[
https://issues.apache.org/jira/browse/HBASE-4430?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13131162#comment-13131162
]
Li Pi commented on HBASE-4430:
------------------------------
This new patch synchronizes things a bit at the SingleSizeCache level. I do
leave the copy to Slab, along with reads, unsynchronized, so it should give a
performance advantage over Todd's super simplified patch.
If anyone was wondering, the race that occurred with the last patch worked like
this.
Steady State: Item A is cached.
Thread B: Starts to cache something, starts eviction of Item A,
evictionlistener is called, but has not been completed it.
Thread A: begins to recache item A, notices an entry for A already.
Thread B: gets around to evicting the entry for A, as it finishes eviction,
wakes all the sleeping threads up so they have a chance to recache.
Thread A: continues onwards, sleeps the thread, and waits for something to wake
it up. But B has already finished, and called notifier.notifyAll();
End result: A ends up sleeping indefinitely. Nothing notifies it.
I got rid of the jumble in the end. This should make maintenance much easier.
> Disable TestSlabCache and TestSingleSizedCache temporarily to see if these
> are cause of build box failure though all tests pass
> -------------------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-4430
> URL: https://issues.apache.org/jira/browse/HBASE-4430
> Project: HBase
> Issue Type: Task
> Components: test
> Reporter: stack
> Assignee: Li Pi
> Fix For: 0.92.0
>
> Attachments: HBase-4430.txt, TestSlabCache.trace
>
>
--
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
