…tateTransferLargeObjectTest to fail randomly Remove some unused code in DistributionInterceptor.remoteGetBeforeWrite()
* Wait indefinitely for forked threads to terminate. This is needed to ensure the test does not fail even under high load during parallel run. * Add "protected <T> Future<T> fork(Runnable r, T result)" method in AbstractInfinispanTest.
* StateConsumerImpl.doApplyState should use CACHE_MODE_LOCAL for put commands to avoid potential StateTransferLocks during processing of StateResponseCommands * Update StateTransferLargeObjectTest to use new config style * Use fetchInMemoryState==true, otherwise the expected behaviour is incorrect
Because the MarshalledValue keeps a reference to its marshaller, it can't be used after the cache has been stopped. I changed DummyInMemoryCacheStore to serialize its entries to a byte. The store also keeps a reference to the marshaller, so I changed CacheLoaderFunctionalTest and PassivatePersistentTest to use the new store after a cache restart.
…reads We can't extend MarshallingConfiguration, so I'm using a proxy for ExternalizerTable and I'm resetting the proxy.
…nformation I created an alternative consistent hash factory that uses the JGroups addresses to keep the CHs of different caches in sync. It's not the default, so it needs to be configured explicitly with <hash factory="org.infinispan.distribution.ch.SyncConsistentHashFactory"/> In order to ignore the order of nodes in the members list, I made Address implement Comparable - the JGroups Address was already Comparable.
… doesn't appear any more.
…n and the Invalidation interceptors.
… a remote site.
* Also fixed a bunch of test to avoid leaking cache managers and make sure they don't end up with random NPEs.
…nections" This reverts commit 2f1b245.
When a node becomes the coordinator, it should not include nodes that that didn't respond to the cache status request in the consistent hash.