You are viewing a plain text version of this content. The canonical link for it is here.
Posted to muse-user@ws.apache.org by Andreas Loew <An...@Sun.COM> on 2007/05/08 14:09:14 UTC
Adding clustering/high availability capabilities to the Muse WS-N
implementation?
All,
as I am currently facing some load balancing and high availability
requirements regarding a WS-N implementation to be created (which will
have to provide a Web Services adapter/"profile" for existing JMS
destinations), I am looking for comments/best practices/input about
using Muse in a clustered environment:
By "clustered" environment I mean an environment that consists of two or
more "deployment units" (instances) of Muse within two or more instances
of an app server/web container (ideally stateless by itself, but using a
cluster-wide backing store for stateful data).
So the main question is: Is it possible at all/what does it take to add
clustering capabilities to a Muse (or Muse WS-N) deployment scenario?
(1) The two stateful entities that would have to be managed in a
cluster-complient way are
(1a) Subscriptions (implement NotificationProducerFilePersistence)
(1b) PullPointDatastores (implement PullPointDataStore)
But I doubt that (1a) and (1b) will be sufficient to create the same
notion of state within all instances at all times:
E.g. a NotificationConsumer freshly subscribes with cluster node 1 - how
would cluster node 2 be noticed about the state change (i.e. share the
state of the HashMap _subscriptionsByEPR of SimpleNotificationProducer)?
(2) Are there other issues or incompatibilities that - after having
implemented the management of state for (1a) and (1b) in a cluster-aware
way - would break a cluster deployment scenario?
Many thanks in advance for any comments, pointers or feedback!
Best regards,
Andreas
--
Andreas Loew
Java Architect
Sun Microsystems (Germany)
---------------------------------------------------------------------
To unsubscribe, e-mail: muse-user-unsubscribe@ws.apache.org
For additional commands, e-mail: muse-user-help@ws.apache.org
Re: Adding clustering/high availability capabilities to the Muse WS-N
implementation?
Posted by Daniel Jemiolo <da...@us.ibm.com>.
You may want to look at the ResourceRouter code (in muse-core,
org/apache/muse/core/routing), which handles the resolution of EPRs to
resource instances. Additionally, the concepts in this article may be
helpful:
http://www-128.ibm.com/developerworks/library/ac-muse.html
Dan
Andreas.Loew@Sun.COM wrote on 05/09/2007 11:43:09 AM:
> Daniel,
>
> many thanks for your truly helpful reply.
>
> Daniel Jemiolo schrieb:
> > To my knowledge, WSN and WS-ServiceGroups are the two areas where the
> > implementation would break in a clustered environment.
>
> Just to confirm:
> Are there any other cluster-related issues with the existing WS-N
> implementation in Muse you can think of, *other than* related to the two
> issues that I have already mentioned:
>
> - NotificationProducer (or rather, Subscription state) and
> - PullPoint (or rather PullPointDataStore)?
>
> Any particular pieces of WSN code you'd recommend me to investigate
> further because they might cause additional issues in a cluster?
>
> Thanks again heaps & best regards,
>
> Andreas
>
> --
> Andreas Loew
> Java Architect
> Sun Microsystems (Germany)
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: muse-user-unsubscribe@ws.apache.org
> For additional commands, e-mail: muse-user-help@ws.apache.org
>
Re: Adding clustering/high availability capabilities to the Muse WS-N
implementation?
Posted by Andreas Loew <An...@Sun.COM>.
Daniel,
many thanks for your truly helpful reply.
Daniel Jemiolo schrieb:
> To my knowledge, WSN and WS-ServiceGroups are the two areas where the
> implementation would break in a clustered environment.
Just to confirm:
Are there any other cluster-related issues with the existing WS-N
implementation in Muse you can think of, *other than* related to the two
issues that I have already mentioned:
- NotificationProducer (or rather, Subscription state) and
- PullPoint (or rather PullPointDataStore)?
