-
1. Re: Failover-Cluster
kurtstam Apr 25, 2007 10:39 AM (in response to psourisseau)Yes basically the UDDI/EPR handling will give you this. You can run duplicate services on different machines (JVMs), who will for instance listen to the same JMS queue. So whichever server has bandwidth will pick up the incoming message first. Now you will say; but it's JMS your single point of failure? And it is not since JBossMessaging can run clustered. So now you are fully redundant.
You can play the similar tricks with other protocols we use:
File - use a NAS or something like it.
S/FTP - use multiple servers with the files on the NAS
HTTP - route your request through mod_jk, or a hardware loadbalancer.
You should be able to run the ESB in the 'all' (clustered profile). But I'm not sure what it would give you more then what you already have, but it should all just work.
Hope this helps,
--Kurt -
2. Re: Failover-Cluster
psourisseau Apr 25, 2007 12:03 PM (in response to psourisseau)Hi,
Thx for your answer ;-)
I would like to understand UDDI/EPR. Have you some url or document about this or perhaps some basic explanation? is it standard UDDI or something else...
For clustering queuing management, i'm sure that's OK with jbossmessage.. a goob product.
In a "bpel" process, a state should be necessary, how it can be shared by all esb instance?
Regards,
Pascal -
3. Re: Failover-Cluster
kurtstam Apr 25, 2007 12:12 PM (in response to psourisseau)You can take a look at:
http://anonsvn.labs.jboss.com/labs/jbossesb/trunk/product/docs/services/Registry.odt
Each service registers its EPR to the UDDI.
I'm not sure I understand the second question. You are running bpel processes on jboss? Or orchestrating the ESB using BPEL? I'm not sure how this relates to failover..
--Kurt -
4. Re: Failover-Cluster
marklittle Apr 25, 2007 5:47 PM (in response to psourisseau)"psourisseau" wrote:
In a "bpel" process, a state should be necessary, how it can be shared by all esb instance?
Are you talking about replicating BPEL processes for fail-over? Typically you'd cope with fail-over by using a combination of replicated stateless services, compensation handlers, transactions (compensation transactions being more useful than ACID in this case) and re-running the process in the event of a failure. -
5. Re: Failover-Cluster
psourisseau Apr 26, 2007 4:09 AM (in response to psourisseau)Hi Mark,
Thx for your answer,
yes, it's about bpel process failover..
I've downloaded a white paper about this...
I must compile it and do a sample...
It seems that there's not a standard mecanism implemented by jBpm for cluster environment, is it correct ? but a pattern exist...
Do you know when the final release of bpm integration in jbossesb was done ?
Thx,
Regards,
Pascal -
6. Re: Failover-Cluster
marklittle Apr 26, 2007 8:40 AM (in response to psourisseau)jBPM+ESB integration will probably continue for some time. However, the first official release of what we have so far will be in the next couple of weeks. We should have a good level of integration for the 4.2 GA release around July.