Description of problem: rgmanager will hang during shutdown if two nodes get into a relocate-to-eachother deadlock. Version-Release number of selected component (if applicable): CVS/Devel How reproducible: 50% on 3-node cluster Steps to Reproduce: 1. run rgmanager on 3 nodes 2. run "while [ 0 ]; do clusvcadm -r foo; done 3. kill rgmanager with SIGTERM on all nodes at near-the-same-time Actual results: Two nodes end up with threads trying to relocate the service to one-another. Expected results: Clean shutdown. It does not matter what the return value from the relocate request is. Additional info: This is probably a simple logic error somewhere in the code.
One way to solve this (and all future problems like it) is to only allow stop & disable requests during shutdown.
I have (mostly) a fix for this. Still testing though.
Fixed in CVS
These have been fixed in CVS for some time; closing