broker cluster node failure recovery

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

broker cluster node failure recovery

jonb
I have a network of several brokers where a couple nodes are producers to one queue and a couple of nodes are consumers on this same queue.  We are testing high availability and in doing so pull the network cable on one of the consumers.  While this is happening there is a steady message rate on this queue of about 100 messages per second.  When the cable is pulled, shortly after some messages queue up on the producer nodes up to 60 seconds and then about 60 seconds after the network cable is pulled, all these queued messages rush to the consumers nodes where they are processed in a burst.  Then the system normalizes and all seems to function fine/normally without the lost consumer.  Is there any way to shrink this time down to something shorter like, say, 5 seconds?  If so, what parameter(s) can/should I tweak?

Thanks in advance for any suggestions you might have!

FYI - we statically network brokers in the activemq.xml config file like so:

    <networkConnector uri="static:(tcp://producer1:61616,tcp://producer2:61616,tcp://consumer1:61616,tcp://consumer2:61616)"/>
Reply | Threaded
Open this post in threaded view
|

Re: broker cluster node failure recovery

Tim Bain
Have you tried shortening the WireFormat's maxInactivityDuration URI option
described at http://activemq.apache.org/configuring-wire-formats.html?

On Thu, Oct 30, 2014 at 9:08 PM, jonb <[hidden email]> wrote:

> I have a network of several brokers where a couple nodes are producers to
> one
> queue and a couple of nodes are consumers on this same queue.  We are
> testing high availability and in doing so pull the network cable on one of
> the consumers.  While this is happening there is a steady message rate on
> this queue of about 100 messages per second.  When the cable is pulled,
> shortly after some messages queue up on the producer nodes up to 60 seconds
> and then about 60 seconds after the network cable is pulled, all these
> queued messages rush to the consumers nodes where they are processed in a
> burst.  Then the system normalizes and all seems to function fine/normally
> without the lost consumer.  Is there any way to shrink this time down to
> something shorter like, say, 5 seconds?  If so, what parameter(s)
> can/should
> I tweak?
>
> Thanks in advance for any suggestions you might have!
>
> FYI - we statically network brokers in the activemq.xml config file like
> so:
>
>     <networkConnector
>
> uri="static:(tcp://producer1:61616,tcp://producer2:61616,tcp://consumer1:61616,tcp://consumer2:61616)"/>
>
>
>
> --
> View this message in context:
> http://activemq.2283324.n4.nabble.com/broker-cluster-node-failure-recovery-tp4686819.html
> Sent from the ActiveMQ - User mailing list archive at Nabble.com.
>