Use exponential backoff for failing migrations #8530

acardace · 2022-09-27T13:26:05Z

What this PR does / why we need it:

When migrating a VM fails an increasingly exponential backoff will be applied before retrying.

Which issue(s) this PR fixes (optional, in fixes #<issue number>(, fixes #<issue_number>, ...) format, will close the issue(s) when PR gets merged):
Fixes #
https://bugzilla.redhat.com/show_bug.cgi?id=2124528

Special notes for your reviewer:

Release note:

Use exponential backoff for failing migrations

acardace · 2022-09-27T13:31:15Z

/uncc @maiqueb
/uncc @marceloamaral

acardace · 2022-09-27T13:31:33Z

/cc @vladikr
/cc @xpivarc
/cc @jean-edouard

acardace · 2022-09-27T13:32:20Z

Can you guys give a look and share thoughts about the approach/default values etc?

I still have to implement unit and e2e tests.

larrydewey · 2022-10-05T14:13:03Z

I will take a look at this today.

jean-edouard · 2022-10-05T14:23:59Z

/retest

pkg/virt-controller/watch/migration-backoff.go

pkg/virt-controller/watch/migration-backoff_test.go

kubevirt-bot · 2022-10-06T16:49:11Z

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: jean-edouard

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:

~~OWNERS~~ [jean-edouard]

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

enp0s3

@acardace Hi, PR looks great! I have a few questions, please see below

enp0s3 · 2022-11-08T20:41:25Z

pkg/virt-controller/watch/migration.go

@@ -1113,6 +1170,12 @@ func (c *MigrationController) sync(key string, migration *virtv1.VirtualMachineI
 			return c.handlePreHandoffMigrationCancel(migration, vmi, pod)
 		}

+		if err = c.handleMigrationBackoff(key, vmi, migration); err != nil {


we can have an err return if the c.listMigrationsMatchingVMI(vmi.Namespace, vmi.Name) fails, in that case the error type is not MigrationBackoffReason, we will need to check the returned error type before we generate an event

Good catch!

fixed, thank you for spotting that!

enp0s3 · 2022-11-08T21:07:29Z

pkg/virt-controller/watch/migration.go

+		return metav1.Time{}
+	}
+
+	outOffBackoffTS := getFailedTS(migrations[1]).Add(backoff)


@xpivarc Hi, you mean to use status.migrtaionState.endTimestamp instead of PhaseTransitionTimestamp?

enp0s3 · 2022-11-08T21:08:12Z

tests/migration_test.go

+			deleteEvents(eventListOpts)
+		})
+
+		It("[Serial] after a successful migration backoff should be cleared", func() {


Why wouldn't run it in parallel?

We are deleting events. While with some effort we can make this parallel.

I'm just sticking to the existing functions.

enp0s3 · 2022-11-09T13:20:20Z

pkg/virt-controller/watch/migration.go

+		return metav1.Time{}
+	}
+
+	outOffBackoffTS := getFailedTS(migrations[1]).Add(backoff)


@acardace Can we address the case of a cancelled migration?

enp0s3 · 2022-11-09T13:20:39Z

pkg/virt-controller/watch/migration.go

@@ -1112,6 +1172,11 @@ func (c *MigrationController) sync(key string, migration *virtv1.VirtualMachineI
 		if migration.DeletionTimestamp != nil {
 			return c.handlePreHandoffMigrationCancel(migration, vmi, pod)
 		}
+		if err = c.handleMigrationBackoff(key, vmi, migration); errors.Is(err, migrationBackoffError) {


acardace · 2022-11-09T15:38:01Z

@enp0s3 isn't the canceled migration case handled at

kubevirt/pkg/virt-controller/watch/migration.go

Line 1172 in c0e4a6c

if migration.DeletionTimestamp != nil {

?

