[
https://issues.apache.org/jira/browse/HBASE-1421?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12710438#action_12710438
]
stack commented on HBASE-1421:
------------------------------
MSG_REPORT_PROCESS_OPEN doesn't seem to do anything master-side any more (I
thought it used to update timers on master-side)? This means that if the open
message is lost, then we don't try open again seemingly? Seems broke.
Otherwise, I went through the processing of messages returned by the
regionserver and tried remove all places where we threw unchecked exceptions in
particular. I also changed the process message signatures so they don't throw
even IOExceptions. Instead we just log warnings since most of the time these
are non-fatal anyway and even if they are damaging, we probably want to keep
going with a warning log rather than throw an exception that can possibly do
even more damage.
> Processing a regionserver message -- OPEN, CLOSE, SPLIT, etc. -- and if we're
> carrying more than one message in payload, if exception, all messages that
> follow are dropped on floor
> ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-1421
> URL: https://issues.apache.org/jira/browse/HBASE-1421
> Project: Hadoop HBase
> Issue Type: Bug
> Reporter: stack
> Priority: Blocker
> Fix For: 0.20.0
>
> Attachments: 1421.patch
>
>
> Just saw this in pset cluster. Marking blocker.
> We had an incidence of HBASE-1344 on our 0.19.x era hbase cluster. The
> report from the regionserver was carrying at least two open messages. The
> first provoked the exception, the second open message was never processed.
> Regionserver thought it had successfully opened region. Master didn't know
> anything about it.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
