[hibernate-dev] Regression in AS7 (master) cluster cache invalidation test after moving from Hibernate ORM 4.1.6 to 4.1.9

[Scott Marlow]

 From the TRACE logs, I can see that 
org.hibernate.cache.infinispan.access.TransactionalAccessDelegate.remove(Object) 
invoked the writeCache.remove(key) with 
key=org.jboss.as.test.clustering.cluster.ejb3.xpc.bean.Employee#1 but 
the other node isn't notified of this change (as far as I can tell from 
the Infinispan trace output).

I looked closer at TransactionalAccessDelegate.remove and looks like we 
are removing the entry via the writeCache.  If the AS 2lc entity cache 
definition is using invalidation (which I think it is), 
TransactionalAccessDelegate will also use CACHE_MODE_LOCAL.

I tried changing TransactionalAccessDelegate.writeCache from:

  this.writeCache =
   Caches.isInvalidationCache(cache) ?
   Caches.ignoreReturnValuesCache(cache, Flag.CACHE_MODE_LOCAL) :
   Caches.ignoreReturnValuesCache(cache);

To:

   this.writeCache = Caches.ignoreReturnValuesCache(cache);

And my simplified version of the AS7 clustered 2lc test passed.

Why would we want to use Flag.CACHE_MODE_LOCAL when the cache is 
configured to be an invalidation cache?

Scott

On 12/14/2012 11:52 AM, Scott Marlow wrote:
> I'll add TRACE logging to
> org.hibernate.cache.infinispan.access.TransactionalAccessDelegate to
> show more information about what is going on in there when we fail to
> invalidate the 2lc on the other node.
>
> Any other ideas?
>
>
> On 12/14/2012 10:13 AM, Scott Marlow wrote:
>> On 12/14/2012 09:46 AM, Scott Marlow wrote:
>>> I enabled org.jboss.as.jpa TRACE logging as well as org.infinispan TRACE
>>> (I know overly verbose) logging.  Output is here
>>> http://dl.dropbox.com/u/35343318/jboss/serverlogs.zip
>>>
>>> In the node1 log, I see that we delete the entity and then try to
>>> perform the remove command.  I don't seem to see this command executed
>>> on node0:
>>>
>>> 09:10:22,416 INFO  [stdout] (http-/127.0.0.1:8180-1) Hibernate: delete
>>> from Employee where id=?
>>> ...
>>> 09:10:22,419 TRACE [org.infinispan.interceptors.CallInterceptor]
>>> (http-/127.0.0.1:8180-1) Executing command:
>>> RemoveCommand{key=org.jboss.as.test.clustering.cluster.ejb3.xpc.bean.Employee#1,
>>> value=null, flags=[CACHE_MODE_LOCAL, SKIP_CACHE_LOAD, SKIP_REMOTE_LOOKUP]}.
>>> ...
>>>
>>> I don't see anything in the node0 log about
>>> "RemoveCommand{key=org.jboss.as.test.clustering.cluster.ejb3.xpc.bean.Employee#1"
>>> being received or executed.
>>>
>>> Are there any other TRACE logging settings that I should enable?
>>>
>>> I'm going to make a small test change to show the 2lc cache statistics
>>> for node1 after the entity is deleted from the database.  The elementCountInMemory went from one to zero (as expected).
>>
>> The 2lc statistics on node1, show that the node1 2lc has been
>> invalidated to remove the deleted entity.
>>
>> 09:56:30,071 INFO  [stdout] (http-/127.0.0.1:8180-1) 2lc for
>> stateful.war#mypc.org.jboss.as.test.clustering.cluster.ejb3.xpc.bean.Employee:
>> SecondLevelCacheStatistics[hitCount=1,missCount=0,putCount=1,elementCountInMemory=0,elementCountOnDisk=-1,sizeInMemory=-1]
>>
>> Its node0, that didn't receive the "RemoveCommand" that also still has
>> the elementCountInMemory as one:
>>
>> 09:56:30,172 INFO  [stdout] (http-/127.0.0.1:8080-1) 2lc for
>> stateful.war#mypc.org.jboss.as.test.clustering.cluster.ejb3.xpc.bean.Employee:
>> SecondLevelCacheStatistics[hitCount=0,missCount=0,putCount=1,elementCountInMemory=1,elementCountOnDisk=-1,sizeInMemory=-1]
>>
>>
>>>
>>> Scott
>>>
>>> On 12/14/2012 12:03 AM, Scott Marlow wrote:
>>>> http://hudson.jboss.org/hudson/job/as7-param-pull/5123/ shows a failure
>>>> with the second level cache not being invalidated as expected across a
>>>> two node cluster.
>>>>
>>>> I'm able to reproduce locally, which is good.  We will continue to use
>>>> Hibernate 4.1.6 until we get past this regression (hopefully we can
>>>> understand the test failure cause soon).
>>>>
>>>> https://github.com/jbossas/jboss-as/blob/master/testsuite/integration/clust/src/test/java/org/jboss/as/test/clustering/cluster/ejb3/xpc/StatefulWithXPCFailoverTestCase.java#L276
>>>> is the line that is failing (yeah, the arguments to assertEquals are
>>>> backwards but that isn't the root issue ;)
>>>>
>>>> It is possible that its a bug in the test also.
>>>>
>>>> The cluster has two nodes, each with their own in-memory database.  The
>>>> test assumes that it can create the same entity on both nodes (same
>>>> primary key) and that for the purpose of testing the second level cache,
>>>> we can expect cache invalidation to occur on both nodes when deleting
>>>> the entity on one of the nodes.
>>>>
>>>> I believe that the cache invalidation should be synchronous but I'll
>>>> probably try adding a sleep to see if it could be an async/timing issue.
>>>>
>>>> I hope to have more information soon (probably in the morning on Friday).
>>>>
>>>> Scott
>>>> _______________________________________________
>>>> hibernate-dev mailing list
>>>> hibernate-dev at lists.jboss.org
>>>> https://lists.jboss.org/mailman/listinfo/hibernate-dev
>>>>
>>>
>>> _______________________________________________
>>> hibernate-dev mailing list
>>> hibernate-dev at lists.jboss.org
>>> https://lists.jboss.org/mailman/listinfo/hibernate-dev
>>>
>>
>> _______________________________________________
>> hibernate-dev mailing list
>> hibernate-dev at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/hibernate-dev
>>
>
> _______________________________________________
> hibernate-dev mailing list
> hibernate-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/hibernate-dev
>