutput, errorLogs)
+		return r.Client.Status().Update(reqCtx.Ctx, backup)
 	}
-	backup.Status.FailureReason = errorLogs
-	return r.Client.Status().Update(reqCtx.Ctx, backup)
+	return nil
 }
 
 func (r *LogCollectionReconciler) patchRestoreStatus(reqCtx intctrlutil.RequestCtx, job *batchv1.Job, restoreName string) error {