[
https://issues.apache.org/jira/browse/HBASE-10101?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13843637#comment-13843637
]
Jeffrey Zhong commented on HBASE-10101:
---------------------------------------
{quote}
However, the region could have been open on the dead server for a while, so we
need to wait till the current dead server is done with log splitting to avoid
possible data loss.
{quote}
I thought this already last night. This can't be because the RIT state is in
transition not (opened) so the region can't get requests. During region
assignment, the same check(in am.forceRegionStateToOffline for offline case)
happens to prevent us from assigning a region prematurely(before logsplitting
work). Setting the last region location is logically same as removing the last
assignment because last assignment is only used to skip region assign inside
SSH.
{quote}
expire the server as long as it is not known to be dead
{quote}
I c. It's a good change but unrelated to this issue, right?
In overall, I'm good(+1) on your trunk v3 patch while I still prefer to my
hbase-10101-v3-update.patch because it has less changes and readable.
[~jxiang] If you check in your trunk v3 patch, please assign the JIRA to
yourself. Thanks.
> testOfflineRegionReAssginedAfterMasterRestart times out sometimes.
> ------------------------------------------------------------------
>
> Key: HBASE-10101
> URL: https://issues.apache.org/jira/browse/HBASE-10101
> Project: HBase
> Issue Type: Bug
> Reporter: Jimmy Xiang
> Assignee: Jeffrey Zhong
> Attachments: hbase-10101-v2.patch, hbase-10101-v3-update.patch,
> hbase-10101-v3.patch, hbase-10101.patch, test.log, trunk-10101.patch,
> trunk-10101_v2.patch, trunk-10101_v3.patch
>
>
> Sometimes, I got this test timed out. The log is attached. It could be
> because the new cluster takes a while to process the dead server, or assign
> meta.
--
This message was sent by Atlassian JIRA
(v6.1.4#6159)
