jackrabbit-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jukka Zitting (JIRA)" <j...@apache.org>
Subject [jira] Commented: (JCR-2786) Cluster sync not always done when calling session.refresh(..)
Date Fri, 15 Oct 2010 15:44:33 GMT

    [ https://issues.apache.org/jira/browse/JCR-2786?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12921410#action_12921410
] 

Jukka Zitting commented on JCR-2786:
------------------------------------

My original thinking behind the syncCount mechanism from JCR-1753 was to skip the cluster
sync if another thread completes the sync after the sync() method was entered. I missed the
case where a thread performs the sync but is then delayed before it gets to the syncCount++
statement.

Your fix changes the logic from checking whether a sync was completed to whether a sync was
*started* after the sync() method was entered, which raises the likelihood of extra cluster
syncs. However, of the top of my head I don't see any good way to reliably track the completion
of a cluster sync, so for now I think your solution is the best. At least it can only causes
one extra cluster sync even if n threads were blocked waiting on syncLock.

PS: AtomicInteger enables a more elegant way to implement the check-and-increment operation:

if (count == syncCount.get()) {
    syncCount.incrementAndGet();
    ...
}

vs.

if (syncCount.compareAndSet(count, count + 1)) {
    ...;
}


> Cluster sync not always done when calling session.refresh(..)
> -------------------------------------------------------------
>
>                 Key: JCR-2786
>                 URL: https://issues.apache.org/jira/browse/JCR-2786
>             Project: Jackrabbit Content Repository
>          Issue Type: Bug
>          Components: clustering
>            Reporter: Thomas Mueller
>            Assignee: Thomas Mueller
>
> Session.refresh(..) is supposed to synchronize cluster changes, but this doesn't always
happen, specially if the syncDelay is low. The reason is a wrong assumption in ClusterNode.sync:
The code there to avoid duplicate sync calls doesn't always work as expected. The following
algorithm is used:
>         int count = syncCount;
>         syncLock.acquire();
>         if (count == syncCount) {
>             journalSync();
>             syncCount++;
>         }
>         syncLock.release();
> The problem is that the background thread might be at the line "syncCount++" when Session.refresh(..)
is called, so that the main thread believes journalSync was already called and thus doesn't
call it.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message