store: upgrade the CheckTxnStatus API #13123
Conversation
CheckTxnStatus introduces a non-block read mode. In this mode, TiDB can ignore the secondary lock TTL check and send the CheckTxnStatus request.
Codecov Report
@@ Coverage Diff @@
## master #13123 +/- ##
================================================
- Coverage 80.8733% 80.1926% -0.6807%
================================================
Files 469 469
Lines 115012 111519 -3493
================================================
- Hits 93014 89430 -3584
- Misses 15054 15197 +143
+ Partials 6944 6892 -52 |
|
We don't need |
Removed. @coocood Note, we have no switch, so after this PR the resolve lock logic is changed. |
|
/run-all-tests |
| @@ -919,7 +919,7 @@ func rollbackKey(db *leveldb.DB, batch *leveldb.Batch, key []byte, startTS uint6 | |||
| return nil | |||
| } | |||
|
|
|||
| func rollbackLock(batch *leveldb.Batch, lock mvccLock, key []byte, startTS uint64) error { | |||
There was a problem hiding this comment.
This parameter is not used in the function, so I remove it
| return lr.getTxnStatus(bo, l.TxnID, l.Primary, callerStartTS, currentTS) | ||
|
|
||
| rollbackIfNotExist := false | ||
| for { |
There was a problem hiding this comment.
There is already a for loop in getTxnStatus why not handle retry in getTxnStatus directly?
There was a problem hiding this comment.
To make the unit test easier.
| } | ||
| // Call getTxnStatusFromLock to cover the retry logic. | ||
| status, err := resolver.getTxnStatusFromLock(bo, lock, currentTS) | ||
| c.Assert(err, IsNil) |
There was a problem hiding this comment.
At this time, the lock may still not written, we can only be sure it is written after <-errCh.
There was a problem hiding this comment.
The lock may still not written, and getTxnStatusFromLock should retry in that case, so we cover the retry logic.
The lock TTL is set to 10s, long enough for the goroutine to finish, the result should be stable.
There was a problem hiding this comment.
or we add some sleep before errCh <- xxx to ensure covering the retry path
There was a problem hiding this comment.
or we add some sleep before
errCh <- xxxto ensure covering the retry path
So as to make the CI slower?
|
/run-all-tests |
|
/rebuild |
|
LGTM |
| @@ -663,22 +663,22 @@ func (s *testMVCCLevelDB) TestErrors(c *C) { | |||
| func (s *testMVCCLevelDB) TestCheckTxnStatus(c *C) { | |||
| s.mustPrewriteWithTTLOK(c, putMutations("pk", "val"), "pk", 5, 666) | |||
|
|
|||
| ttl, commitTS, err := s.store.CheckTxnStatus([]byte("pk"), 5, 0, 666) | |||
| ttl, commitTS, err := s.store.CheckTxnStatus([]byte("pk"), 5, 0, 666, false) | |||
There was a problem hiding this comment.
I think you need to cover the case where the last parameter is true.
| if _, ok := errors.Cause(err).(txnNotFoundErr); !ok { | ||
| return TxnStatus{}, err | ||
| } |
There was a problem hiding this comment.
I think here it needs comments to explain it.
|
/run-all-tests |
1 similar comment
|
/run-all-tests |
|
/run-unit-test |
|
/run-unit-test |
|
/run-all-tests |
tiancaiamao
added
status/LGT3
CheckTxnStatus introduces a non-block read mode. In this mode, TiDB can ignore the secondary lock TTL check and send the CheckTxnStatus request.
What problem does this PR solve?
CheckTxnStatus introduces a non-block read mode.
In this mode, TiDB can ignore the secondary lock TTL check and send the CheckTxnStatus request.
What is changed and how it works?
The 2PC prewrite operation writes the primary and the secondary regions concurrently.
When the CheckTxnStatus request arrives, the primary key lock may be still ongoing (secondary locks exist while the primary lock doesn't).
We can back off and retry CheckTxnStatus in that case.
noWaitparameter to the ResolveLocks() functionTxnNotFounderror and retry in getTxnStatusFromLock()RollbackIfNotExistfield of theCheckTxnStatusrequest in the mocktikvCheck List
Tests
Code changes
Side effects