What were you trying to do that didn't work?

 
In 2 node cluster with qdevice server as a arbitrator when node1's network is blocked and recovered after token timeout but within the consensus time, both nodes lose quorum. Issue occurs no matter what qdevice algorithm is used (lms/fsplit).

Please provide the package NVR for which bug is seen:

corosynclib-3.1.7-1.el9.x86_64                            
corosync-3.1.7-1.el9.x86_64                                
corosync-qdevice-3.0.2-2.el9_2.x86_64 

How reproducible:

Always

Steps to reproduce

Create cluster consisting of:

• node1
• node2
• qdevice
  Set totem token to 20000ms (in order to trigger the issue more easily as it is heavily depending on timing so to have enough time to recover the network just in between totem expiring but consensus not expiring yet - so after 35s in this reproducer)

[root@c-rhel9-node1 ~]# date; logger -t ========== algorighm=ffsplit, sleep 35 ==========; ./net_breaker.sh BreakCommCmd node1; sleep 35; ./net_breaker.sh FixCommCmd node1;
(net_breaker.sh is from https://access.redhat.com/solutions/79523)
 

Expected results

No quorum loss on node2

Actual results

Both nodes lost quorum temporarily, and node2 was fenced.