Description of problem: The totem protocol doesn't properly filter messages that are ordered from rings outside its membership. This can result in a segfault in newly starting nodes if messages were transmitted in the previous configuration. Nodes that temporarily leave and then rejoin may also suffer incorrect messages delivered to the totempg layer resulting in segfault. Version-Release number of selected component (if applicable): How reproducible: run modified mp5/aisexec program that creates 16 copies of aisexec on one node. The mp5 program will kill and restart nodes. Between 6-12 hrs there are a few crashes running modfified mp5 program. Steps to Reproduce: 1. simulate 16 node environment with just openais running mp5. 2. wait several hours until mp5 program fails. 3. Actual results: segfault and core files in /var/lib/openais/core.* Expected results: no failures reported by mp5 - no segfaults as a result of this scenario. Additional info:
file:///usr/share/doc/HTML/en-US/index.html
I have tested this case and the code does what it is supposed to - there wss a different problem which I believed was this problem. That problem has been fixed in the rhel 5.1 branch.