[
https://issues.apache.org/jira/browse/SLING-11620?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17618223#comment-17618223
]
Carsten Ziegeler commented on SLING-11620:
------------------------------------------
[~sagarmiglani] I think changing to an explicit lock does not make a
difference. I haven't looked deeply into the problem, but I guess it is based
on registering / unregistering services within the synchronized block. So I
think all calls for register/unregister needs to move out of that block and the
synchronized block should only be used for calculating the state change
> Synchronization issue in Webconsole Security Provider
> -----------------------------------------------------
>
> Key: SLING-11620
> URL: https://issues.apache.org/jira/browse/SLING-11620
> Project: Sling
> Issue Type: Bug
> Components: Extensions
> Affects Versions: Web Console Security Provider 1.2.6
> Reporter: Sagar Miglani
> Priority: Major
> Attachments: BlockedThreads.png, SLING-11620_fix.patch,
> WebconsoleServiceListeners.patch
>
>
> WebconsoleSecurityProvider's
> [ServiceListeners|https://github.com/apache/sling-org-apache-sling-extensions-webconsolesecurityprovider/blob/master/src/main/java/org/apache/sling/extensions/webconsolesecurityprovider/internal/ServicesListener.java]
> class has three fields of Listener (an inner class of {{ServiceListeners}}
> implementing {{{}ServiceListener{}}}) {{{}repositoryListener{}}},
> {{{}authSupportListener{}}}, {{{}authListener{}}}.
> Each of listener uses synchronized methods and on any {{ServiceChanged}}
> event it may call {{notifyChange}} method of {{ServiceListeners}} parent
> class which is also a synchronized method and in this method we are again
> calling all of the listener fields' {{getService}} sychronized methods
> If two service events for these listeneres fields occur simulatenously a
> deadlock situation may arise.
> eg:
> Thread1:
> - repositoryListener -> serviceChanged
> - synchronized retainSerivce - (self lock acquired)
> - synchronized notifyChange - (parent's lock acquired)
> Thread2:
> - authSupportListener -> serviceChanged
> - synchronized retainService - (self lock aquired)
> - synchronized notifyChange - (wait for thread 1 to release parent's lock)
> Thread1:
> - synchronized (authSupportListener) getService - (wait for thread 2 to
> release the lock)
> Tried to replicate the scenario [^WebconsoleServiceListeners.patch]. The test
> in this patch file fails due to timeout.
> On taking thread snapshot you may see the thread blocked and waiting for each
> other. [^BlockedThreads.png]
> Shouldn't these all listener fields and parent class be using the shared lock
> instead of sychronized methods?
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