acardace · 2022-11-14T10:08:00Z

/retest-required

With this patch when migrating a VM fails an increasingly exponential backoff will be applied before retrying. Signed-off-by: Antonio Cardace <[email protected]>

enp0s3 · 2022-11-14T16:33:18Z

/lgtm
/unhold

@acardace Awesome, thanks for addressing the comments!

kubevirt-bot · 2022-11-14T23:47:30Z

@acardace: The following tests 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-kubevirt-e2e-k8s-1.23-operator-nonroot	`544e68e`	link	true	`/test pull-kubevirt-e2e-k8s-1.23-operator-nonroot`
pull-kubevirt-e2e-k8s-1.25-sig-network	`58ad400`	link	false	`/test pull-kubevirt-e2e-k8s-1.25-sig-network`

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.

acardace · 2022-11-15T05:36:41Z

/cherrypick release-0.58

kubevirt-bot · 2022-11-15T05:37:24Z

@acardace: new pull request created: #8784

In response to this:

/cherrypick release-0.58

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.

rmohr · 2022-11-16T10:50:41Z

Following the bugzilla discussion, I am curious on why we did not solve the pod accumulation and instead went for a back-off.

Here some thoughts where I would be happy to hear your thought-line:

I could not identify a maximum back-off, do you rely in the gc for migration objects? Let's suppose the backoff kicked in 65 times (next retry would be in 10 minutes), I resolve a node problem, and for 10 minutes nothing would happen and delay my upgrade work? I would suggest to not rely on the garbage collection and add clear limits.
What if the migration goes to another node? Would we try it faster?
It feels slightly out-of-band with the default k8s behaviour. We also don't say: "user A posted now 10 times the same failing VM yaml, let's back-off". I know that maintenance with drains is tough. How fast does it retry? Did you consider it to retry too fast, compared to this back-off and just GCing the pods as insufficient?

xpivarc · 2022-11-16T14:10:44Z

Stating for others looking into this:

Following the bugzilla discussion, I am curious on why we did not solve the pod accumulation and instead went for a back-off.

Here some thoughts where I would be happy to hear your thought-line:
* I could not identify a maximum back-off, do you rely in the gc for migration objects? Let's suppose the backoff kicked in 65 times (next retry would be in 10 minutes), I resolve a node problem, and for 10 minutes nothing would happen and delay my upgrade work? I would suggest to not rely on the garbage collection and add clear limits.

Maximum backoff is 20 sec * 2 ^ (n-1) where n is const defaultFinalizedMigrationGarbageCollectionBuffer = 5. So we can expose this and make it tunable.

The second part. Right now the backoff would be applied even if you would fix the problem. Here we could only recommend a workaround - remove the last VMIM object.

* What if the migration goes to another node? Would we try it faster?

No. As of now we can't even know if the migration is targeted to another node.

* It feels slightly out-of-band with the default k8s behaviour. We also don't say: "user A posted now 10 times the same failing VM yaml, let's back-off".  I know that maintenance with drains is tough. How fast does it retry? Did you consider it to retry too fast, compared to this back-off and just GCing the pods as insufficient?

Eviction is often retried continuously. We can try to look into suggesting backoff to clients but they could still ignore the suggestion. Even with the pod GC the problem would remain. It seems reasonable to constrain ourselves only to VMIM created by eviction for now and revisit the overall backoff later.

acardace · 2022-11-18T12:10:49Z

Following the bugzilla discussion, I am curious on why we did not solve the pod accumulation and instead went for a back-off.

Here some thoughts where I would be happy to hear your thought-line:
* I could not identify a maximum back-off, do you rely in the gc for migration objects? Let's suppose the backoff kicked in 65 times (next retry would be in 10 minutes), I resolve a node problem, and for 10 minutes nothing would happen and delay my upgrade work? I would suggest to not rely on the garbage collection and add clear limits.

Yes it probably makes sense to make it explicit in code, the limit as Lubo said is implicit and governed by the migration object GC, it anyway is 320 seconds.

* What if the migration goes to another node? Would we try it faster?

