add `--concurrent-horizontal-pod-autoscaler-syncs` flag to kube-controller-manager #108501

zroubalik · 2022-03-03T20:10:18Z

Signed-off-by: Zbynek Roubalik zroubalik@gmail.com

What type of PR is this?

/kind feature
/kind api-change

What this PR does / why we need it:

While many other controllers like deployment have flag --concurrent-xxx-syncs to set the number of workers, the HPA controller doesn't and just hardcoded with 1 worker. It is very slow to deal with a large number of HPAs in the cluster. This PR exports a flag --concurrent-horizontal-pod-autoscaler-syncs with the default value of 5 for the HPA controller.

Which issue(s) this PR fixes:

Fixes #96242

Special notes for your reviewer:

Does this PR introduce a user-facing change?

kube-controller-manager supports '--concurrent-horizontal-pod-autoscaler-syncs' flag to set the number of horizontal pod autoscaler controller workers.

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

k8s-ci-robot · 2022-03-03T20:10:26Z

Hi @zroubalik. Thanks for your PR.

I'm waiting for a kubernetes member to verify that this patch is reasonable to test. If it is, they should reply with /ok-to-test on its own line. Until that is done, I will not automatically test new commits in this PR, but the usual testing commands by org members will still work. Regular contributors should join the org to skip this step.

Once the patch is verified, the new status will be reflected by the ok-to-test label.

I understand the commands that are listed here.

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.

pkg/controller/podautoscaler/config/v1alpha1/defaults.go

k8s-triage-robot · 2022-03-03T21:10:22Z

This PR may require API review.

If so, when the changes are ready, complete the pre-review checklist and request an API review.

Status of requested reviews is tracked in the API Review project.

yusank · 2022-03-04T07:03:19Z

You should add Mutex Lock for all maps in HorizontalController, otherwise it will be cause of concurrent map writes.

zroubalik · 2022-03-04T08:38:13Z

You should add Mutex Lock for all maps in HorizontalController, otherwise it will be cause of concurrent map writes.

Good catch, fixed in 95612b5

PTAL

fedebongio · 2022-03-08T21:04:54Z

/remove-sig api-machinery

pigletfly · 2022-03-09T14:18:17Z

/ok-to-test

mwielgus

/approve

pbetkier · 2022-09-30T10:24:38Z

@liggitt @deads2k asking for your approves.

liggitt

@zroubalik I would like to see that the feature improves the recalculation speed. There is a qps-limit that also can be a bottle-neck when using multiple HPAs.

Could you please provide some benchmarks/measurements that show the improvement? I'm asking that because the change introduces additional complexity to the algorithm and it's debbugability, so a justification that the feature is useful would be great!

Unfortunately I don't have any specific numbers, but as you can see on the linked issues, there are users facing exactly this kind of problem.

given that there are several shared objects in the controller that have to be locked and introduce contention between the workers, the benchmarks to show the actual impact of this still seems like a good idea

I also noted places where the current locking is not safely protecting against races, which also indicates more test coverage is needed

pkg/controller/podautoscaler/horizontal.go

zroubalik · 2022-10-14T15:51:11Z

I extended the test a little bit, now it also access scaleUpEvents and scaleDownEvents maps.

I run the test serveral times, it reconciles 5000 HPAs and on my machine:

with 1 worker the average test execution is ~26s
with 5 workers the average test execution is ~7.5s

PTAL

pbetkier · 2022-10-14T16:51:22Z

I run the test serveral times, it reconciles 5000 HPAs and on my machine:

with 1 worker the average test execution is ~26s

with 5 workers the average test execution is ~7.5s

Since the test stubs calls to api-server (including metrics-server) it cannot discover bottlenecks like qps-limit or other coming from how the system behaves as a whole.

I think the best way to measure the improvement would be to run a comparison in a large cluster – the promise of this feature is to make HPA reconcile faster in large clusters. Creating a large cluster for tests would be costly though. Perhaps you can emulate the performance problem in a smaller cluster, by simulating a high latency of a custom metrics backend – that's another use case where we expect to see better results after this change (#96242).

