dra: scheduler queueing hints #119078

pohly · 2023-07-04T15:55:43Z

What type of PR is this?

/kind feature

What this PR does / why we need it:

Scheduling with less overhead and without delays when ResourceClaims are involved.

Which issue(s) this PR fixes:

Related-to: #118893 (comment)

Special notes for your reviewer:

This is a combination of two related enhancements:

By implementing a PreEnqueue check, the initial pod scheduling
attempt for a pod with a claim template gets avoided when the claim
does not exist yet.
By implementing cluster event checks, only those pods get
scheduled for which something changed, and they get scheduled
immediately without delay.

Does this PR introduce a user-facing change?

faster scheduling when ResourceClaims are involved

Additional documentation e.g., KEPs (Kubernetes Enhancement Proposals), usage docs, etc.:

- [KEP]: https://github.com/kubernetes/enhancements/issues/3063

k8s-ci-robot · 2023-07-04T15:55:52Z

This issue is currently awaiting triage.

If a SIG or subproject determines this is a relevant issue, they will accept it by applying the triage/accepted label and provide further guidance.

The triage/accepted label can be added by org members by writing /triage accepted in a comment.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

pohly · 2023-07-04T15:56:49Z

/assign @sanposhiho

sanposhiho

I haven't got enough time to review it, will do the full review later

sanposhiho · 2023-07-05T00:56:23Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

-		// TODO (#113702): can we change this so that such an event does not trigger *all* pods?
-		// Yes: https://github.com/kubernetes/kubernetes/blob/abcbaed0784baf5ed2382aae9705a8918f2daa18/pkg/scheduler/eventhandlers.go#L70
-		{Event: framework.ClusterEvent{Resource: framework.PodSchedulingContext, ActionType: framework.Add | framework.Update}},
+		{Event: framework.ClusterEvent{Resource: framework.PodSchedulingContext, ActionType: framework.Add | framework.Update}, QueueingHintFn: pl.isSchedulableAfterPodSchedulingContextChange},
 		// A resource might depend on node labels for topology filtering.
 		// A new or updated node may make pods schedulable.
 		{Event: framework.ClusterEvent{Resource: framework.Node, ActionType: framework.Add | framework.UpdateNodeLabel}},


Cannot we add QueueingHint on Node add event/update event? (Or do you want to do it in another PR?)

It should be doable, but definitely in a separate PR. Getting the checks wrong is a potential source of bugs, so I don't want to do too much at once.

carlory · 2023-07-07T07:31:01Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+}
+
+// isSchedulableAfterClaimChange is invoked whenever a claim changed. It checks whether
+// that change made a previously unschedulable pod schedulable. It errs on the side of


Suggested change

// that change made a previously unschedulable pod schedulable. It errs on the side of

// that change made a previous unschedulable pod schedulable. It errs on the side of

Previously refers to unschedulable, not pod. I believe previously is correct here - but I am not a native speaker.

sanposhiho · 2023-07-07T07:59:06Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+		return framework.QueueSkip
+	}
+
+	modifiedClaim, ok := newObj.(*resourcev1alpha2.ResourceClaim)


Can we use As function to convert oldObj and newObj.
https://github.com/kubernetes/kubernetes/blob/master/pkg/scheduler/util/utils.go#L167

Looking... As doesn't handle cache.DeletedFinalStateUnknown. That is okay here (can't occur for newObj, but for oldObj in isSchedulableAfterPodSchedulingContextChange it is relevant.

Shall I add support for it in this PR?

I've added it, will be in the next push.

sanposhiho · 2023-07-07T08:45:24Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+	return nil
+}
+
+// isSchedulableAfterClaimChange is invoked whenever a claim changed. It checks whether


nit

Suggested change

// isSchedulableAfterClaimChange is invoked whenever a claim changed. It checks whether

// isSchedulableAfterClaimChange is invoked whenever a claim added or changed. It checks whether

... or deleted. Instead of listing all possible reasons, I changed this to "is invoked for all claim events reported by an informer".

This also explains why handling cache.DeletedFinalStateUnknown is required.

sanposhiho · 2023-07-08T05:10:42Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+	// Deleted? That can happen because we ourselves delete the PodSchedulingContext while
+	// working on the pod. This can be ignored.
+	if oldObj != nil && newObj == nil {
+		pl.logger.V(4).Info("PodSchedulingContext for pod got deleted", "pod", klog.KObj(pod))
+		return framework.QueueSkip
+	}


This deleted event handling can be done around L372 (before converting type).

