2:44 a.m.
On Aug 16, 2011, at 4:26 PM, Sanne Grinovero wrote:
Hi Scott,
demanding people to configure a new cache for each application is very
tricky because of ISPN-658, unless you're referring to creating a
whole new instance of Infinispan, with a dedicated new CacheManager
and either starting a new JGroups channel or using the JGroups
multiplexer Paul had written. In case of the multiplexer I'm not sure
if the channel identifiers could be assigned dynamically according to
application deployments and at the same time be consistent with the
identifier numbers chosen on other application server instances, where
different applications might be deployed or in different order, so my
guess is that people would need to deal with additional JGroups
configurations and handle the burden of extra network ports to be
configured and managed.
Just to remind everyone about a previous discussion related to ISPN-658 and 2LC use case
which has an impact on this.
Even though currently entities are stored in caches named using the entity name (or region
name if configured), a few months back we decided that we''d move to store all
entities under the same cache. However, I'm not sure this will end up happening if we
can finally get a proper workaround for ISPN-658 in place, i.e.
https://issues.jboss.org/browse/ISPN-1324
Using a rich key containing also the PU and application name seems the
best option to me, and I wouldn't worry too much about the additional
payload as that needs to be transferred anyway: I don't see how the
receiving side could route the command to the proper application
otherwise.
With my previous comment in mind, Sanne's suggestion here looks fine. It's about
building the cache region names, which currently become actual cache names, correclly,
taking the PU and app name into account.
If if the future we end up putting all entities under the same cache, the region name will
become the key in the cache, as opposed to the cache name.
I think it would be nice - if possible - to not change the
regionfactory contract but in some way wrap the keys normally
produced, to keep it simple and avoid the payload for those cases it's
not deployed in AS7.
Sanne
2011/8/16 Scott Marlow <smarlow(a)redhat.com>:
> I'm looking at o.h.c.i.InfinispanRegionFactory and wondering how best to
> configure the 2lc for AS 7.1. After a recent forum discussion and
> opening AS7-1544, I need to make some corrections to properly
> accommodate multiple JPA persistence units sharing the same underlying
> cache definition or having their own.
>
> Currently, with AS 7.0/7.0.1, I think each application should have its
> own Infinispan cache definition. In theory, the same cache definition
> could be shared if entity class names are unique across all persistence
> units (within apps and across different applications). This can be
> accommodated currently by adding additional Infinispan configuration
> entries to the AS7 standalone.xml (or domain.xml).
>
> One question that I have, is could we add an additional key for
> identifying the 2lc cache. The additional key that I have at the time
> of creating the container managed entity manager, is the "application
> name scoped persistence unit name" (includes the PU name and the
> application/containing jar name). Perhaps that could be passed into
> InfinispanRegionFactory via a property representing a unique way to
> refer to all 2lc entries for a specific applications persistence unit
> definition. I'm not sure how desirable it would be to further qualify
> the cache entries with this additional information (especially if it
> increases the payload transferred during replication).
>
> I think the alternative would be to have each application, configure
> their own cache definition (either to share with all applications PU or
> create a separate definition per PU). This might be close to what I
> already described as a workaround in the
> http://community.jboss.org/message/620903 forum post.
>
> What do you think?
>
> Scott
> _______________________________________________
> hibernate-dev mailing list
> hibernate-dev(a)lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/hibernate-dev
>
_______________________________________________
hibernate-dev mailing list
hibernate-dev(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/hibernate-dev
--
Galder Zamarreño
Sr. Software Engineer
Infinispan, JBoss Cache