*: fix scheduling can not immediately start after transfer leader #4875

rleungx · 2022-04-29T02:59:48Z

Signed-off-by: Ryan Leung rleungx@gmail.com

What problem does this PR solve?

Issue Number: Close #4769.

What is changed and how does it work?

This PR uses the flag to distinguish if the region's heartbeat is new when transferring the PD leader.

Check List

Tests

Unit test

Release note

Fix the issue that scheduling cannot immediately start after PD leader transfers

ti-chi-bot · 2022-04-29T02:59:49Z

[REVIEW NOTIFICATION]

This pull request has been approved by:

AndreMouche
JmPotato

To complete the pull request process, please ask the reviewers in the list to review by filling /cc @reviewer in the comment.
After your PR has acquired the required number of LGTMs, you can assign this pull request to the committer in the list by filling /assign @committer in the comment to help you merge this pull request.

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

Reviewer can indicate their review by submitting an approval review.
Reviewer can cancel approval by submitting a request changes review.

codecov · 2022-04-29T07:10:38Z

Codecov Report

Merging #4875 (54e8fc1) into master (562586c) will decrease coverage by 0.04%.
The diff coverage is 83.33%.

❗ Current head 54e8fc1 differs from pull request most recent head f297bf2. Consider uploading reports for the commit f297bf2 to get more accurate results

@@            Coverage Diff             @@
##           master    #4875      +/-   ##
==========================================
- Coverage   75.45%   75.40%   -0.05%     
==========================================
  Files         298      297       -1     
  Lines       29532    29506      -26     
==========================================
- Hits        22282    22250      -32     
- Misses       5318     5328      +10     
+ Partials     1932     1928       -4

Flag	Coverage Δ
unittests	`75.40% <83.33%> (-0.05%)`	⬇️

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

Impacted Files	Coverage Δ
server/cluster/cluster.go	`83.30% <33.33%> (-1.39%)`	⬇️
server/region_syncer/client.go	`85.82% <50.00%> (+0.10%)`	⬆️
server/cluster/coordinator.go	`72.74% <100.00%> (-0.92%)`	⬇️
server/cluster/prepare_checker.go	`100.00% <100.00%> (ø)`
server/core/region.go	`91.17% <100.00%> (+0.05%)`	⬆️
server/core/region_option.go	`82.85% <100.00%> (+0.33%)`	⬆️
server/grpc_service.go	`53.00% <100.00%> (ø)`
server/schedulers/shuffle_hot_region.go	`55.55% <0.00%> (-10.11%)`	⬇️
server/id/id.go	`76.19% <0.00%> (-4.77%)`	⬇️
server/member/member.go	`64.21% <0.00%> (-3.16%)`	⬇️
... and 24 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7088c95...f297bf2. Read the comment docs.

server/cluster/cluster.go

JmPotato

LGTM. BTW, do we need to triage this issue first to see which version it affects before merging it?

rleungx · 2022-05-16T06:26:18Z

LGTM. BTW, do we need to triage this issue first to see which version it affects before merging it?

Done

AndreMouche · 2022-05-16T06:32:55Z

server/cluster/cluster.go

@@ -746,6 +746,11 @@ func (c *RaftCluster) processReportBuckets(buckets *metapb.Buckets) error {
 	return nil
 }

+// IsPrepared return if the prepare checker is ready.


Suggested change

// IsPrepared return if the prepare checker is ready.

// IsPrepared returns true if the prepare checker is ready.

AndreMouche · 2022-05-16T07:03:15Z

server/cluster/cluster.go

@@ -812,7 +817,7 @@ func (c *RaftCluster) processRegionHeartbeat(region *core.RegionInfo) error {
 		regionEventCounter.WithLabelValues("update_cache").Inc()
 	}

-	if isNew {
+	if !c.IsPrepared() && isNew {


Could we check the condition in the function collect directly?

AndreMouche · 2022-05-16T07:08:09Z

server/cluster/coordinator.go

@@ -310,6 +311,9 @@ func (c *coordinator) runUntilStop() {
 }

 func (c *coordinator) run() {
+	failpoint.Inject("runSchedulerCheckInterval", func() {


Could we change the Ticker instead of runSchedulerCheckInterval? So we can keep runSchedulerCheckInterval as a constant.
For example , re define the ticker after L317

failpoint.Inject("runSchedulerCheckInterval", func() { ticker = time.NewTicker(100 * time.Millisecond) })

ti-chi-bot · 2022-05-17T09:10:48Z

@rleungx: Your PR was out of date, I have automatically updated it for you.

At the same time I will also trigger all tests for you:

/run-all-tests

If the CI test fails, you just re-trigger the test that failed and the bot will merge the PR for you after the CI passes.

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 ti-community-infra/tichi repository.