No, but what if it's the source node that has issues? not sure if it makes sense to tweak it like this.

* It feels slightly out-of-band with the default k8s behaviour. We also don't say: "user A posted now 10 times the same failing VM yaml, let's back-off".  I know that maintenance with drains is tough. How fast does it retry? Did you consider it to retry too fast, compared to this back-off and just GCing the pods as insufficient?

stu-gott · 2022-11-30T13:06:45Z

/cherry-pick release-0.53

kubevirt-bot · 2022-11-30T13:07:25Z

@stu-gott: #8530 failed to apply on top of branch "release-0.53":

Applying: migration: use exponential backoff for failing migrations
Using index info to reconstruct a base tree...
M	pkg/virt-controller/watch/migration.go
M	pkg/virt-controller/watch/migration_test.go
M	pkg/virt-controller/watch/vmi.go
M	staging/src/kubevirt.io/api/core/v1/types.go
M	tests/migration_test.go
Falling back to patching base and 3-way merge...
Auto-merging tests/migration_test.go
CONFLICT (content): Merge conflict in tests/migration_test.go
Auto-merging staging/src/kubevirt.io/api/core/v1/types.go
Auto-merging pkg/virt-controller/watch/vmi.go
Auto-merging pkg/virt-controller/watch/migration_test.go
CONFLICT (content): Merge conflict in pkg/virt-controller/watch/migration_test.go
Auto-merging pkg/virt-controller/watch/migration.go
CONFLICT (content): Merge conflict in pkg/virt-controller/watch/migration.go
error: Failed to merge in the changes.
hint: Use 'git am --show-current-patch=diff' to see the failed patch
Patch failed at 0001 migration: use exponential backoff for failing migrations
When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".

In response to this:

/cherry-pick release-0.53

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.

kubevirt-bot added release-note Denotes a PR that will be considered when it comes time to generate release notes. dco-signoff: yes Indicates the PR's author has DCO signed all their commits. size/L labels Sep 27, 2022

kubevirt-bot requested review from maiqueb and marceloamaral September 27, 2022 13:26

acardace marked this pull request as draft September 27, 2022 13:30

kubevirt-bot added the do-not-merge/work-in-progress Indicates that a PR should not merge because it is a work in progress. label Sep 27, 2022

kubevirt-bot removed request for marceloamaral and maiqueb September 27, 2022 13:31

kubevirt-bot requested review from jean-edouard, vladikr and xpivarc September 27, 2022 13:31

acardace force-pushed the failed-migration-backoff branch 2 times, most recently from b8605a4 to efad57d Compare October 3, 2022 15:30

acardace marked this pull request as ready for review October 3, 2022 15:32

kubevirt-bot removed the do-not-merge/work-in-progress Indicates that a PR should not merge because it is a work in progress. label Oct 3, 2022

kubevirt-bot requested review from marceloamaral and stu-gott October 3, 2022 15:32

acardace force-pushed the failed-migration-backoff branch from efad57d to 3cce3f1 Compare October 4, 2022 08:47

acardace force-pushed the failed-migration-backoff branch from 3cce3f1 to 544e68e Compare October 5, 2022 15:53

jean-edouard reviewed Oct 5, 2022

View reviewed changes

pkg/virt-controller/watch/migration-backoff.go Outdated Show resolved Hide resolved

pkg/virt-controller/watch/migration-backoff_test.go Outdated Show resolved Hide resolved

pkg/virt-controller/watch/migration-backoff_test.go Outdated Show resolved Hide resolved

acardace force-pushed the failed-migration-backoff branch from 544e68e to 4ac36d0 Compare October 6, 2022 16:40

kubevirt-bot added the approved Indicates a PR has been approved by an approver from all required OWNERS files. label Oct 6, 2022

acardace force-pushed the failed-migration-backoff branch from 4ac36d0 to 05e8997 Compare October 10, 2022 09:10

kubevirt-bot added the lgtm Indicates that a PR is ready to be merged. label Nov 8, 2022

