[
https://issues.apache.org/jira/browse/HBASE-27366?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17603609#comment-17603609
]
Duo Zhang commented on HBASE-27366:
---
The LockProcedure will take the table exclusive lock of ProcedureScheduler,
wile split/merge will take the shared lock, so I do not think the problem is we
do not use LockProcudure in split/merge procedure. Maybe the problem is how we
deal with master restart? And on master and branch-2, we introduced a
SnapshotProcedure. Maybe you could try to see if this approach can solve the
problem.
> split or merge removed region under snapshot
>
>
> Key: HBASE-27366
> URL: https://issues.apache.org/jira/browse/HBASE-27366
> Project: HBase
> Issue Type: Bug
> Components: snapshots
>Affects Versions: 2.4.10
>Reporter: Huaxiang Sun
>Priority: Major
>
> We run into snapshot failures for one table with large number of regions. The
> event sequence is like the following:
>
> # Snapshot process lists all regions for one table.
> # Normalize kicks in to split some regions for the table under snapshot.
> # split finishes and major compaction finishes. The parent region is moved
> to archive.
> # When the Snapshot processes the parent region, it does not exist and
> snapshot fails.
> Since snapshot process acquires the table lock, but there is no table lock
> acquired in split or merge process, they crash into each other.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)