- Previously, if the main rgmanager process died, either by an unexpected termination with a segmentation fault or by killing it manually, any service running on it, was immediately recovered on another node rather than waiting for fencing, like the rgmanager process did in previous versions. This was problematic for services containing Highly Available Logical Volume Manager (HA-LVM) resources using tagging, because the start operation failed if the tag that was found belonged to a node that was still a member of the cluster. With this update, service recovery is delayed until after the node is removed from the configuration and fenced, which allows the LVM resource to recover properly.
- Previously, attempts to start an MRG Messaging (MRG-M) broker caused rgmanager to terminate unexpectedly with a segmentation fault. This was caused by subtle memory corruption introduced by calling the pthread_mutex_unlock() function on a mutual exclusion that was not locked. This update addresses scenarios where memory could be corrupted when calling pthread_mutex_unlock(), and rgmanager no longer terminates unexpectedly in the described situation.