[jira] Commented: (AMQ-1116) deadlock when shutting down client that is configured with failover=true and is presently disconnected from broker

["Jay Repko (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/activemq/browse/AMQ-1116?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=48011#action_48011 ] 

Jay Repko commented on AMQ-1116:
--------------------------------

I'm not sure that this is actually fixed.   I am currently trying it out with a broker in one VM and a client in another.  If I kill the broker forcibly and then try to execute the close method of ActiveMQConnection, I get stuck in the failover transport until the reconnect tries expires after 10 minutes. 

It gets stuck in the close when it calls doSyncSendPacket() 
<code>
                    if (isConnectionInfoSentToBroker) {
                        // If we announced ourselfs to the broker.. Try to let
                        // the broker
                        // know that the connection is being shutdown.
                        doSyncSendPacket(info.createRemoveCommand(), closeTimeout);
                        doAsyncSendPacket(new ShutdownInfo());
                    }
</code>

This is because another message ack is waiting for the transport to reconnect to be sent and so is holding the writerMutex in MutexTransport.



> deadlock when shutting down client that is configured with failover=true and is presently disconnected from broker
> ------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-1116
>                 URL: https://issues.apache.org/activemq/browse/AMQ-1116
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Transport
>    Affects Versions: 4.0.2
>            Reporter: Chris Hofstaedter
>            Assignee: Rob Davies
>             Fix For: 5.1.0
>
>         Attachments: ConnectionPool.java, DemandForwardingBridgeSupport.java, FailoverTransport.java, FailoverTransport.java, FailoverTransport_AMQ-1116.patch
>
>
> Sounds like a great idea to me :) Please go ahead and submit a jira with the patch and we'll get it integrated ASAP
> On 1/4/07, Chris Hofstaedter <ch...@nmwco.com> wrote:
> > I ran into this issue as well on <= 4.0.2.  I never tested it against
> > 4.1.    The FailoverTransport doesn't shutdown if it is disconnected at
> > the time of the shutdown.
> >
> > I believe that the root cause is in the 
> > FailoverTransport.oneway(Command
> > command) function which will hold onto the command and keep trying to
> > reconnect until the message is sent or the transport is disposed.
> >
> > I applied a local patch that seems to solve the problem.  Note that 
> > I've only tested the patch against 4.0.2.
> >
> > Specifically, I early return from the oneway function if the command 
> > is ShutdownInfo and the connectedTransport is null.  This seems to 
> > solve the problem in that I get clean shutdowns on the client when 
> > failover is true and the client is presently disconnected.
> >
> > Does this seem like a reasonable patch?  Should I create a JIRA with 
> > this info?
> >
> >
> >
> > -----Original Message-----
> > From: James Strachan [mailto:james.strachan@gmail.com]
> > Sent: Tuesday, December 12, 2006 4:46 AM
> > To: activemq-users@geronimo.apache.org
> > Subject: Re: failover mode and client shutdown
> >
> > There could be a bug relating to closes with the failover transport 
> > possibly, but the ActiveMQConnection does wait up to the closeTimeout 
> > for a close to succeed before shutting down - so you could try reduce 
> > the timeout.
> >
> > http://incubator.apache.org/activemq/maven/activemq-core/apidocs/org/a
> > pa
> > che/activemq/ActiveMQConnection.html#setCloseTimeout(int)
> >
> >
> > On 12/12/06, Keith Irwin <ke...@gmail.com> wrote:
> > > Folks--
> > >
> > > When we have clients running and we take down AMQ (<= 4.1.0), then 
> > > attempt to shutdown the clients with Control-C (rather than kill the 
> > > JVM with a -9), the clients won't shut down.  It's as if a "close" 
> > > on the failover connection never reaches the amq client classes.
> > >
> > > I note that in the 4.1.0 release notes, this issue is referenced, 
> > > and the advice is to set the maxReconnectAttempts (or similar) 
> > > property to something non-zero.
> > >
> > > The problem is that we don't want there to be a max number of 
> > > attempts.  Unless we specifically want to take down the client (say, 
> > > for an apt-get package upgrade), we want it to keep on trying 
> > > forever.
> > >
> > > SO, my question: Is there an architectural reason for not being able 
> > > to close a failover connection if AMQ is down?
> > >
> > > If it isn't impossible due to tradeoffs elsewhere in the code base, 
> > > we might be willing to submit a patch to fix the issue.
> > >
> > > Our only other recourse is to attempt to close the connections in 
> > > separate threads, then timeout those threads after a while and fall 
> > > out the end of the java process.
> > >
> > > For instance:
> > >
> > >   Thread th = new Thread(new Runnable() {
> > >       public void run() {
> > >          connection.close();
> > >       }
> > >    });
> > >    th.start();
> > >
> > >    // give up after 2 seconds
> > >    Thread.currentThread().join(2000);
> > >
> > > I guess this is do-able, but it seems, you know, some how, well,
> > wrong.
> > >
> > > So, is it worth investigating a patch to AMQ?
> > >
> > > Keith
> > >

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