*: fix bug that the kill command doesn't work when the killed session is waiting for the pessimistic lock#12852
Conversation
|
/run-all-tests |
Codecov Report
@@ Coverage Diff @@
## master #12852 +/- ##
================================================
- Coverage 80.0759% 80.0376% -0.0384%
================================================
Files 465 465
Lines 107649 106721 -928
================================================
- Hits 86201 85417 -784
+ Misses 14967 14908 -59
+ Partials 6481 6396 -85 |
| tk.MustExec("insert test_kill values (1, 1)") | ||
| tk.MustExec("begin pessimistic") | ||
| tk2.MustExec("begin pessimistic") | ||
| tk.MustQuery("select * from test_kill where id = 1 for update") |
There was a problem hiding this comment.
How about adding a test case for MAX_EXECUTION_TIME?
There was a problem hiding this comment.
I test it manually and currently MAX_EXECUTION_TIME works when the session is waiting for a lock.
Does MySQL have different behavior? @jackysp
There was a problem hiding this comment.
MySQL has the same behavior.
store/tikv/2pc.go
Outdated
| if err1 := bo.BackoffWithMaxSleep(boTxnLockFast, int(expire), errors.New(locks[0].String())); err1 != nil { | ||
| return err1 | ||
| } |
There was a problem hiding this comment.
why do we need to add this backoff here?
There was a problem hiding this comment.
emm.... maybe I know: do you just want to check bo.ctx.Done()?
first batch is canceled by action.kill, then other batchs will use bo.ctx.Done(), batchExecutor will return until all batch be cancelled.. 🤔
There was a problem hiding this comment.
... it seems doesn't work @tiancaiamao
Line 1222 in 2ec341e
this line, we just can cancel actionPrewrite, actionPessimisticLock also need cancel?
There was a problem hiding this comment.
We can call time.Sleep or bo.Backoff, I choose backoff because it may generate better error. @lysu
There was a problem hiding this comment.
The coprocessor also use backoff like this https://github.com/pingcap/tidb/blob/master/store/tikv/coprocessor.go#L792
There was a problem hiding this comment.
normally, we think about TxnLockFast as read-conflict in metric
|
efddcf6#diff-4ef55fe1237c7e6bb2217ccca37b9f5fR439 @tiancaiamao do we need to make |
|
LGTM |
1c1f6ee to
d0525b3
Compare
The problem out of the scope of this PR means to solve. @lysu |
|
/run-all-tests |
|
/rebuild |
|
cherry pick to release-3.0 failed |
|
cherry pick to release-3.1 failed |
…ect/tidb into feature-add-udf-support * 'feature-add-udf-support' of https://github.com/JustProject/tidb: (26 commits) *: fix bug that the kill command doesn't work when the killed session is waiting for the pessimistic lock (pingcap#12852) executor: fix the projection upon the indexLookUp in indexLookUpJoin can't get result. (pingcap#12889) planner, executor: support create view on union (pingcap#12595) planner/cascades: introduce TransformationID in cascades planner (pingcap#12879) executor: fix data race in test (pingcap#12910) executor: reuse chunk row for insert on duplicate update (pingcap#12847) ddl: speed up tests (pingcap#12888) executor: speed up test (pingcap#12896) expression: implement vectorized evaluation for `builtinSecondSig` (pingcap#12886) expression: implement vectorized evaluation for `builtinJSONObjectSig` (pingcap#12663) expression: speed up builtinRepeatSig by using MergeNulls (pingcap#12674) expression: speed up unit tests under the expression package (pingcap#12887) store,kv: snapshot doesn't cache the non-exists kv entries lead to poor 'insert ignore' performance (pingcap#12872) executor: fix data race in `GetDirtyTable()` (pingcap#12767) domain: increase TTL to reduce the occurrence of reporting min startTS errors (pingcap#12578) executor: split test for speed up (pingcap#12881) executor: fix inconsistent of grants privileges with MySQL when executing `grant all on ...` (pingcap#12330) expression: implement vectorized evaluation for `builtinJSONUnquoteSig` (pingcap#12841) tune grpc connection count between tidb and tikv (pingcap#12884) Makefile: change test parallel to 8 (pingcap#12885) ...
|
@tiancaiamao need cherry pick to 3.1? |
… is waiting for the pessimistic lock (pingcap#12852)
What problem does this PR solve?
What is changed and how it works?
Fix bug that the kill command doesn't work when the killed session is waiting for the pessimistic lock.
The
killedflag is checked when the code runs toexecutor.Next(), but when the transaction is acquiring the pessimistic lock, it blocks and never runs toexecutor.Next(), so the kill command can't work.Check List
Tests
Code changes
Related changes
Release note