Re: [xmlblaster] locking bug

[Marcel Ruff <mr at marcelruff.info>]

Póka Balázs schrieb:
> Hi there,
>
> Just a thought: what if you modified SessionInfo.timeout(Object) like 
> this:
>
>    public final void timeout(Object userData) {
>       synchronized (this.EXPIRY_TIMER_MONITOR) {
>          this.timerKey = null;
>          log.warning(ME+": Session timeout for " + getLoginName() + " 
> occurred, session '" + getSecretSessionId() + "' is expired, autologout");
>       }
>       DisconnectQosServer qos = new DisconnectQosServer(glob);
>       qos.deleteSubjectQueue(true);
>       try {
>          glob.getAuthenticate().disconnect(getAddressServer(), 
> getSecretSessionId(), qos.toXml());
>       } catch (XmlBlasterException e) {
>          log.severe(ME+": Internal problem with disconnect: " + 
> e.toString());
>       }
>    }
>
> It seems as if timerKey is protected by EXPIRY_TIMER_MONITOR 
> elsewhere, but not here (this might be an error). I simply removed the 
> locking of "lock" from the method. Does this make it 
> thread-unsafe? :)
No, i think the original sync did not make sense,
as the disconnect (coming here from a timeout thread)
can also come from a remote client which disconnects at any time,
so the core must be capable to handle this situation anyhow.

I have removed the sync here and also in shutdown() to prevent the deadlock.
The current code is now committed to svn.

thanks again
Marcel
> regards,
> Balázs


--
Marcel Ruff
http://www.xmlBlaster.org
http://watchee.net
Phone: +49 7551 309371