store/mockstore: Some wg.Wait() operations may block forever

[BurtonQin]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

We found 2 blocking bugs through a WIP Fuzzing project by system-pclub, PSU.

2. What did you expect to see? (Required)

The wg.Wait() operation shall not block forever whatever the order of the concurrent messages.

3. What did you see instead (Required)

We found the 2 bugs by fuzzing the code from entry point:

tidb/ddl/ddl_test.go

Line 60 in fa71a0e

func TestT(t *testing.T) {

1. Blocked at Wait() in function acquireOne at
   

   tidb/store/mockstore/unistore/tikv/region.go

   Lines 107 to 111 in fa71a0e

   	if ok {
   	w.Wait()
   	waitCnt++
   	continue
   	}

2. Blocked at Wait() in function Write at
   

   tidb/store/mockstore/unistore/tikv/write.go

   Lines 227 to 233 in fa71a0e

   	if len(wb.lockBatch.entries) > 0 {
   	// We must delete lock after commit succeed, or there will be inconsistency.
   	wb.lockBatch.wg.Add(1)
   	writer.lockCh <- &wb.lockBatch
   	wb.lockBatch.wg.Wait()
   	return wb.lockBatch.err
   	}

4. What is your TiDB version? (Required)

branch: master
commit: 7e6690d

[ChenPeng2013]

@sticnarf PTAL, is it a bug, thanks

[charlesxsh]

Hi there, kindly ask do we have any update on this PR?

[sticnarf]

I don't have ideas about how this really happens at this time...

The mockstore is only used in tests instead of production. Probably these forever wg.Wait() are due to incomplete resource cleanups after tests.

Sorry, I'm afraid we will not spend much time investigating the issue unless it greatly affects daily development by decreasing the test success rate.

[jebter]

The issue has not been updated for too long, so I will close it. If there are any updates, you can reopen it.