Why did a clustered qpid broker shut down after a network problem?

Issue

• Running MRG / qpid brokers in a cluster
• Some or all qpid brokers terminating unexpectedly

For example,

• We met some networking problems for around 2 seconds. Several seconds later, the whole qpid cluster went down with the following message being logged:

critical Error delivering frames: Cluster timer drop non-existent task ManagementAgent::periodicProcessing

• One broker went down with a log entry

critical Error delivering frames: Unknown connection: Frame[BEbe; channel=0; {MessageFlushBody: destination=1; }] data 10.1.2.3:7727-1 read-credit=1 (qpid/cluster/Cluster.cpp:542)