…sponding to the join request This is caused by the way JGroups immediately throws SuspectExceptions for unicast requests like our join request. We work around it by forcing commands through the anycast branch when the response mode is SYNCHRONOUS_IGNORE_LEAVERS.
Don't modify the builder instance when building the configuration object.
…d to do is to upgrade that to an WL, otherwise the WL acquisition will block forever.
Either, the client is getting messed up with the socket, or the server is writing to the wrong endpoint here. This commit is to help narrow what the cause is.
Also fixed some cache manager leaks.
…tsuite ConfigurationOverrideTest was not stopping its cache managers properly (and also didn't use the test transport settings).
… node names) but not in class names or in stack traces.
…tsuite Some tests were creating their own executors with Executors.newSingleThreadPool(). I added a new method fork(Callable) returning a Future to AbstractInfinispanTest to replace this usage.
…tsuite Replace KeyAffinityService with MagicKey in the test suite to minimize the number of threads used. When using KeyAffinityService, make sure we shut down the executor properly with shutdownNow.
…thread per connection
* Don't repeatedly call CacheManager.getCacheNames(). This is called for every request even if the cache does exist, and each invocation involves checking concurrent maps, creating a new Set, etc. This can be cached locally. * Don't call CacheManager.getCache() for every call. Again, this is expensive since latches are checked for concurrent startup of caches and this can be stored elsewhere again.
…fferent cluster addresses Will throw an IllegalStateException if messages with a @Test(timeout = x) annotation don't call TestCacheManagerFactory.backgroundTestStarted(this) at the beginning.