You could e.g. configure a http://github.com/Shopify/toxiproxy sitting in front of custom metrics backend, with a latency fault configured.

liggitt · 2022-10-14T18:07:00Z

needs gofmt run and the added unit test is failing:

    horizontal_test.go:4449: 
        	Error Trace:	/home/prow/go/src/k8s.io/kubernetes/_output/local/go/src/k8s.io/kubernetes/pkg/controller/podautoscaler/horizontal_test.go:4449
        	Error:      	Not equal: 
        	            	expected: 5000
        	            	actual  : 1484
        	Test:       	TestMultipleHPAs
        	Messages:   	Expected to process all HPAs

zroubalik · 2022-10-14T18:55:25Z

needs gofmt run and the added unit test is failing:

    horizontal_test.go:4449: 
        	Error Trace:	/home/prow/go/src/k8s.io/kubernetes/_output/local/go/src/k8s.io/kubernetes/pkg/controller/podautoscaler/horizontal_test.go:4449
        	Error:      	Not equal: 
        	            	expected: 5000
        	            	actual  : 1484
        	Test:       	TestMultipleHPAs
        	Messages:   	Expected to process all HPAs

Fixed, 5000 HPAs was too optimistic number for this infra, so it has fallen into timeout trap.

pkg/controller/podautoscaler/horizontal.go

…oller-manager Signed-off-by: Zbynek Roubalik <zroubalik@gmail.com>

zroubalik · 2022-10-17T18:36:00Z

/retest

liggitt · 2022-10-17T18:41:05Z

/lgtm
/approve

k8s-ci-robot · 2022-10-17T18:45:32Z

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: liggitt, mwielgus, pbetkier, zroubalik

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:

~~api/OWNERS~~ [liggitt]
~~cmd/kube-controller-manager/OWNERS~~ [liggitt]
~~pkg/controller/podautoscaler/OWNERS~~ [liggitt,mwielgus]
~~pkg/generated/openapi/OWNERS~~ [liggitt]
~~staging/src/k8s.io/kube-controller-manager/config/OWNERS~~ [liggitt]

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

…quest !873) 1.20:移植1.26支持多个HPAControlller worker add `--concurrent-horizontal-pod-autoscaler-syncs` flag to kube-controller-manager 为满足作业帮需求，移植PR：kubernetes#108501

k8s-ci-robot added needs-priority Indicates a PR lacks a `priority/foo` label and requires one. cncf-cla: yes Indicates the PR's author has signed the CNCF CLA. sig/api-machinery Categorizes an issue or PR as relevant to SIG API Machinery. labels Mar 3, 2022

k8s-ci-robot requested review from deads2k and liggitt March 3, 2022 20:10

k8s-ci-robot added sig/apps Categorizes an issue or PR as relevant to SIG Apps. sig/autoscaling Categorizes an issue or PR as relevant to SIG Autoscaling. and removed do-not-merge/needs-sig Indicates an issue or PR lacks a `sig/foo` label and requires one. labels Mar 3, 2022

zroubalik commented Mar 3, 2022

View reviewed changes

pkg/controller/podautoscaler/config/v1alpha1/defaults.go Outdated Show resolved Hide resolved

k8s-ci-robot added size/M Denotes a PR that changes 30-99 lines, ignoring generated files. and removed size/S Denotes a PR that changes 10-29 lines, ignoring generated files. labels Mar 4, 2022

k8s-ci-robot removed the sig/api-machinery Categorizes an issue or PR as relevant to SIG API Machinery. label Mar 8, 2022

k8s-ci-robot added ok-to-test Indicates a non-member PR verified by an org member that is safe to test. and removed needs-ok-to-test Indicates a PR that requires an org member to verify it is safe to test. labels Mar 9, 2022

zroubalik force-pushed the hpa branch from 95612b5 to 90e1667 Compare March 10, 2022 12:00

k8s-ci-robot added area/code-generation sig/api-machinery Categorizes an issue or PR as relevant to SIG API Machinery. labels Mar 10, 2022

mwielgus approved these changes Sep 30, 2022

View reviewed changes

liggitt self-assigned this Oct 12, 2022

liggitt added this to In progress in API Reviews Oct 13, 2022

liggitt reviewed Oct 13, 2022

View reviewed changes

pkg/controller/podautoscaler/horizontal.go Show resolved Hide resolved

pkg/controller/podautoscaler/horizontal.go Show resolved Hide resolved

liggitt moved this from In progress to Changes requested in API Reviews Oct 13, 2022

