affinity: add scatter filter, add more evict check and avoid statistic miss #10080

lhy1024 · 2025-12-17T08:08:40Z

What problem does this PR solve?

Issue Number: Ref #9764

What is changed and how does it work?

Check List

Tests

Unit test

Release note

None.

Signed-off-by: lhy1024 <admin@liudos.us>

codecov · 2025-12-17T08:17:22Z

Codecov Report

❌ Patch coverage is 88.23529% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 78.49%. Comparing base (35b9458) to head (6e50d3d).
⚠️ Report is 2 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master   #10080   +/-   ##
=======================================
  Coverage   78.49%   78.49%           
=======================================
  Files         515      515           
  Lines       69236    69253   +17     
=======================================
+ Hits        54344    54358   +14     
- Misses      10948    10957    +9     
+ Partials     3944     3938    -6

Flag	Coverage Δ
unittests	`78.49% <88.23%> (+<0.01%)`	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

lhy1024 · 2025-12-17T10:56:48Z

/retest

lhy1024 · 2025-12-17T11:07:57Z

/retest

rleungx · 2025-12-17T11:38:10Z

pkg/schedule/scatter/region_scatterer.go

+	// Check if region is in an affinity group that doesn't allow regular scheduling
+	if !r.affinityFilter.Select(region).IsOK() {
+		scatterSkipAffinityCounter.Inc()
+		return nil, errors.Errorf("region %d is in affinity group", region.GetID())


Will it trigger the retry?

it should be the same logic with hot region or no leader?

pkg/schedule/filter/region_filters.go

Signed-off-by: lhy1024 <admin@liudos.us>

Signed-off-by: HunDunDM <hundundm@gmail.com>

lhy1024 · 2025-12-18T04:00:46Z

@bufferflies @rleungx PTAL

lhy1024 · 2025-12-18T04:00:48Z

/retest

Signed-off-by: lhy1024 <admin@liudos.us>

lhy1024 · 2025-12-18T05:48:03Z

/retest

lhy1024 · 2025-12-18T06:00:45Z

/retest

lhy1024 · 2025-12-18T06:30:12Z

/retest

rleungx · 2025-12-18T06:30:46Z

pkg/schedule/affinity/manager.go


 // GetRegionAffinityGroupState returns the affinity group state and isAffinity for a region.
-func (m *Manager) GetRegionAffinityGroupState(region *core.RegionInfo) (group *GroupState, isAffinity bool) {
+// If skipSaveCache is not set to true, InvalidCache must be called at the appropriate time to prevent stale cache entries.


How about adding comments about when we need to update the cache?

pkg/schedule/affinity/policy.go

Signed-off-by: lhy1024 <admin@liudos.us>

bufferflies

rest lgtm

bufferflies · 2025-12-18T07:21:58Z

pkg/schedule/handler/handler.go

 	}

-	ops, err := operator.CreateMergeRegionOperator("admin-merge-region", c, region, target, operator.OpAdmin|operator.OpMerge)
+	ops, err := operator.CreateMergeRegionOperator("admin-merge-region", c, region, target, operator.OpAdmin)


Why remove the OpAdmin type?

#10076 (comment)

bufferflies · 2025-12-18T07:29:45Z

pkg/schedule/checker/affinity_checker.go

+
+	// A Region may no longer exist in the RegionTree due to a merge.
+	// In this case, clear the cache in affinity manager for that Region and skip processing it.
+	if c.cluster.GetRegion(region.GetID()) == nil {


Can it be put in line 87?

Cannot move the check before GetAndCacheRegionAffinityGroupState.

Problem with moving it earlier:

If check passes, then region is deleted, then HandleOverlaps calls InvalidCache (but cache doesn't
exist yet - no-op)

Then we save cache → stale cache leak (no cleanup path)

Current approach guarantees:

Save cache first, then check

Either HandleOverlaps (when region deleted) OR AffinityChecker (when check fails) will clean up

At least one cleanup point always executes

lhy1024 · 2025-12-18T07:40:20Z

@bufferflies @rleungx PTAL

Signed-off-by: lhy1024 <admin@liudos.us>

lhy1024 · 2025-12-18T08:22:43Z

/retest

ti-chi-bot · 2025-12-18T08:29:11Z

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: bufferflies, rleungx

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

The pull request process is described here

Details

Needs approval from an approver in each of these files:

~~OWNERS~~ [bufferflies,rleungx]

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

ti-chi-bot · 2025-12-18T08:29:14Z

[LGTM Timeline notifier]

Timeline:

2025-12-18 08:02:08.985471197 +0000 UTC m=+1719273.799248759: ☑️ agreed by rleungx.
2025-12-18 08:29:13.735428916 +0000 UTC m=+1720898.549206488: ☑️ agreed by bufferflies.

ref tikv#9764 Signed-off-by: ti-chi-bot <ti-community-prow-bot@tidb.io>

ti-chi-bot · 2025-12-18T08:41:11Z

In response to a cherrypick label: new pull request created to branch release-8.5: #10088.
But this PR has conflicts, please resolve them!

