[
https://issues.apache.org/jira/browse/HBASE-5875?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13266522#comment-13266522
]
chunhui shen commented on HBASE-5875:
-------------------------------------
bq.What if the RS went down just after processing the znode to OPENED? So only
SSH will come and try to assign root?
Yes, SSH will assign root. Also it remind me to the bug HBASE-5918, would you
take a see?
With the current patch, I think there is possibility of data loss mentioned in
HBASE-4880.
My approach is just a thought, since ROOT region is online in the
AssignmentManager when initializing, it must been assigned.
However, it also has a hole where remove hregioninfo from RIT but not add the
region to AssignmentManager.regions in AssignmentManager#regionOnline().
> Process RIT and Master restart may remove an online server considering it as
> a dead server
> ------------------------------------------------------------------------------------------
>
> Key: HBASE-5875
> URL: https://issues.apache.org/jira/browse/HBASE-5875
> Project: HBase
> Issue Type: Bug
> Affects Versions: 0.92.1
> Reporter: ramkrishna.s.vasudevan
> Assignee: ramkrishna.s.vasudevan
> Fix For: 0.94.1
>
> Attachments: HBASE-5875.patch
>
>
> If on master restart it finds the ROOT/META to be in RIT state, master tries
> to assign the ROOT region through ProcessRIT.
> Master will trigger the assignment and next will try to verify the Root
> Region Location.
> Root region location verification is done seeing if the RS has the region in
> its online list.
> If the master triggered assignment has not yet been completed in RS then the
> verify root region location will fail.
> Because it failed
> {code}
> splitLogAndExpireIfOnline(currentRootServer);
> {code}
> we do split log and also remove the server from online server list. Ideally
> here there is nothing to do in splitlog as no region server was restarted.
> So master, though the server is online, master just invalidates the region
> server.
> In a special case, if i have only one RS then my cluster will become non
> operative.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
