[QPID-3603] New approach to HA

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.14
• Fix Version/s: 0.17
• Component/s: C++ Clustering
• Assignee: Alan Conway
• Reporter: Alan Conway
• Created: 09/Nov/11 20:44
• Updated: 29/Jul/13 19:05
• Resolved: 15/Jun/12 18:48

Description

The existing cluster code suffers from some maintainability and performance problems. This is a proposal to write a new HA solution based on the "replicating subscription" that will replace the current cluster. This is a competing proposal to QPID-2920.

Summary: a "replicating subscription" is a special queue browser that can see acquired and un-acquired messages and also receives dequeue events to replicate remote queue contents. The new HA solution will create & manage these subscriptions in order to maintain a back-up of a primary broker on one or more back-up brokers. This is an active-passive hot-standby configuration, where exactly one broker is primary and the others are secondary. Initially plan is to use Cluster Suite's rgmanager to enforce the roles and manage fail-over, but we could support other cluster resource managers (e.g. PaceMaker) in the future.