[GEODE-3024] race condition between server and restarted locator preparing membership views

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.2.0
• Component/s: membership
• Assignee: Unassigned
• Reporter: Bruce J Schuchardt
• Created: 02/Jun/17 16:03
• Updated: 17/Jun/17 18:04
• Resolved: 06/Jun/17 23:12

Description

When a locator is restarted & recovers from disk it will try to take over the role of membership coordinator for the cluster if it finds the current coordinator is a cache server. If the cache server is in the process of sending out a new view it may get into a race with the locator in sending out view preparation messages.

The locator will send out a view-prep message and the server will also send one. Responses to the view-prep message will include the conflicting view and each of the two processes will create a new view and send it out. This repeats ad-infinitum.

This problem was observed in a system that was shutting down at the same time a locator was being restarted.