From b19b01b2851ddd83ac89884f4e34968ce9dfcec7 Mon Sep 17 00:00:00 2001
From: RidRisR <79858083+RidRisR@users.noreply.github.com>
Date: Tue, 7 Jan 2025 04:37:26 +0100
Subject: [PATCH] infinate backoff for create job

---
 pkg/controller/compact_status_updater.go      |  9 +--
 .../compact_backup_controller.go              | 63 ++++++++++---------
 2 files changed, 35 insertions(+), 37 deletions(-)

diff --git a/pkg/controller/compact_status_updater.go b/pkg/controller/compact_status_updater.go
index c084dd3277..a602094518 100644
--- a/pkg/controller/compact_status_updater.go
+++ b/pkg/controller/compact_status_updater.go
@@ -146,15 +146,8 @@ func (r *CompactStatusUpdater) OnCreateJob(ctx context.Context, compact *v1alpha
 	if err != nil {
 		newStatus.State = string(v1alpha1.BackupRetryTheFailed)
 		newStatus.Message = err.Error()
-		newStatus.RetryStatus = []v1alpha1.CompactRetryRecord{
-			{
-				RetryNum:       len(compact.Status.RetryStatus),
-				DetectFailedAt: metav1.NewTime(time.Now()),
-				RetryReason:    err.Error(),
-			},
-		}
 	} else {
-		newStatus.State = string(v1alpha1.BackupRunning)
+		newStatus.State = string(v1alpha1.BackupPrepare)
 	}
 	return r.UpdateStatus(compact, newStatus)
 }
diff --git a/pkg/controller/compactbackup/compact_backup_controller.go b/pkg/controller/compactbackup/compact_backup_controller.go
index 14f95cd88a..f60e020b8f 100644
--- a/pkg/controller/compactbackup/compact_backup_controller.go
+++ b/pkg/controller/compactbackup/compact_backup_controller.go
@@ -42,6 +42,10 @@ import (
 	"k8s.io/utils/ptr"
 )
 
+const (
+	maxInterval = 6 * time.Minute
+)
+
 // Controller controls backup.
 type Controller struct {
 	deps *controller.Dependencies
@@ -281,21 +285,21 @@ func (c *Controller) sync(key string) (err error) {
 		return nil
 	}
 
-	ok, err := c.precheckCompact(compact)
+	ok, err := c.checkJobStatus(compact)
 	if err != nil {
 		return err
 	}
 	if !ok {
-		klog.Infof("Compact %s/%s is not allowed, skip", ns, name)
+		klog.Infof("Compact %s/%s is not allowed to create job, skip", ns, name)
 		return nil
 	}
 
-	err = c.doCompact(compact.DeepCopy())
+	err = c.createJob(compact.DeepCopy())
 	c.statusUpdater.OnCreateJob(context.TODO(), compact, err)
 	return err
 }
 
-func (c *Controller) doCompact(compact *v1alpha1.CompactBackup) error {
+func (c *Controller) createJob(compact *v1alpha1.CompactBackup) error {
 	ns := compact.GetNamespace()
 	name := compact.GetName()
 	compactJobName := compact.GetName()
@@ -474,17 +478,34 @@ func (c *Controller) makeCompactJob(compact *v1alpha1.CompactBackup) (*batchv1.J
 	return job, "", nil
 }
 
-// precheckCompact checks if doCompact is allowed to run
+func (c *Controller) validate(compact *v1alpha1.CompactBackup) error {
+	spec := compact.Spec
+	if spec.StartTs == "" {
+		return errors.NewNoStackError("start-ts must be set")
+	}
+	if spec.EndTs == "" {
+		return errors.NewNoStackError("end-ts must be set")
+	}
+	if spec.Concurrency <= 0 {
+		return errors.NewNoStackError("concurrency must be greater than 0")
+	}
+	if spec.MaxRetryTimes < 0 {
+		return errors.NewNoStackError("maxRetryTimes must be greater than or equal to 0")
+	}
+	return nil
+}
+
+// checkJobStatus checks if doCompact is allowed to run
 // Only if there is no other compact job existing, doCompact is allowed
 // If the existing job failed, update compact status
-func (c *Controller) precheckCompact(compact *v1alpha1.CompactBackup) (bool, error) {
+func (c *Controller) checkJobStatus(compact *v1alpha1.CompactBackup) (bool, error) {
 	ns := compact.GetNamespace()
 	name := compact.GetName()
 
 	job, err := c.deps.KubeClientset.BatchV1().Jobs(ns).Get(context.TODO(), name, metav1.GetOptions{})
 	if err != nil {
 		if errors.IsNotFound(err) {
-			return c.allowCompact(compact), nil
+			return c.allowCreateJob(compact), nil
 		}
 		klog.Errorf("Failed to get job %s for compact %s/%s, error %v", name, ns, name, err)
 		return false, err
@@ -509,24 +530,7 @@ func (c *Controller) precheckCompact(compact *v1alpha1.CompactBackup) (bool, err
 	return false, nil
 }
 
-func (c *Controller) validate(compact *v1alpha1.CompactBackup) error {
-	spec := compact.Spec
-	if spec.StartTs == "" {
-		return errors.NewNoStackError("start-ts must be set")
-	}
-	if spec.EndTs == "" {
-		return errors.NewNoStackError("end-ts must be set")
-	}
-	if spec.Concurrency <= 0 {
-		return errors.NewNoStackError("concurrency must be greater than 0")
-	}
-	if spec.MaxRetryTimes < 0 {
-		return errors.NewNoStackError("maxRetryTimes must be greater than or equal to 0")
-	}
-	return nil
-}
-
-func (c *Controller) allowCompact(compact *v1alpha1.CompactBackup) bool {
+func (c *Controller) allowCreateJob(compact *v1alpha1.CompactBackup) bool {
 	ns := compact.GetNamespace()
 	name := compact.GetName()
 
@@ -535,15 +539,16 @@ func (c *Controller) allowCompact(compact *v1alpha1.CompactBackup) bool {
 		if attempts <= 1 {
 			return 0
 		}
-		return 10 * time.Duration(math.Pow(10, float64(attempts-1))) * time.Second
+		interval := 10 * time.Duration(math.Pow(10, float64(attempts-1))) * time.Second
+		if interval > maxInterval {
+			return maxInterval
+		}
+		return interval
 	}
 
 	attempts := len(compact.Status.RetryStatus)
 	if attempts > 0 {
 		lastRetry := compact.Status.RetryStatus[attempts-1]
-		if lastRetry.RetryNum >= int(compact.Spec.MaxRetryTimes) {
-			return false
-		}
 		backoff := expBackoff(attempts)
 		if time.Since(lastRetry.DetectFailedAt.Time) < backoff {
 			klog.Infof("Compact: [%s/%s] backoff in effect, skipping retry.", ns, name)