Cache Invalidation seems to fail when redeploy CMP beans

[mfrost]

We have in place a clustering architecture as described in the JBoss clustering book.

It currently consists of 3 nodes. 2 "read-only" nodes whose CMP entity beans have read-only accessor methods and a "read-write" node where the entity beans have "read-write" accessor methods.

We use the cache invalidation architecture (based on JGroups) to link the read-only and read-write beans together. Changes to the RW beans cause invalidation in the RO beans and the data remains consistent.

This all appears to work fine until we redeploy our application jar files. The system continues to function but we get out of date data in our RO beans. Basically the RO beans are not asked to ejbLoad and hence re-read the data from the DB.

Bounce all 3 nodes and the system works fine again.

Is it possible that the re-deploy has broken some part of the cache invalidation process? Anyone any ideas?

Many thanks

mark

[slaboure]

Which node do you redeploy when it doesn't work? the RO or the RW? anyway, I think you can add a bug report on sf.net/projects/jboss as we should support redeploy of such apps.

Cheers,


sacha

[mfrost]

hi Sacha

If I re-deploy to the RW node, cache invalidation still works fine.

If I re-deploy to any of the RO nodes the invalidation mechanism fails.

I will add a bug report as suggested (sourceforge site is down at present)

cheers

mark

[mfrost]

Please see bug:

http://sourceforge.net/tracker/index.php?func=detail&aid=973234&group_id=22866&atid=376685