Manik Surtani wrote:

On 3 Mar 2009, at 22:14, Brian Stansberry wrote:

I kind of got lost in the details there, but I don't think that affects my opinion. ;)

IMHO trying to do complex stuff in toString() is asking for trouble. Either DataContainerImpl.toString() or CacheInvocationDelegate.toString() should stop calling into arbitrarily complex code and just output basic info.

Yup.  Also, this is a known issue and is fixed in 3.0.3.GA:

https://jira.jboss.org/jira/browse/JBCACHE-1478

I'm already using 3.0.3.GA...

Galder Zamarreno wrote:

Hi,

Bear with me on this one, it's a bit long :)

I've been running some tests creating Cache instances via MC and I've noticed that when TRACE logging is enabled on org.jboss.kernel, my Cache beans are not deployed whereas when org.jboss.kernel is set to DEBUG or higher, Cache beans deploy fine.

The error I'm getting is the following:

2009-03-03 21:40:51,505 ERROR [AbstractKernelController] (main) Error installing to Instantiated: name=CacheClusterA1 state=Described

java.lang.UnsupportedOperationException: Not supported in UnversionedNode

  at org.jboss.cache.AbstractNode.getChildrenDirect(AbstractNode.java:277)

  at org.jboss.cache.mvcc.NodeReference.getChildrenDirect(NodeReference.java:272)     at org.jboss.cache.invocation.NodeInvocationDelegate.getChildrenDirect(NodeInvocationDelegate.java:163)     at org.jboss.cache.DataContainerImpl.numNodes(DataContainerImpl.java:461)

  at org.jboss.cache.DataContainerImpl.getNumberOfNodes(DataContainerImpl.java:438)     at org.jboss.cache.DataContainerImpl.toString(DataContainerImpl.java:394)

  at org.jboss.cache.DataContainerImpl.toString(DataContainerImpl.java:372)

  at org.jboss.cache.invocation.CacheInvocationDelegate.toString(CacheInvocationDelegate.java:144)     at java.lang.String.valueOf(String.java:2615)

  at org.jboss.util.JBossStringBuilder.append(JBossStringBuilder.java:114)

  at org.jboss.dependency.plugins.AbstractControllerContext.toString(AbstractControllerContext.java:362) Looks like in TRACE mode, MC is trying to print the Cache when this is not fully started.

To try to understand this, let's look at what DataContainerImpl logs when org.jboss.kernel is set to DEBUG:

log4j-jboss-kerneldebug.log:2894:2009-03-03 21:40:24,249 2198  TRACE [org.jboss.cache.DataContainerImpl] (main:) Starting data container. Using MVCC? false

log4j-jboss-kerneldebug.log:2895:2009-03-03 21:40:24,274 2223  TRACE [org.jboss.cache.DataContainerImpl] (main:) Setting root node to an instance of class org.jboss.cache.mvcc.NodeReference

...

log4j-jboss-kerneldebug.log:3038:2009-03-03 21:40:24,336 2285  TRACE [org.jboss.cache.DataContainerImpl] (main:) Starting data container. Using MVCC? true

log4j-jboss-kerneldebug.log:3039:2009-03-03 21:40:24,338 2287  DEBUG [org.jboss.cache.DataContainerImpl] (main:) Setting rootInternal to NodeReference{delegate=UnversionedNode[ /]}

After MVCC has been set to true, DataContainerImpl.getNumberOfNodes() is able to get the number of nodes successfully:

    if (!usingMvcc) return numNodes(root) - 1;

    return numNodesMvcc(rootInternal) - 1;

However, if in between the two "Starting data container" messages, someone (i.e. MC), tries to get the number of nodes (via toString), then Cache tries to get number of nodes of a MVCC node as a non MVCC node failing with the exception above. This is exactly what happens when you enable TRACE on MC kernel.

numNodesMvcc() calls getChildren(). numNodes calls getChildrenDirect().

With this TRACE enabled, log only show the first two lines:

log4j-jboss-kerneltrace.log:6731:2009-03-03 21:40:51,468 2817  TRACE [org.jboss.cache.DataContainerImpl] (main:) Starting data container. Using MVCC? false

log4j-jboss-kerneltrace.log:6732:2009-03-03 21:40:51,475 2824  TRACE [org.jboss.cache.DataContainerImpl] (main:) Setting root node to an instance of class org.jboss.cache.mvcc.NodeReference

So, how do we solve this?

I'm not sure I understand why DataContainerImpl.createRootNode() is called twice. The very first call, which seems to come from DataContainerImpl.injectDependencies() method (annotated with @Inject) looks suspicious. Why does it look suspicious? Because calls like this expect a config to be set. And when they're called initially from @Inject, the config object is not still set (when the call comes as part of the start operation, config is indeed set)

usingMvcc = config != null && config.getNodeLockingScheme() == NodeLockingScheme.MVCC;

And this returning false, when in fact the config is set to MVCC is what results of number of nodes being calculated incorrectly.

Not sure on how to fix this though.

Thoughts?

------------------------------------------------------------------------

_______________________________________________

jbosscache-dev mailing list

jbosscache-dev@lists.jboss.org

https://lists.jboss.org/mailman/listinfo/jbosscache-dev

--

Brian Stansberry

Lead, AS Clustering

JBoss, a division of Red Hat

brian.stansberry@redhat.com

_______________________________________________

jbosscache-dev mailing list

jbosscache-dev@lists.jboss.org

https://lists.jboss.org/mailman/listinfo/jbosscache-dev

--

Manik Surtani

Lead, JBoss Cache

http://www.jbosscache.org

manik@jboss.org

------------------------------------------------------------------------

_______________________________________________

jbosscache-dev mailing list

jbosscache-dev@lists.jboss.org

https://lists.jboss.org/mailman/listinfo/jbosscache-dev

--
Galder Zamarreņo
Sr. Software Maintenance Engineer
JBoss, a division of Red Hat