perf: Improve resmap performance with AppendAll and Transform functions #5427
Conversation
|
This PR has multiple commits, and the default merge method is: merge. 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. |
k8s-ci-robot
added
the
do-not-merge/work-in-progress
k8s-ci-robot
added
the
cncf-cla: yes
chlunde
changed the title
|
Hi @chlunde. Thanks for your PR. I'm waiting for a kubernetes-sigs member to verify that this patch is reasonable to test. If it is, they should reply with Once the patch is verified, the new status will be reflected by the 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. |
k8s-ci-robot
added
needs-ok-to-test
chlunde
changed the title
|
@annasong20 would you like this as one PR, or do you prefer to split some of the commits into different three PRs? I've attempted to make it easier to remove by having a logical story if you start at the first commit, except for perf: improve applyOrdering by avoid call to GetByCurrentId which is actually another PR that this depends on. |
|
The Kubernetes project currently lacks enough contributors to adequately respond to all PRs. This bot triages PRs according to the following rules:
You can:
Please send feedback to sig-contributor-experience at kubernetes/community. /lifecycle stale |
k8s-ci-robot
added
the
lifecycle/stale
|
/remove-lifecycle stale |
k8s-ci-robot
removed
the
lifecycle/stale
k8s-ci-robot
added
the
needs-rebase
Introduce a new function, Transform, which can be used to safely modify entries an resmap. The callback function is called for each entry in the map and handles ID conflict detection, which was the most expensive operation causing significant performance issues. The ID cache is local to this function as we cannot guarantee that all mutations are performed using this function.
…tion This avoids expensive calls to GetMatchingResourcesByCurrentId
Introduce a new function, AppendMany, which can be used to safely add entries to a resmap. The function handles ID conflict detection, which was the most expensive operation causing significant performance issues. The ID cache is local to this function as we cannot guarantee that all mutations are performed using this function.
This shaves of 14 seconds (one third) of the execution time for a
kustomization tree with 4000 documents, from 40.68s to 27.41s
0 0% 5.44% 18.42s 40.56% sigs.k8s.io/kustomize/api/krusty.(*Kustomizer).applySortOrder
0 0% 5.44% 18.40s 40.52% sigs.k8s.io/kustomize/api/internal/builtins.applyOrdering
before
(pprof) top20 -cum
Showing nodes accounting for 5.85s, 12.88% of 45.41s total
Dropped 622 nodes (cum <= 0.23s)
Showing top 20 nodes out of 157
flat flat% sum% cum cum%
0 0% 0% 40.68s 89.58% github.com/spf13/cobra.(*Command).Execute
0 0% 0% 40.68s 89.58% github.com/spf13/cobra.(*Command).ExecuteC
0 0% 0% 40.68s 89.58% github.com/spf13/cobra.(*Command).execute
0 0% 0% 40.68s 89.58% main.main
0 0% 0% 40.68s 89.58% runtime.main
0 0% 0% 40.68s 89.58% sigs.k8s.io/kustomize/kustomize/v5/commands/build.NewCmdBuild.func1
0 0% 0% 40.12s 88.35% sigs.k8s.io/kustomize/api/krusty.(*Kustomizer).Run
0.51s 1.12% 1.12% 33.20s 73.11% sigs.k8s.io/kustomize/api/resource.(*Resource).CurId
0 0% 1.12% 26.95s 59.35% sigs.k8s.io/kustomize/api/resmap.(*resWrangler).GetMatchingResourcesByCurrentId
0.35s 0.77% 1.89% 26.95s 59.35% sigs.k8s.io/kustomize/api/resmap.(*resWrangler).filteredById
0.07s 0.15% 2.05% 25.53s 56.22% sigs.k8s.io/kustomize/api/resmap.GetCurrentId
0 0% 2.05% 21.68s 47.74% sigs.k8s.io/kustomize/api/internal/target.(*KustTarget).MakeCustomizedResMap (inline)
0 0% 2.05% 21.68s 47.74% sigs.k8s.io/kustomize/api/internal/target.(*KustTarget).makeCustomizedResMap
0.54s 1.19% 3.24% 19.75s 43.49% sigs.k8s.io/kustomize/api/resource.(*Resource).GetGvk (inline)
1s 2.20% 5.44% 19.21s 42.30% sigs.k8s.io/kustomize/kyaml/resid.GvkFromNode
0 0% 5.44% 18.42s 40.56% sigs.k8s.io/kustomize/api/internal/builtins.(*SortOrderTransformerPlugin).Transform
0 0% 5.44% 18.42s 40.56% sigs.k8s.io/kustomize/api/krusty.(*Kustomizer).applySortOrder
0 0% 5.44% 18.40s 40.52% sigs.k8s.io/kustomize/api/internal/builtins.applyOrdering
0.87s 1.92% 7.36% 16.55s 36.45% sigs.k8s.io/kustomize/kyaml/yaml.visitMappingNodeFields
2.51s 5.53% 12.88% 15.68s 34.53% sigs.k8s.io/kustomize/kyaml/yaml.visitFieldsWhileTrue
after
(pprof) top20 -cum
Showing nodes accounting for 1.23s, 3.85% of 31.98s total
Dropped 584 nodes (cum <= 0.16s)
Showing top 20 nodes out of 184
flat flat% sum% cum cum%
0 0% 0% 27.41s 85.71% github.com/spf13/cobra.(*Command).Execute
0 0% 0% 27.41s 85.71% github.com/spf13/cobra.(*Command).ExecuteC
0 0% 0% 27.41s 85.71% github.com/spf13/cobra.(*Command).execute
0 0% 0% 27.41s 85.71% main.main
0 0% 0% 27.41s 85.71% runtime.main
0 0% 0% 27.41s 85.71% sigs.k8s.io/kustomize/kustomize/v5/commands/build.NewCmdBuild.func1
