[
https://issues.apache.org/jira/browse/HBASE-20920?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16553633#comment-16553633
]
Duo Zhang commented on HBASE-20920:
-----------------------------------
Actually the second persist is inside the lock... And we need to persist three
times, lock, execute, and free. And I think the execute and free can be
unified, but the lock part is not easy. But there are still space to optimize.
If holdLock is false, then it means that we do not care whether the lock is
restored when start up, so we can skip the persist of the lock operation. And
if holdLock is true, then for each step we will not do release and free every
time so no problem at all.
Thanks.
> Merge the update procedure store on locking with the general persist after a
> procedure execution
> ------------------------------------------------------------------------------------------------
>
> Key: HBASE-20920
> URL: https://issues.apache.org/jira/browse/HBASE-20920
> Project: HBase
> Issue Type: Sub-task
> Components: amv2, Performance
> Reporter: Duo Zhang
> Priority: Critical
>
> For each procedure step, we now update the store twice; once to persist lock
> state and then later to persist the change in the procedure's dataset.
> Writing the store is the long-pole in procedure operation. Can we unify the
> two steps? (It is awkward because currently second persist is done outside of
> lock because it slow... and then updating lock state must be done inside lock
> else....). TODO!
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