The log message then shouldn't mention the pod. It's simpler, so I moved the check.

sanposhiho · 2023-07-08T05:49:45Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+	// here is that in a situation where the cluster really is
+	// full, backoff won't be used because the scheduler keeps
+	// trying different nodes. This should not happen when it has
+	// full knowledge about resource availability (=
+	// PodSchedulingContext.*.UnsuitableNodes is complete) but may happen
+	// when it doesn't (= PodSchedulingContext.*.UnsuitableNodes had to be
+	// truncated).
+	if podScheduling.Spec.SelectedNode != "" {


I'd like to understand this part correctly.

Let's say Node1 - 4 are in the cluster and Node2-4 doesn't have enough capacity. So, the scheduler set podScheduling.Spec.SelectedNode=Node1. But, later, the device driver said claimStatus.UnsuitableNodes = Node1. And we reach here and the Pod will get QueueImmediately.

In this case, in the next scheduling cycle, Node2-4 are still rejected by the resource fit plugin or whatever, and Node1 is rejected by the DRA plugin. Meaning, this Pod won't be schedulable until Node2-4 get free space.
But, the Pod could keep getting rescheduled by QueueImmediately.

You mentioned such risk in the above comment, right?
How can we avoid this situation? Could you elaborate This should not happen when it has full knowledge about resource availability?

But, the Pod could keep getting rescheduled by QueueImmediately.

There will only be one scheduling attempt, the one triggered by this cluster event. After that, there won't be any further updates of the PodSchedulingContext unless a driver reports a change in resource availability, so there will be no further "immediate" queueing.

You mentioned such risk in the above comment, right?

That risk is slightly different: UnsuitableNodes has a size limit. If a driver cannot report all unsuitable nodes, then the scheduler might alternate between different "selected nodes" that are all unsuitable.

This is an issue that we need to keep an eye on, but I think it can be ignored for now in this PR (alpha stage, not limited to scheduler events).

There will only be one scheduling attempt, the one triggered by this cluster event.

Ah, true.

UnsuitableNodes has a size limit. If a driver cannot report all unsuitable nodes, then the scheduler might alternate between different "selected nodes" that are all unsuitable.

OK, that's what you said "truncated". Understood.

Can we have a comment to describe the risk is acceptable for now?

Added. I also added a few more comments in other places.

sanposhiho · 2023-07-08T05:51:25Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+		}
+	}
+
+	if oldPodScheduling != nil &&


If we reach here, oldPodScheduling != nil is always true due to L442.

True, changed.

Maybe you have forgotten to fix this?

No, oldPodScheduling is nil for a "create" event.

Nah, OK never mind. 🙏

sanposhiho · 2023-07-08T05:52:26Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+	if loggerV := pl.logger.V(6); loggerV.Enabled() {
+		loggerV.Info("PodSchedulingContext for pod with no relevant changes, maybe schedule", "pod", klog.KObj(pod), "podSchedulingDiff", cmp.Diff(oldPodScheduling, podScheduling))
+	} else {
+		pl.logger.V(5).Info("PodSchedulingContext for pod with no relevant changes, maybe schedule", "pod", klog.KObj(pod))
+	}
+	return framework.QueueAfterBackoff


In which case we could reach here? Cannot we return QueueSkip if we know it's non relevant change?

We don't know for sure that the change can be ignored. When in doubt, "backoff" (and thus try later) IMHO is the better choice and matches what the scheduler would do without scheduling hints.

I'll change the wording to "with unknown changes" instead of "no relevant changes", that's more accurate for this situation. The diff is included at V(6) precisely because a developer might want to know more.

That makes sense

pohly · 2023-07-08T11:54:46Z

@sanposhiho : I pushed an update, please take another look.

sanposhiho

Only nits

sanposhiho · 2023-07-08T12:38:33Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+		}
+	}
+
+	if oldPodScheduling != nil &&


Maybe you have forgotten to fix this?

sanposhiho · 2023-07-08T12:39:24Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+		loggerV.Info("PodSchedulingContext for pod with unknown changes, maybe schedule", "pod", klog.KObj(pod), "podSchedulingDiff", cmp.Diff(oldPodScheduling, podScheduling))
+	} else {
+		pl.logger.V(5).Info("PodSchedulingContext for pod with unknown changes, maybe schedule", "pod", klog.KObj(pod))
+	}


