[
https://issues.apache.org/jira/browse/HBASE-10101?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13842857#comment-13842857
]
Jimmy Xiang commented on HBASE-10101:
-------------------------------------
+1 on v3. For those two failed unit tests, if they are just flaky, I can take a
look in HBASE-10090.
bq. We are setting the servername but didn't we just do it up in
createOrForceNodeOffline?
createOrForceNodeOffline is just set the info in ZK, not the region states map
in the mater. Actually, this line is not much use, which will removed in in
HBASE-10090.
bq. We are doing the above because region could have come from another RS
altogether, one that did not get fully processed? Now we force it to be on the
current dead server?
We are doing this because the region is transitioning on the new/current dead
server now. This happens when the master restarts. In meta, the region could
be still assigned to the old dead server (as recorded in the lastAssignment
when the master loads the meta after restart). Since ZK shows the region is
already transition on the new dead server, we should not worry about log
splitting of the old dead server. That's why I agree with [~jeffreyz] on clear
the last assignment info here.
> testOfflineRegionReAssginedAfterMasterRestart times out sometimes.
> ------------------------------------------------------------------
>
> Key: HBASE-10101
> URL: https://issues.apache.org/jira/browse/HBASE-10101
> Project: HBase
> Issue Type: Test
> Reporter: Jimmy Xiang
> Assignee: Jeffrey Zhong
> Priority: Minor
> Attachments: hbase-10101-v2.patch, hbase-10101-v3.patch,
> hbase-10101.patch, test.log
>
>
> 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)
