On Tue, Oct 2, 2012 at 6:06 PM, Sanne Grinovero <sanne@infinispan.org> wrote:

I just updated my local copy of Infinispan and this is the result in
the Query testsuite:

Failed tests:
testIndexingWorkDistribution(org.infinispan.query.distributed.MultiNodeDistributedTest):
expected:<5> but was:<4>
Tests run: 128, Failures: 1, Errors: 0, Skipped: 0

OR:

Failed tests:
testIndexingWorkDistribution(org.infinispan.query.distributed.MultiNodeDistributedTest):
expected:<5> but was:<4>
testQueryAfterAddingNewNode(org.infinispan.query.statetransfer.StateTransferQueryDistributedIndexTest):
org.infinispan.CacheException: Unable to invoke method public void
org.infinispan.statetransfer.StateTransferManagerImpl.waitForInitialStateTransferToComplete()
throws java.lang.InterruptedException on object of type
StateTransferManagerImpl
Tests run: 128, Failures: 2, Errors: 0, Skipped: 0

It doesn't fail all the time, but almost every time. It passed the
testsuite once, and it timeouts initial state transfer very often
(attached thread dump)

Doesn't look like NBST is non blocking after all?

DefaultCacheManager.getCache() is still blocking until the joiner receives some state (i.e. owns at least one segment in the consistent hash).

The pre-existing nodes are not blocked during this state transfer, and in fact we didn't block the joiner either at the beginning but I re-added the wait to fix some tests "the easy way". I created a task in JIRA to re-enable this, although I'm not sure if it should be the default: https://issues.jboss.org/browse/ISPN-2366