Can we have some comment to describe that if we reach here, something unknown happens and that's why we return QueueAfterBackoff

sanposhiho · 2023-07-08T12:40:05Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+	// here is that in a situation where the cluster really is
+	// full, backoff won't be used because the scheduler keeps
+	// trying different nodes. This should not happen when it has
+	// full knowledge about resource availability (=
+	// PodSchedulingContext.*.UnsuitableNodes is complete) but may happen
+	// when it doesn't (= PodSchedulingContext.*.UnsuitableNodes had to be
+	// truncated).
+	if podScheduling.Spec.SelectedNode != "" {


Can we have a comment to describe the risk is acceptable for now?

sanposhiho · 2023-07-08T12:43:46Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources_test.go

+		},
+	}
+
+	for name, tc := range testcases {


nit: do you want to make this test parallel as well? (because other UTs in this test do that.

pohly · 2023-07-08T15:04:31Z

@sanposhiho: updated once more.

sanposhiho

Everything is clear to me. Thanks!
/lgtm

k8s-ci-robot · 2023-07-09T03:19:34Z

LGTM label has been added.

Git tree hash: d9fa124cb687dd64b7c920d41e754a031f5a256a

sanposhiho · 2023-07-09T03:22:11Z

@kubernetes/sig-scheduling-leads @kerthcet
Could anyone else give another review for /approve on it?

sanposhiho · 2023-07-09T08:49:17Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+	// TODO: the runtime should set up logging for each plugin, including
+	// adding a name for each one (same as in kube-controller-manager).
+	return NewWithLogger(klog.TODO(), plArgs, fh, fts)


Do we have an issue for this TODO by the way?

Not specifically for this one, but there are PRs pending for the conversion to contextual logging. Once those are complete, we can search for klog.TODO and fix it if not done yet.

Seems better to pass-in a logger here, rather than a context like this https://github.com/kubernetes/kubernetes/pull/116884/files#diff-b5d92dd0897babb368df8c1aad9e95c957eca378bd46481d6d8501011f722715R77-R81

Whatever gets done here must get coordinated when converting kube-scheduler to contextual logging. Let's not open that rathole right now.

carlory · 2023-07-10T08:09:14Z

pkg/scheduler/util/utils.go

@@ -164,6 +165,8 @@ func IsScalarResourceName(name v1.ResourceName) bool {
 // As converts two objects to the given type.
 // Both objects must be of the same type. If not, an error is returned.
 // nil objects are allowed and will be converted to nil.
+// For oldObj, cache.DeletedFinalStateUnknown is handled and the
+// pobject stored in it will be converted instead.


Suggested change

// pobject stored in it will be converted instead.

// pobject stored in it will be converted instead.

// pod object stored in it will be converted instead.

kerthcet · 2023-07-10T12:55:13Z

I will take a look tomorrow, I still have jobs at hand right now.

kerthcet · 2023-07-10T12:49:27Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+func (pl *dynamicResources) PreEnqueue(ctx context.Context, pod *v1.Pod) (status *framework.Status) {
+	logger := klog.FromContext(ctx)
+	defer func() {
+		logger.V(5).Info("PreEnqueue", "pod", klog.KObj(pod), "status", status)


I would remove this because we logged them at the calling side.

kerthcet · 2023-07-10T13:10:10Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+
+	if !usesClaim {
+		// This was not the claim the pod was waiting for.
+		pl.logger.V(6).Info("unrelated claim got modified", "pod", klog.KObj(pod), "claim", klog.KObj(modifiedClaim))


why not still v(4), I think it's similar with other logs.

Because this is going to much more common than the other ones and not as interesting as those. If a developer wants to know everything, they should use -v6.

Some other examples:

pkg/kubelet/pleg/evented.go: if klog.V(6).Enabled() { pkg/kubelet/pleg/evented.go- klog.ErrorS(err, "Evented PLEG: error generating pod status from the received event", "podUID", podID, "podStatus", status) -- pkg/kubelet/pleg/evented.go: if klogV := klog.V(6); klogV.Enabled() { pkg/kubelet/pleg/evented.go- klogV.InfoS("Evented PLEG: Generated pod status from the received event", "podUID", podID, "podStatus", status) -- pkg/kubelet/pleg/generic.go: if klog.V(6).Enabled() { pkg/kubelet/pleg/generic.go- klog.ErrorS(err, "PLEG: Write status", "pod", klog.KRef(pod.Namespace, pod.Name), "podStatus", status) -- pkg/kubelet/pleg/generic.go: if klogV := klog.V(6); klogV.Enabled() { pkg/kubelet/pleg/generic.go- klogV.InfoS("PLEG: Write status", "pod", klog.KRef(pod.Namespace, pod.Name), "podStatus", status) -- staging/src/k8s.io/apiserver/pkg/util/flowcontrol/fairqueuing/queueset/queueset.go: if klogV := klog.V(6); klogV.Enabled() { staging/src/k8s.io/apiserver/pkg/util/flowcontrol/fairqueuing/queueset/queueset.go- chosenQueue := qs.queues[bestQueueIdx]

kerthcet · 2023-07-10T13:15:32Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+	}
+
+	pl.logger.V(4).Info("status of claim for pod got updated", "pod", klog.KObj(pod), "claim", klog.KObj(modifiedClaim))
+	return framework.QueueImmediately


I think we should be strict about QueueImmediately, I didn't see any special reason here, do we have?

It is needed to make the scheduler respond without backoff delay when a claim got allocated.

Also note that this was discussed in one of the previous PRs which had this code and the conclusion there was that queuing immediately is okay - I can try to find that discussion again if you want.

QueueImmediately in this case should be OK.
Previous discussion: #117561 (comment)

kerthcet · 2023-07-10T13:19:27Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+	oldPodScheduling, newPodScheduling, err := schedutil.As[*resourcev1alpha2.PodSchedulingContext](oldObj, newObj)
+	if err != nil {
+		// Shouldn't happen.
+		pl.logger.Error(nil, "isSchedulableAfterPodSchedulingChange")


Let's keep the same with isSchedulableAfterClaimChange as isSchedulableAfterPodSchedulingContextAddOrUpdate?

Ah, the function name needs to be updated - fixed.

kerthcet · 2023-07-10T13:22:30Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+	podScheduling := newPodScheduling // Never nil because deletes are handled above.
+
+	if podScheduling.Name != pod.Name || podScheduling.Namespace != pod.Namespace {
+		pl.logger.V(7).Info("PodSchedulingContext for unrelated pod got modified", "pod", klog.KObj(pod), "podScheduling", klog.KObj(podScheduling))


Seems we never use v(7) in scheduler before, I think it's ok but why v(5) or v(10) not work for you? I generally use v(4) for debugging, v(5) for tracing as suggested by https://github.com/kubernetes/community/blob/master/contributors/devel/sig-instrumentation/logging.md#what-method-to-use

Nothing >5 is standardized. Each additional increment just adds more information. Therefore V(6) seems fine to me. V(10) tends to be extremely noisy, which would make debugging this code at -v10 not very practical.

See also #119078 (comment) for other examples.

kerthcet · 2023-07-10T13:50:18Z

pkg/scheduler/framework/plugins/dynamicresources/dynamicresources.go

+		return framework.QueueSkip
+	}
+
+	if oldPodScheduling == nil /* create */ ||


One question about AllocationMode, is it immutable?

pohly · 2023-07-10T15:11:30Z

/retest

This makes some complex values a bit more readable.

Informer callbacks must be prepared to get cache.DeletedFinalStateUnknown as the deleted object. They can use that as hint that some information may have been missed, but typically they just retrieve the stored object inside it.

This is a combination of two related enhancements: - By implementing a PreEnqueue check, the initial pod scheduling attempt for a pod with a claim template gets avoided when the claim does not exist yet. - By implementing cluster event checks, only those pods get scheduled for which something changed, and they get scheduled immediately without delay.

sanposhiho · 2023-07-12T10:03:35Z

/retest

sanposhiho

/lgtm
/approve

k8s-ci-robot · 2023-07-12T11:17:34Z

LGTM label has been added.

Git tree hash: bea34d1676accc5534cbf00a91996c9f1de486f6

k8s-ci-robot · 2023-07-12T11:17:56Z

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: pohly, sanposhiho

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

~~pkg/scheduler/OWNERS~~ [sanposhiho]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

k8s-ci-robot · 2023-07-12T12:18:58Z

@pohly: The following test failed, say /retest to rerun all failed tests or /retest-required to rerun all mandatory failed tests:

Test name	Commit	Details	Required	Rerun command
pull-kubernetes-conformance-kind-ga-only-parallel	`6f1a295`	link	unknown	`/test pull-kubernetes-conformance-kind-ga-only-parallel`

Full PR test history. Your PR dashboard. Please help us cut down on flakes by linking to an open issue when you hit one in your PR.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. I understand the commands that are listed here.

