[GEODE-6802] Data/bucket (among redundant copies) inconsistence could occur after a member is lost

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.10.0
• Component/s: persistence, regions
• Assignee: Eric Shu
• Reporter: Eric Shu
• Created: 23/May/19 23:42
• Updated: 26/Sep/19 18:05
• Resolved: 06/Jun/19 20:28

Description

When a prime bucket(or replicate region) is lost during the operation, it could occur at a time it only sent the operation to only part of the replicates (redundant copies), but not to others. Geode does region synchronization after some delay using a timed task to sync with all other members.

When the lost member restarted (or new nodes joined), it could request GII from a source (before region sync occurred) did not receive the original operation. This restarted member or new member does not have the timed task which triggered by the original member departed event. This could lead to data inconsistence among the members.

In addition, for persistent members, there could be some exception occurred during GII, and region synchronization provider would not send necessary data to the requester and thus lead to data inconsistency.

Issue Links

• links to: GitHub Pull Request #3623