Follow scenario
1. Stop Broker3a. Broker3b becomes master. (unless hit ENTMQBR-2546)
2. Start Broker3a. Broker3b goes in the backup mode, waiting for the live server.
3. Restart Broker3b. Broker3b announces as backup of Broker3a
4. Hang Broker2a using "kill -STOP <pid of Broker2a>" (SIGSTOP)
Here, Broker2b sends voting request only to broker1a, it does not send the voting request to broker3a. Eventually, Broker2b does not become master.
and Broker2a is in a stopped state.

==> Worked for me, but after issuing SIGCONT to "paused" broker2a, slave did not get down at all, and we had 4 live brokers.

Originally reported via email thread Re: Failover from Master to Slave Stopped Working in AMQ 7.4 by msidda1@redhat.com, shchavan.