zroubalik force-pushed the hpa branch from 9696485 to 601ff9d Compare October 14, 2022 09:28

k8s-ci-robot removed the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Oct 14, 2022

zroubalik force-pushed the hpa branch from 601ff9d to 1200abe Compare October 14, 2022 15:46

zroubalik force-pushed the hpa branch from 1200abe to d9f3973 Compare October 14, 2022 18:54

liggitt reviewed Oct 17, 2022

View reviewed changes

pkg/controller/podautoscaler/horizontal.go Outdated Show resolved Hide resolved

add --concurrent-horizontal-pod-autoscaler-syncs flag to kube-contr…

1cefcde

…oller-manager Signed-off-by: Zbynek Roubalik <zroubalik@gmail.com>

zroubalik force-pushed the hpa branch from d9f3973 to 1cefcde Compare October 17, 2022 15:39

k8s-ci-robot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Oct 17, 2022

k8s-ci-robot added the approved Indicates a PR has been approved by an approver from all required OWNERS files. label Oct 17, 2022

k8s-ci-robot merged commit 85643c0 into kubernetes:master Oct 17, 2022

k8s-ci-robot added this to the v1.26 milestone Oct 17, 2022

liggitt mentioned this pull request Oct 24, 2022

Added workerCount flag to node controller in cloud controller manager #113104

Merged

liggitt moved this from Changes requested to API review completed, 1.26 in API Reviews Nov 10, 2022

ialidzhikov mentioned this pull request Dec 16, 2022

☂️-Issue for "Support for Kubernetes v1.26” gardener/gardener#6773

Closed

20 tasks

pbetkier mentioned this pull request Jun 26, 2023

Add a jitter when requeuing HPA objects #118788

Open

lanshunfang mentioned this pull request Jun 28, 2023

HPA scheduling implementation doesn't scale well with custom/external metrics - all checks are sequential by one goroutine #96242

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

add `--concurrent-horizontal-pod-autoscaler-syncs` flag to kube-controller-manager #108501

add `--concurrent-horizontal-pod-autoscaler-syncs` flag to kube-controller-manager #108501

zroubalik commented Mar 3, 2022 •

edited

k8s-ci-robot commented Mar 3, 2022

k8s-triage-robot commented Mar 3, 2022

yusank commented Mar 4, 2022

zroubalik commented Mar 4, 2022

fedebongio commented Mar 8, 2022

pigletfly commented Mar 9, 2022

mwielgus left a comment

pbetkier commented Sep 30, 2022

liggitt left a comment

zroubalik commented Oct 14, 2022

pbetkier commented Oct 14, 2022 •

edited

liggitt commented Oct 14, 2022

zroubalik commented Oct 14, 2022 •

edited

zroubalik commented Oct 17, 2022

liggitt commented Oct 17, 2022

k8s-ci-robot commented Oct 17, 2022

add --concurrent-horizontal-pod-autoscaler-syncs flag to kube-controller-manager #108501

add --concurrent-horizontal-pod-autoscaler-syncs flag to kube-controller-manager #108501

Conversation

zroubalik commented Mar 3, 2022 • edited

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 Mar 3, 2022

k8s-triage-robot commented Mar 3, 2022

yusank commented Mar 4, 2022

zroubalik commented Mar 4, 2022

fedebongio commented Mar 8, 2022

pigletfly commented Mar 9, 2022

mwielgus left a comment

Choose a reason for hiding this comment

pbetkier commented Sep 30, 2022

liggitt left a comment

Choose a reason for hiding this comment

zroubalik commented Oct 14, 2022

pbetkier commented Oct 14, 2022 • edited

liggitt commented Oct 14, 2022

zroubalik commented Oct 14, 2022 • edited

zroubalik commented Oct 17, 2022

liggitt commented Oct 17, 2022

k8s-ci-robot commented Oct 17, 2022

add `--concurrent-horizontal-pod-autoscaler-syncs` flag to kube-controller-manager #108501

add `--concurrent-horizontal-pod-autoscaler-syncs` flag to kube-controller-manager #108501

zroubalik commented Mar 3, 2022 •

edited

pbetkier commented Oct 14, 2022 •

edited

zroubalik commented Oct 14, 2022 •

edited