Details
-
Bug
-
Status: Reopened
-
Critical
-
Resolution: Unresolved
-
3.6.0
-
None
Description
There was a regression introduced after inline the write in CommitProcessor with changes intorduced in ZOOKEEPER-3359, which didn't wait the in-flight write to finish before calling shutdown on the nextProcessor.
So it's possible that CommitProcessor thread and QuorumPeer thread will update the DataTree concurrently if we're doing fastForwardDataBase at the end of ZooKeeperServer.shutdown, which will cause inconsistent issue.
This JIRA is going to make sure we wait on the CommitProcessor to shutdown gracefully before calling shutdown on next processor, and exit if we cannot finish it gracefully to avoid potential inconsistency.