acardace requested review from jean-edouard and removed request for vladikr, marceloamaral and stu-gott November 8, 2022 13:31

enp0s3 reviewed Nov 8, 2022

View reviewed changes

acardace force-pushed the failed-migration-backoff branch from 50a96a0 to c0e4a6c Compare November 9, 2022 11:06

kubevirt-bot removed the lgtm Indicates that a PR is ready to be merged. label Nov 9, 2022

enp0s3 reviewed Nov 9, 2022

View reviewed changes

acardace force-pushed the failed-migration-backoff branch from c0e4a6c to 65eece2 Compare November 10, 2022 16:08

acardace force-pushed the failed-migration-backoff branch from 65eece2 to 12c8c50 Compare November 14, 2022 15:52

migration: use exponential backoff for failing migrations

58ad400

With this patch when migrating a VM fails an increasingly exponential backoff will be applied before retrying. Signed-off-by: Antonio Cardace <[email protected]>

acardace force-pushed the failed-migration-backoff branch from 12c8c50 to 58ad400 Compare November 14, 2022 16:30

kubevirt-bot removed the do-not-merge/hold Indicates that a PR should not merge because someone has issued a /hold command. label Nov 14, 2022

kubevirt-bot assigned enp0s3 Nov 14, 2022

kubevirt-bot added the lgtm Indicates that a PR is ready to be merged. label Nov 14, 2022

enp0s3 approved these changes Nov 14, 2022

View reviewed changes

kubevirt-bot merged commit 6701a21 into kubevirt:main Nov 15, 2022

kubevirt-bot mentioned this pull request Nov 15, 2022

[release-0.58] Use exponential backoff for failing migrations #8784

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use exponential backoff for failing migrations #8530

Use exponential backoff for failing migrations #8530

acardace commented Sep 27, 2022 •

edited

Loading

acardace commented Sep 27, 2022

acardace commented Sep 27, 2022

acardace commented Sep 27, 2022

larrydewey commented Oct 5, 2022

jean-edouard commented Oct 5, 2022

kubevirt-bot commented Oct 6, 2022

enp0s3 left a comment

enp0s3 Nov 8, 2022

xpivarc Nov 9, 2022

acardace Nov 9, 2022

enp0s3 Nov 8, 2022

enp0s3 Nov 8, 2022

xpivarc Nov 9, 2022

acardace Nov 9, 2022

enp0s3 Nov 9, 2022

enp0s3 Nov 9, 2022

acardace commented Nov 9, 2022

acardace commented Nov 14, 2022

enp0s3 commented Nov 14, 2022

kubevirt-bot commented Nov 14, 2022

acardace commented Nov 15, 2022

kubevirt-bot commented Nov 15, 2022

rmohr commented Nov 16, 2022

xpivarc commented Nov 16, 2022

acardace commented Nov 18, 2022

stu-gott commented Nov 30, 2022

kubevirt-bot commented Nov 30, 2022

Use exponential backoff for failing migrations #8530

Use exponential backoff for failing migrations #8530

Conversation

acardace commented Sep 27, 2022 • edited Loading

acardace commented Sep 27, 2022

acardace commented Sep 27, 2022

acardace commented Sep 27, 2022

larrydewey commented Oct 5, 2022

jean-edouard commented Oct 5, 2022

kubevirt-bot commented Oct 6, 2022

enp0s3 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

acardace commented Nov 9, 2022

acardace commented Nov 14, 2022

enp0s3 commented Nov 14, 2022

kubevirt-bot commented Nov 14, 2022

acardace commented Nov 15, 2022

kubevirt-bot commented Nov 15, 2022

rmohr commented Nov 16, 2022

xpivarc commented Nov 16, 2022

acardace commented Nov 18, 2022

stu-gott commented Nov 30, 2022

kubevirt-bot commented Nov 30, 2022

acardace commented Sep 27, 2022 •

edited

Loading