[QPID-3369] Loss of cluster elder can result in loss of whole cluster

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Cannot Reproduce
• Affects Version/s: 0.7
• Component/s: C++ Clustering
• Assignee: Alan Conway
• Reporter: Jason Dillaman
• Created: 25/Jul/11 17:06
• Updated: 06/Sep/16 16:22
• Resolved: 20/Dec/13 19:09

Description

If the cluster elder is lost, it is possible that the remaining nodes of the
cluster will fail with the following errors:

Error delivering frames: Cluster timer wakeup non-existent task
ManagementAgent::periodicProcessing (qpid/cluster/ClusterTimer.cpp:93)

– or –

Error delivering frames: Cluster timer drop non-existent task
ManagementAgent::periodicProcessing (qpid/cluster/ClusterTimer.cpp:109)

When a member is promoted to be the elder, the ClusterTimer::becomeElder()
method will add all the current cluster tasks to the Timer. However, there is
a potential race condition where CPG can deliver the timer wakeup from the
original elder.