Set DefaultRequeuingBackoffBaseSeconds to 60s. (kubernetes-sigs#2251)

apis/config/v1beta1/configuration_types.go

@@ -247,7 +247,7 @@ type RequeuingStrategy struct {
 	// - "Rand" represents the random jitter.
 	// During this time, the workload is taken as an inadmissible and
 	// other workloads will have a chance to be admitted.
-	// By default, the consecutive requeue delays are around: (10s, 20s, 40s, ...).
+	// By default, the consecutive requeue delays are around: (60s, 120s, 240s, ...).
 	//
 	// Defaults to null.
 	// +optional
@@ -256,7 +256,7 @@ type RequeuingStrategy struct {
 	// BackoffBaseSeconds defines the base for the exponential backoff for
 	// re-queuing an evicted workload.
 	//
-	// Defaults to 10.
+	// Defaults to 60.
 	// +optional
 	BackoffBaseSeconds *int32 `json:"backoffBaseSeconds,omitempty"`
 }

apis/config/v1beta1/defaults.go

@@ -47,7 +47,7 @@ const (
 	DefaultMultiKueueGCInterval                         = time.Minute
 	DefaultMultiKueueOrigin                             = "multikueue"
 	DefaultMultiKueueWorkerLostTimeout                  = 15 * time.Minute
-	DefaultRequeuingBackoffBaseSeconds                  = 10
+	DefaultRequeuingBackoffBaseSeconds                  = 60
 )
 
 func getOperatorNamespace() string {

charts/kueue/values.yaml

@@ -85,7 +85,7 @@ managerConfig:
     #  requeuingStrategy:
     #    timestamp: Eviction
     #    backoffLimitCount: null # null indicates infinite requeuing
-    #    backoffBaseSeconds: 10
+    #    backoffBaseSeconds: 60
     #manageJobsWithoutQueueName: true
     #internalCertManagement:
     #  enable: false

config/components/manager/controller_manager_config.yaml

@@ -29,7 +29,7 @@ clientConnection:
 #  requeuingStrategy:
 #    timestamp: Eviction
 #    backoffLimitCount: null # null indicates infinite requeuing
-#    backoffBaseSeconds: 10
+#    backoffBaseSeconds: 60
 #manageJobsWithoutQueueName: true
 #internalCertManagement:
 #  enable: false

keps/1282-pods-ready-requeue-strategy/README.md

@@ -143,12 +143,12 @@ type RequeuingStrategy struct {
 	// Once the number is reached, the workload is deactivated (`.spec.activate`=`false`).
 	// When it is null, the workloads will repeatedly and endless re-queueing.
 	//
-	// Every backoff duration is about "10s*2^(n-1)+Rand" where:
+	// Every backoff duration is about "60s*2^(n-1)+Rand" where:
 	// - "n" represents the "workloadStatus.requeueState.count",
 	// - "Rand" represents the random jitter.
 	// During this time, the workload is taken as an inadmissible and
 	// other workloads will have a chance to be admitted.
-	// By default, the consecutive requeue delays are around: (10s, 20s, 40s, ...).
+	// By default, the consecutive requeue delays are around: (60s, 120s, 240s, ...).
 	//
 	// Defaults to null.
 	// +optional
@@ -157,7 +157,7 @@ type RequeuingStrategy struct {
 	// BackoffBaseSeconds defines the base for the exponential backoff for
 	// re-queuing an evicted workload.
 	//
-	// Defaults to 10.
+	// Defaults to 60.
 	// +optional
 	BackoffBaseSeconds *int32 `json:"backoffBaseSeconds,omitempty"`
 }
@@ -230,16 +230,16 @@ Duration this time, other workloads will have a chance to be admitted.
 
 The queueManager calculates an exponential backoff duration by [the Step function](https://pkg.go.dev/k8s.io/apimachinery/pkg/util/[email protected]#Backoff.Step)
 according to the $b*2^{(n-1)}+Rand$ where:
-- $b$ represents the base delay, configured by `baseDelaySeconds`
+- $b$ represents the base delay, configured by `backoffBaseSeconds`
 - $n$ represents the `workloadStatus.requeueState.count`,
 - $Rand$ represents the random jitter.
 
 It will spend awaiting to be requeued after eviction:
 $$\sum_{k=1}^{n}(b*2^{(k-1)} + Rand)$$
 
-Assuming `backoffLimitCount` equals 10, and `baseDelaySeconds` equals 10 (default) the workload is requeued 10 times
+Assuming `backoffLimitCount` equals 10, and `backoffBaseSeconds` equals 60 (default) the workload is requeued 10 times
 after failing to have all pods ready, then the total time awaiting for requeue
-will take (neglecting the jitter): `10s+20s+40s +...+7680s=2h 8min`.
+will take (neglecting the jitter): `60s+120s+240s +...+30720s=8h 32min`.
 Also, considering `.waitForPodsReady.timeout=300s` (default),
 the workload will spend `50min` total waiting for pods ready.
 

pkg/controller/core/workload_controller.go

@@ -455,7 +455,7 @@ func (r *WorkloadReconciler) triggerDeactivationOrBackoffRequeue(ctx context.Con
 			"Deactivated Workload %q by reached re-queue backoffLimitCount", klog.KObj(wl))
 		return true, nil
 	}
-	// Every backoff duration is about "10s*2^(n-1)+Rand" where:
+	// Every backoff duration is about "60s*2^(n-1)+Rand" where:
 	// - "n" represents the "requeuingCount",
 	// - "Rand" represents the random jitter.
 	// During this time, the workload is taken as an inadmissible and other

site/content/en/docs/reference/kueue-config.v1beta1.md

@@ -719,7 +719,7 @@ When it is null, the workloads will repeatedly and endless re-queueing.</p>
 <li>&quot;Rand&quot; represents the random jitter.
 During this time, the workload is taken as an inadmissible and
 other workloads will have a chance to be admitted.
-By default, the consecutive requeue delays are around: (10s, 20s, 40s, ...).</li>
+By default, the consecutive requeue delays are around: (60s, 120s, 240s, ...).</li>
 </ul>
 <p>Defaults to null.</p>
 </td>
@@ -730,7 +730,7 @@ By default, the consecutive requeue delays are around: (10s, 20s, 40s, ...).</li
 <td>
    <p>BackoffBaseSeconds defines the base for the exponential backoff for
 re-queuing an evicted workload.</p>
-<p>Defaults to 10.</p>
+<p>Defaults to 60.</p>
 </td>
 </tr>
 </tbody>

site/content/en/docs/tasks/manage/setup_sequential_admission.md

@@ -47,7 +47,7 @@ fields:
       requeuingStrategy:
         timestamp: Eviction | Creation
         backoffLimitCount: 5
-        backoffBaseSeconds: 10
+        backoffBaseSeconds: 60
 ```
 
 {{% alert title="Note" color="primary" %}}
@@ -99,8 +99,8 @@ _The `backoffBaseSeconds` is available in Kueue v0.7.0 and later_
 {{% /alert %}}
 The time to re-queue a workload after each consecutive timeout is increased
 exponentially, with the exponent of 2. The first delay is determined by the
-`backoffBaseSeconds` parameter (defaulting to 10). So, after the consecutive timeouts
-the evicted workload is re-queued after approximately `10, 20, 40, ...` seconds.
+`backoffBaseSeconds` parameter (defaulting to 60). So, after the consecutive timeouts
+the evicted workload is re-queued after approximately `60, 120, 240, ...` seconds.
 
 ## Example