0 0% 0% 26.85s 83.96% sigs.k8s.io/kustomize/api/krusty.(*Kustomizer).Run
0 0% 0% 22.07s 69.01% sigs.k8s.io/kustomize/api/internal/target.(*KustTarget).MakeCustomizedResMap (inline)
0 0% 0% 22.07s 69.01% sigs.k8s.io/kustomize/api/internal/target.(*KustTarget).makeCustomizedResMap
0.38s 1.19% 1.19% 20.69s 64.70% sigs.k8s.io/kustomize/api/resource.(*Resource).CurId
0 0% 1.19% 13.64s 42.65% sigs.k8s.io/kustomize/api/resmap.(*resWrangler).Append
0 0% 1.19% 13.55s 42.37% sigs.k8s.io/kustomize/api/resmap.(*resWrangler).GetMatchingResourcesByCurrentId (inline)
0.12s 0.38% 1.56% 13.55s 42.37% sigs.k8s.io/kustomize/api/resmap.(*resWrangler).filteredById
0.01s 0.031% 1.59% 12.67s 39.62% sigs.k8s.io/kustomize/api/resmap.GetCurrentId
0.21s 0.66% 2.25% 12.49s 39.06% sigs.k8s.io/kustomize/api/resource.(*Resource).GetGvk (inline)
0.51s 1.59% 3.85% 12.28s 38.40% sigs.k8s.io/kustomize/kyaml/resid.GvkFromNode
0 0% 3.85% 11.52s 36.02% sigs.k8s.io/kustomize/api/internal/target.(*KustTarget).IgnoreLocal
0 0% 3.85% 10.53s 32.93% sigs.k8s.io/kustomize/api/internal/target.(*KustTarget).AccumulateTarget
0 0% 3.85% 10.53s 32.93% sigs.k8s.io/kustomize/api/internal/target.(*KustTarget).accumulateResources
0 0% 3.85% 10.53s 32.93% sigs.k8s.io/kustomize/api/internal/target.(*KustTarget).accumulateTarget
…ice/appendAll Append in a loop is very expensive, O(N*M). By using AppendMany we get closer to O(N+M) instead.
Reduce calls to CurId when sorting, and use AppendMany when collecting results.
Regenerate plugins make generate-kustomize-builtin-plugins gofmt -s -w -l plugin/builtin/sortordertransformer/SortOrderTransformer.go etc Possibly split in two, one for appendmany another for transform Run benchmark on both
chlunde
force-pushed
the
resmap-transform
branch
from
ab82018
to
01e154a
Compare
k8s-ci-robot
removed
the
needs-rebase
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: chlunde The full list of commands accepted by this bot can be found here.
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
|
The Kubernetes project currently lacks enough contributors to adequately respond to all PRs. This bot triages PRs according to the following rules:
You can:
Please send feedback to sig-contributor-experience at kubernetes/community. /lifecycle stale |
k8s-ci-robot
added
the
lifecycle/stale
|
/remove-lifecycle stale
/auto-cc
…On Mon, Jul 15, 2024, 4:23 p.m. Kubernetes Triage Robot < ***@***.***> wrote:
The Kubernetes project currently lacks enough contributors to adequately
respond to all PRs.
This bot triages PRs according to the following rules:
- After 90d of inactivity, lifecycle/stale is applied
- After 30d of inactivity since lifecycle/stale was applied,
lifecycle/rotten is applied
- After 30d of inactivity since lifecycle/rotten was applied, the PR
is closed
You can:
- Mark this PR as fresh with /remove-lifecycle stale
- Close this PR with /close
- Offer to help out with Issue Triage
<https://www.kubernetes.dev/docs/guide/issue-triage/>
Please send feedback to sig-contributor-experience at kubernetes/community
<https://github.com/kubernetes/community>.
/lifecycle stale
—
Reply to this email directly, view it on GitHub
<#5427 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAFKQYME3E5I422CCVKRGXDZMQVVZAVCNFSM6AAAAAA6WWTZC2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMRZGMZTKMZZGA>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
|
k8s-ci-robot
removed
the
lifecycle/stale
|
@stormqueen1990 |
|
@koba1t yes, absolutely!
|
|
Hi there, @chlunde! 👋🏻 I noticed this PR is still in draft state. Would you like to have it be reviewed again? |
|
I tried to run CI on GitHub Actions. |
|
The Kubernetes project currently lacks enough contributors to adequately respond to all PRs. This bot triages PRs according to the following rules:
You can:
Please send feedback to sig-contributor-experience at kubernetes/community. /lifecycle stale |
k8s-ci-robot
added
the
lifecycle/stale
Personal notes, keeping as draft :)
Rebased, since the SortOrderTransformer PR is in
With the benchmark suite, this PR reduces the runtime from 48s to 2.5s on my machine.
I think we could split it into multiple independent PRs now. I've benchmarked them on their own to see where the major effect
perf: ResAccumulator: Use map for resmap intersection
standalone. 48s to 47s for the benchmark.
Transform
pair of PRs, 48s to 47s for the benchmark
AppendMany
commit f5cdc84
perf: resmap: Add AppendMany function
no effect on its own
commit e06a969
perf: improve applyOrdering by avoid call to GetByCurrentId
little effect on its own
commit f6ed5d5
perf: Improve SortOrderTransformer and use AppendMany
commit 0fe5421
perf: Use AppendMany for SortOrderTransformer/newResMapFromResourceSlice/appendAll
should split this. Here's the major change 48s->2.5s Could split this into more pieces.