… lock contention * Block new transactions before updating the CH instead of after. * Changed TransactionLoggerImpl to use a single ReentrantReadWriteLock instead of two gates, this will ensure we have no ongoing modifications during the RebalanceTask execution. * Updated StaleTransactionCleanup to use the new TopologyChanged event to roll back invalid transactions (along with ViewChanged for replicated caches). * ClusteredGetCommand should never do a remote get call.
Ensure that JGroupsTransport populates the members list properly before doing allowing DistributionManagerImpl to start. Ensure that DistributionManagerImpl doesn't attempt to use the consistent hash before it has initialized it.
* All non-essential build plugins have been moved to a separate profile which can easily be disabled at build time. On top of that, created a packaging property to able to switch between default 'bundle' packaging and 'jar' packaging which is faster.
Preparing transactions were only removed if an exception was thrown during commit. The fix guarantees that the preparing tx is removed regardless of the store commit outcome.
Don't add too many requests to the replication queue, as the scheduler might kick in and replicate all of them.
…ashToComplete for existing nodes Fixed RehashAfterPartitionMergeTest, TopologyAwareStateTransferTest, TopologyInfoBroadcastTest, RemoteLockCleanupTest, ConcurrentOverlappingLeaveTest, ConcurrentNonOverlappingLeaveTest
Call EasyMock.verify at the end of the test for the JChannel mock
For TestNG threads I set the thread name directly in the test listener. For Infinispan threads I added the test name to the node address. This should make it easier to identify the test that a log message belongs to.
Make a random failure on server list easier to diagnose by indicating the contents of the collection of which sometimes is not part of.
* For some weird reason, the @BeforeClass was not being executed here. I think it might have been a result of removing the groups from SingleCacheManagerTest. Making sure alwaysRun is set to true in the @BeforeClass in SCMT made it work. * Also fixed some of the test names.
Fixed the Maven repositories URLs for the time being, as this is critical. The change is small but it took me ages to verify.
same connection * Added SocketTimeoutErrorTest
same connection * Not all IOExceptions are so fatal that the connection is guaranteed to be closed (e.g. SocketTimeoutException). Therefore, we should maintain a boolean flag that tells us if an IOException was raised during I/O to make TcpTransport.isValid() return correct value.