[
https://issues.apache.org/jira/browse/HBASE-6813?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13611990#comment-13611990
]
Amitanand Aiyer commented on HBASE-6813:
----------------------------------------
Looks like creating the file was done outside the lock in trunk already; but
not in 0.89-fb.
So, closing this as a 0.89-fb fix.
Additionally, in 0.89-fb, we also write a "dummy" log record, outside the lock.
This is done
because the first write to the block is typically costly (involves a trip to
NN) and we would
like to avoid it in the HLog sync code path. This might be something, that may
be ported, if needed.
> Optimise the time spent holding the updateLock under log roll
> -------------------------------------------------------------
>
> Key: HBASE-6813
> URL: https://issues.apache.org/jira/browse/HBASE-6813
> Project: HBase
> Issue Type: Improvement
> Reporter: Amitanand Aiyer
> Priority: Minor
> Fix For: 0.89-fb
>
>
> Log roll entails syncing the old log, closing it and creating a new log file.
> We currently do all the 3 steps while holding the updateLock. This causes
> latency spikes for puts during this time.
> We only need to sync the old log under the lock. Creating the new file, can
> be done before grabbing the lock. Closing the old file can be done after we
> release the lock.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