Any particular pieces of WSN code you'd recommend me to investigate
further because they might cause additional issues in a cluster?
Thanks again heaps & best regards,
Andreas
--
Andreas Loew
Java Architect
Sun Microsystems (Germany)
---------------------------------------------------------------------
To unsubscribe, e-mail: muse-user-unsubscribe@ws.apache.org
For additional commands, e-mail: muse-user-help@ws.apache.org
Re: Adding clustering/high availability capabilities to the Muse WS-N
implementation?
Posted by Daniel Jemiolo <da...@us.ibm.com>.
Hi,
We've never tested Muse in a clustered environment, but it has been used
to create ws-resources where all of the state was maintained outside the
Muse application (which is the foundation of your clustering setup). Of
course, WSN has always been tricky. In general, most people don't have a
"subscription manager" service/data store that fits the WSN model - they
either want us to handle all WSN-related things or they want to adapt
something like JMS. The former is easy - we implement all of the WSN data
structures and operations; the latter is harder - we haven't tried to do a
mapping of WSN to X, where X is any number of messaging
frameworks/systems.
It's certainly possible to write a WSN implementation that maps onto JMS
instead of our data structures. If you implement Muse's
NotificationProducer interface (using AbstractWsResourceCapability as a
base class), you can provide an alternate WSN impl that plugs right into
Muse and works with all of the other WS-* capabilities.
To my knowledge, WSN and WS-ServiceGroups are the two areas where the
implementation would break in a clustered environment. Ideally, I guess we
could break off the WSN and WSSG data structures into some kind of
independent service that could be deployed on its own, and which Muse's
implementations could talk to. That sounds like a lot of middleware and
configuration, though, and would hinder our out-of-the-box experience
(which already has a number of imperfections). Perhaps it would be
possible to refactor the code so that you could configure it for a
clustered environment, but then have a simple setup for other users; this
would be a very interesting item to pursue, but I'm afraid it's not the
case right now.
Dan
Andreas.Loew@Sun.COM wrote on 05/08/2007 08:09:14 AM:
> All,
>
> as I am currently facing some load balancing and high availability
> requirements regarding a WS-N implementation to be created (which will
> have to provide a Web Services adapter/"profile" for existing JMS
> destinations), I am looking for comments/best practices/input about
> using Muse in a clustered environment:
>
> By "clustered" environment I mean an environment that consists of two or
> more "deployment units" (instances) of Muse within two or more instances
> of an app server/web container (ideally stateless by itself, but using a
> cluster-wide backing store for stateful data).
>
> So the main question is: Is it possible at all/what does it take to add
> clustering capabilities to a Muse (or Muse WS-N) deployment scenario?
>
> (1) The two stateful entities that would have to be managed in a
> cluster-complient way are
>
> (1a) Subscriptions (implement NotificationProducerFilePersistence)
> (1b) PullPointDatastores (implement PullPointDataStore)
>
> But I doubt that (1a) and (1b) will be sufficient to create the same
> notion of state within all instances at all times:
>
> E.g. a NotificationConsumer freshly subscribes with cluster node 1 - how
> would cluster node 2 be noticed about the state change (i.e. share the
> state of the HashMap _subscriptionsByEPR of SimpleNotificationProducer)?
>
> (2) Are there other issues or incompatibilities that - after having
> implemented the management of state for (1a) and (1b) in a cluster-aware
> way - would break a cluster deployment scenario?
>
> Many thanks in advance for any comments, pointers or feedback!
>
> Best regards,
>
> Andreas
>
> --
> Andreas Loew
> Java Architect
> Sun Microsystems (Germany)
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: muse-user-unsubscribe@ws.apache.org
> For additional commands, e-mail: muse-user-help@ws.apache.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: muse-user-unsubscribe@ws.apache.org
For additional commands, e-mail: muse-user-help@ws.apache.org