TiCDC does not resolve lock in time

[overvenus]

Bug Report

Please answer these questions before submitting your issue. Thanks!

• What did you do? If possible, provide a recipe for reproducing the error.

1. Create table A and B. Create a changefeed that captures A.
2. Kill TiDB instances and leave some locks in B.
3. Wait 10 minutes.
4. Merge regions a (belongs table A) and region b (belongs table B).

• What did you expect to see?

Changefeed is not blocked and locks in B can be resolved by TiCDC.

• What did you see instead?

Changefeed is blocked. And lots of fallen back logs.

[2021/09/17 09:57:25.081 +08:00] [WARN] [region_worker.go:690] ["The resolvedTs is fallen back in kvclient"] ["Event Type"=RESOLVED] [resolvedTs=427759786830332014] [lastResolvedTs=427767830928687109] [regionID=478066658]

Also, we see TiKV resolved ts jump, because of region merge.

• Versions of the cluster

  v5.1.1

[amyangfei]

Whether there exists resolve lock log for regionID=478066658? @overvenus

[amyangfei]

Besides, whether kv client reconnect triggered happens?

[overvenus]

Whether there exists resolve lock log for regionID=478066658? @overvenus

No.

whether kv client reconnect

No.

[maxshuang]

@overvenus why need to wait 10 min? If tidb instance down, prewrite lock can be gc in no situation? so we don't need to wait?