sanposhiho · 2023-07-12T12:33:52Z

/retest

k8s-ci-robot requested review from bart0sh and klueska July 4, 2023 15:56

k8s-ci-robot assigned sanposhiho Jul 4, 2023

pohly mentioned this pull request Jul 4, 2023

[Umbrella] Implement QueueingHintFn in in-tree plugins #118893

Closed

12 tasks

sanposhiho reviewed Jul 5, 2023

View reviewed changes

carlory reviewed Jul 7, 2023

View reviewed changes

sanposhiho reviewed Jul 8, 2023

View reviewed changes

pohly force-pushed the dra-scheduler-queueing-hints branch from 09496db to 28d3ae4 Compare July 8, 2023 11:54

sanposhiho reviewed Jul 8, 2023

View reviewed changes

pohly force-pushed the dra-scheduler-queueing-hints branch from 28d3ae4 to e76fcab Compare July 8, 2023 15:04

sanposhiho reviewed Jul 9, 2023

View reviewed changes

k8s-ci-robot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Jul 9, 2023

sanposhiho reviewed Jul 9, 2023

View reviewed changes

carlory reviewed Jul 10, 2023

View reviewed changes

kerthcet reviewed Jul 10, 2023

View reviewed changes

pohly force-pushed the dra-scheduler-queueing-hints branch from e76fcab to fd164e3 Compare July 10, 2023 14:46

k8s-ci-robot removed the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Jul 10, 2023

k8s-ci-robot requested a review from sanposhiho July 10, 2023 14:46

sanposhiho mentioned this pull request Jul 12, 2023

nodeaffinity: scheduler queueing hints #119155

Merged

k8s-ci-robot added the needs-rebase Indicates a PR cannot be merged because it has merge conflicts with HEAD. label Jul 12, 2023

pohly added 3 commits July 12, 2023 11:07

scheduler dynamicresources: minor logging improvements

ef48efc

This makes some complex values a bit more readable.

pohly force-pushed the dra-scheduler-queueing-hints branch from fd164e3 to 6f1a295 Compare July 12, 2023 09:17

k8s-ci-robot removed the needs-rebase Indicates a PR cannot be merged because it has merge conflicts with HEAD. label Jul 12, 2023

sanposhiho approved these changes Jul 12, 2023

View reviewed changes

k8s-ci-robot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Jul 12, 2023

k8s-ci-robot added the approved Indicates a PR has been approved by an approver from all required OWNERS files. label Jul 12, 2023

k8s-ci-robot merged commit 529eeb7 into kubernetes:master Jul 12, 2023

k8s-ci-robot added this to the v1.28 milestone Jul 12, 2023

alculquicondor mentioned this pull request Jul 12, 2023

change the QueueingHintFn to pass a logger #119242

Merged

klueska mentioned this pull request Jul 19, 2023

DRA: control plane controller ("classic DRA") kubernetes/enhancements#3063

Closed

42 tasks

jaskaransarkaria mentioned this pull request Jun 5, 2024

Planning upgrade to EKS 1.28 ministryofjustice/cloud-platform#5570

Closed

	// that change made a previously unschedulable pod schedulable. It errs on the side of
	// that change made a previous unschedulable pod schedulable. It errs on the side of

	// isSchedulableAfterClaimChange is invoked whenever a claim changed. It checks whether
	// isSchedulableAfterClaimChange is invoked whenever a claim added or changed. It checks whether

	// pobject stored in it will be converted instead.
	// pobject stored in it will be converted instead.
	// pod object stored in it will be converted instead.

dra: scheduler queueing hints #119078

dra: scheduler queueing hints #119078

Conversation

pohly commented Jul 4, 2023

What type of PR is this?

What this PR does / why we need it:

Which issue(s) this PR fixes:

Special notes for your reviewer:

Does this PR introduce a user-facing change?

Additional documentation e.g., KEPs (Kubernetes Enhancement Proposals), usage docs, etc.:

k8s-ci-robot commented Jul 4, 2023

pohly commented Jul 4, 2023

sanposhiho left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

pohly commented Jul 8, 2023

sanposhiho left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

pohly commented Jul 8, 2023

sanposhiho left a comment • edited Loading

Choose a reason for hiding this comment

k8s-ci-robot commented Jul 9, 2023

sanposhiho commented Jul 9, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kerthcet commented Jul 10, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

pohly commented Jul 10, 2023

sanposhiho commented Jul 12, 2023

sanposhiho left a comment

Choose a reason for hiding this comment

k8s-ci-robot commented Jul 12, 2023

sanposhiho left a comment •

edited

Loading