A fabric has an MQ gateway feeding a master-slave pair of brokers. The elements of the fabric are on AWS, and geographically quite widely separated.

Quite often, after a deliberate switch-over of the master-slave broker roles brought about by shutting down the master, the gateway is left in a broken state where no brokers seem to be available. Although it is normal for this situation to exist transiently during failover, here it seems to be permanent, and a gateway restart is needed to recover service.