[
https://issues.apache.org/jira/browse/HBASE-25720?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17614598#comment-17614598
]
Xiaolin Ha commented on HBASE-25720:
------------------------------------
Hi, [~MarchWang] , I saw the problem you described in HBASE-27413.
The idea here is to abort the RS ASAP when sync WAL failed, even in the stage
of preparing to flush memstore, while currently only fail when committing the
flush of memstore will abort the RS. The PR is not accepted, but you can
backport it, we are using it on our production environment smoothly.
For the sync wal stuck problems, several issues are helpful, I think they can
solve mostly of your problems, especially HBASE-22301, HBASE-26347, and
HBASE-25905.
> Sync WAL stuck when prepare flush cache will prevent flush cache and cause OOM
> ------------------------------------------------------------------------------
>
> Key: HBASE-25720
> URL: https://issues.apache.org/jira/browse/HBASE-25720
> Project: HBase
> Issue Type: Improvement
> Affects Versions: 1.4.13
> Reporter: Xiaolin Ha
> Assignee: Xiaolin Ha
> Priority: Major
> Attachments: prepare-flush-cache-stuck.png
>
>
> We call HRegion#doSyncOfUnflushedWALChanges when preparing to flush cache.
> But this WAL sync may stuck, and abort the flush of cache.
> !prepare-flush-cache-stuck.png|width=519,height=246!
> If we cannot aware of this problem in time, RS will OOM kill.
> I think we should force abort RS when sync stuck in preparing, like in
> committing snapshots.
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