…c miss (tikv#10080) ref tikv#9764 Signed-off-by: lhy1024 <admin@liudos.us>

…c miss (tikv#10080) ref tikv#9764 Signed-off-by: lhy1024 <admin@liudos.us> Signed-off-by: HunDunDM <hundundm@gmail.com> # Conflicts: # pkg/schedule/operator/operator_controller.go

lhy1024 added 4 commits December 17, 2025 16:05

and add filter in scatter

99b86c9

Signed-off-by: lhy1024 <admin@liudos.us>

affinity: support invalid cache for missing regions

d5c6ddb

Signed-off-by: lhy1024 <admin@liudos.us>

add complete storeCondition, such as evict slow

d9bc56f

Signed-off-by: lhy1024 <admin@liudos.us>

rm TODO

aa3f3e6

Signed-off-by: lhy1024 <admin@liudos.us>

lhy1024 requested review from bufferflies and rleungx December 17, 2025 10:56

rleungx reviewed Dec 17, 2025

View reviewed changes

address comments

0560360

Signed-off-by: lhy1024 <admin@liudos.us>

lhy1024 force-pushed the affinity-pr10 branch from 0003cd2 to 0560360 Compare December 17, 2025 11:55

ti-chi-bot bot added size/L Denotes a PR that changes 100-499 lines, ignoring generated files. and removed size/XL Denotes a PR that changes 500-999 lines, ignoring generated files. labels Dec 17, 2025

lhy1024 mentioned this pull request Dec 17, 2025

Support data affinity schedule for performance optimization #9764

Open

28 tasks

address comments

77ba99e

Signed-off-by: lhy1024 <admin@liudos.us>

ti-chi-bot bot added size/XL Denotes a PR that changes 500-999 lines, ignoring generated files. and removed size/L Denotes a PR that changes 100-499 lines, ignoring generated files. labels Dec 17, 2025

HunDunDM and others added 2 commits December 18, 2025 11:34

remove InvalidCacheForMissingRegion

1849124

Signed-off-by: HunDunDM <hundundm@gmail.com>

Merge branch 'master' of github.com:tikv/pd into affinity-pr10

8b61fcc

remove opMerge in "admin-merge-region"

28d7ebc

Signed-off-by: lhy1024 <admin@liudos.us>

lhy1024 force-pushed the affinity-pr10 branch from 8b76638 to 28d7ebc Compare December 18, 2025 04:10

miss HasRelatedMergeRegion

41fb115

Signed-off-by: lhy1024 <admin@liudos.us>

rleungx reviewed Dec 18, 2025

View reviewed changes

pkg/schedule/affinity/policy.go Outdated Show resolved Hide resolved

split functions and add busy test and rename evictLeader

417c264

Signed-off-by: lhy1024 <admin@liudos.us>

bufferflies reviewed Dec 18, 2025

View reviewed changes

rename

6e50d3d

Signed-off-by: lhy1024 <admin@liudos.us>

lhy1024 force-pushed the affinity-pr10 branch from 72bfcfe to 6e50d3d Compare December 18, 2025 07:41

rleungx approved these changes Dec 18, 2025

View reviewed changes

ti-chi-bot bot added needs-1-more-lgtm Indicates a PR needs 1 more LGTM. approved labels Dec 18, 2025

bufferflies approved these changes Dec 18, 2025

View reviewed changes

ti-chi-bot bot added lgtm and removed needs-1-more-lgtm Indicates a PR needs 1 more LGTM. labels Dec 18, 2025

ti-chi-bot bot merged commit b53de7a into tikv:master Dec 18, 2025
31 checks passed

ti-chi-bot pushed a commit to ti-chi-bot/pd that referenced this pull request Dec 18, 2025

This is an automated cherry-pick of tikv#10080

2859fc1

ref tikv#9764 Signed-off-by: ti-chi-bot <ti-community-prow-bot@tidb.io>

ti-chi-bot mentioned this pull request Dec 18, 2025

affinity: add scatter filter, add more evict check and avoid statistic miss (#10080) #10088

Closed

lhy1024 mentioned this pull request Dec 18, 2025

cherrypick data affinity feature to release-8.5 #10061

Merged

lhy1024 added a commit to lhy1024/pd that referenced this pull request Dec 18, 2025

affinity: add scatter filter, add more evict check and avoid statisti…

d68e18c

…c miss (tikv#10080) ref tikv#9764 Signed-off-by: lhy1024 <admin@liudos.us>

affinity: add scatter filter, add more evict check and avoid statistic miss #10080

affinity: add scatter filter, add more evict check and avoid statistic miss #10080

Uh oh!

Conversation

lhy1024 commented Dec 17, 2025

What problem does this PR solve?

What is changed and how does it work?

Check List

Release note

Uh oh!

codecov bot commented Dec 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

lhy1024 commented Dec 17, 2025

Uh oh!

lhy1024 commented Dec 17, 2025

Uh oh!

rleungx Dec 17, 2025

Choose a reason for hiding this comment

Uh oh!

lhy1024 Dec 17, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

lhy1024 commented Dec 18, 2025

Uh oh!

lhy1024 commented Dec 18, 2025

Uh oh!

lhy1024 commented Dec 18, 2025

Uh oh!

lhy1024 commented Dec 18, 2025

Uh oh!

lhy1024 commented Dec 18, 2025

Uh oh!

rleungx Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

bufferflies left a comment

Choose a reason for hiding this comment

Uh oh!

bufferflies Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

lhy1024 Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

bufferflies Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

lhy1024 Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

lhy1024 commented Dec 18, 2025

Uh oh!

lhy1024 commented Dec 18, 2025

Uh oh!

ti-chi-bot bot commented Dec 18, 2025

Uh oh!

ti-chi-bot bot commented Dec 18, 2025

[LGTM Timeline notifier]

Uh oh!

Uh oh!

ti-chi-bot commented Dec 18, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

codecov bot commented Dec 17, 2025 •

edited

Loading