activemq-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "James Strachan" <>
Subject Re: Failover / Persistence
Date Wed, 23 Jul 2008 10:31:40 GMT
2008/7/23 AD <>:
> Hello,
>  I am relatively new to activemq and I am trying to find the optimal setup
> for a failover cluster of brokers.  What i have is the following
>  2 physical servers each running a broker
>  2 physical servers each running a consumer
>  1 Java application implementing JMS API producer which will likely use
> failover:// protocol.
>  I am trying to figure out from the docs the recommended failover approach.
>  I dont have a SAN but I will have NFS, however it seems there are some
> locking issues with using NFS as a shared storage.
>  What i would ideally like is a basic master/slave setup where all consumers
> connect to master and then fail over to the slave ( I understand current
> release does not support failing back to primary) in addition to message
> persistence so no messages are lost if they are unprocessed on the master
> queue.
>  Any recommendations here?

As Hans says, Pure Master/Slave is a good option - particularly as it
doesn't require any special shared network drive or IP failover

I've worked with a number of customers who have gone with the shared
file system approach; but rather than relying on shared file system
file locking (which is sometimes unreliable), they've used a Virtual
IP system to ensure that only one broker is running on the file system
at any time - and using a unix Virtual IP system to fail over to
another master.

Either approach works fine; the latter is certainly a bit faster and
easier from an ActiveMQ perspective - plus avoids the failback issue;
but requires knowledge of setting up Virtual IP with failover etc.

I tend to recommend the approach which suits your internal expertise &
facilities the best. e.g. if you've a SAN or clustered oracle already;
use that - if you're already using a Virtual IP system - hey use that
too if you like; failing that there's Pure Master/Slave which is fine
too (though does require manual fail back)


Open Source Integration

View raw message