[
https://issues.apache.org/jira/browse/HBASE-14947?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Matteo Bertozzi updated HBASE-14947:
------------------------------------
Description:
We ended up with a deadlock in HBASE-14943, with the storeTracker and lock
acquired in reverse order by syncLoop() and insert/update/delete. In the
syncLoop() with don't need the lock when we try to roll or removeInactive. also
we can move the insert/update/delete tracker check in the syncLoop avoiding to
the extra lock operation.
was:
Currently we rely on locking at every operation level in WALProcedureStore,
this requires use of multiple locks to eliminate potential race conditions.
But, use of multiple locks is a recipe for potential deadlocks.
Investigate if there is a better way to deal with this.
> WALProcedureStore improvements
> ------------------------------
>
> Key: HBASE-14947
> URL: https://issues.apache.org/jira/browse/HBASE-14947
> Project: HBase
> Issue Type: Sub-task
> Components: proc-v2
> Reporter: Ashu Pachauri
> Assignee: Matteo Bertozzi
> Priority: Minor
>
> We ended up with a deadlock in HBASE-14943, with the storeTracker and lock
> acquired in reverse order by syncLoop() and insert/update/delete. In the
> syncLoop() with don't need the lock when we try to roll or removeInactive.
> also we can move the insert/update/delete tracker check in the syncLoop
> avoiding to the extra lock operation.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
