[ https://issues.jboss.org/browse/JGRP-1638?page=com.atlassian.jira.plugin.... ] Bela Ban commented on JGRP-1638: -------------------------------- The problem was caused by the STABILITY message being sent unreliably (Message.Flag.NO_RELIABILITY): if this message was not received by the coordinator, it would not clear its 'votes' set and discard all subsequent STABLE messages from all members already in 'votes'. This means that the coordinator would *not* send new STABILITY messages from then on, blocking all progress and increasing memory due to message accumulation in NAKACK2. The reason this was less likely to happen with send_stable_msgs_to_coord_only=false was that as long as not *every* member drops the STABILITY message, some member will clear 'votes' and will thus send a STABILITY message. So the design relied on the fact that, while STABLE messages might get dropped, STABILITY messages could never be dropped. Changing this invalidated the design assumption. SOLUTION: make STABILITY reliable again. This is OK as we're sending many STABLE messages but only relatively few STABILITY messages. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira