[prev in list] [next in list] [prev in thread] [next in thread]
List: activemq-users
Subject: Re: pure master/slave question - console of slave shows many pending msgs
From: Torsten Mielke <torsten () fusesource ! com>
Date: 2012-10-31 19:36:13
Message-ID: 5DE51232-AC83-417D-A099-A0D2EE1AF570 () fusesource ! com
[Download RAW message or body]
Hello Mike,
5.5.1 or later should not have these problems with pure master/slave setup.
I can also confirm that 5.5.1 and later versions work correctly with JDBC and shared \
file system master/slave. So if somehow possible you should upgrade to the latest \
version. There anyway won't be any bug fixes done on the 5.3 5.4 and 5.5 branches at \
Apache.
The HA and Failover features have been hardened a lot since 5.3.
On the technical side:
> > > Mostly, I want assurance that the slave is not continuing to store messages \
> > > that were consumed/acknowledged from the master. (In other words, should the \
> > > master fail, then clients aren't going to get a bunch of duplicate messages...)
You can verify by taking a copy of the persistence adapter and start an isolated \
broker using that copy. Then check the brokers JMX statistics and try to browse the \
messages. On 5.3 I personally could not give you the assurance you are asking for.
Best Regards,
Torsten Mielke
torsten@fusesource.com
tmielke.blogspot.com
On Oct 31, 2012, at 4:25 PM, Mike L. wrote:
>
> Torsten:
>
> Thank you for your reply. There are two reasons we are still on 5.3.1
>
> 1. In TEST I see we have tried 5.3.1; 5.4.1; 5.5.0; and 5.5.1
> We also use Camel and in testing when I had ActiveMQ 5.4.1 up & running the \
> request-response pattern would very often timeout for us (which was unacceptable). \
> I finally concluded that this did not happen on 5.3.1 - and this was approximately \
> two years ago; I'm pretty sure 5.5 wasn't even out yet.
> 2. In our production environment we have five separate applications using ActiveMQ. \
> We would have to have a coordinated downtime of all 5 in order to upgrade to a more \
> recent version. Also, given my experience with 5.4.1 I was reticent to replace \
> 5.3.1 which has been running like a champ...
> Finally, I would prefer to use a shared JDBC master/slave. However, we are using \
> Oracle RAC and my testing has never been able to have the active ActiveMQ server \
> failover to the secondary oracle rac server. (I have seen similar posts from others \
> when searching on this topic).
> Any thoughts/tips are welcome!
>
> Thanks again,
>
> Mike (aka patzerbud)
>
>
> > Subject: Re: pure master/slave question - console of slave shows many pending \
> > msgs
> > From: torsten@fusesource.com
> > Date: Wed, 31 Oct 2012 14:04:20 +0100
> > To: users@activemq.apache.org
> >
> > Hi,
> >
> > Message replication should work correctly in pure master/slave setup on a decent \
> > version of ActiveMQ. 5.3 is rather old. Any reason for using that old version?
> >
> > Also, pure master/slave has some drawback (which are documented). Are you sure \
> > you don't want to use other models of master/slave?
> >
> > Regards,
> >
> > Torsten Mielke
> > torsten@fusesource.com
> > tmielke.blogspot.com
> >
> >
> >
> >
> > On Oct 30, 2012, at 10:31 PM, Mike L. wrote:
> >
> > >
> > > All:
> > >
> > > I have configured a pure (shared nothing) master/slave setup (both are running \
> > > ActiveMQ 5.3.1).
> > > One of the documentation points for this configuration is:
> > >
> > > A slave of a master broker consumes all message states from the master - \
> > > messages, acknowledgments and transactional states.
> > > However, when I view the admin console of the slave \
> > > (http://slaveip:8161/admin/queues.jsp) for many queues I see the "Number Of \
> > > Pending Messages" is all over the map. That is, sometimes it equals the \
> > > "Messages Enqueued" count; sometimes it is zero; and sometimes it is about \
> > > halfway in between. By the way, if I click on a queue name on this page it \
> > > "spins" but never comes back.
> > > Mostly, I want assurance that the slave is not continuing to store messages \
> > > that were consumed/acknowledged from the master. (In other words, should the \
> > > master fail, then clients aren't going to get a bunch of duplicate messages...) \
> > > TIA,
> > >
> > > Mike (aka patzerbud)
> > >
> > >
>
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic