Details
Description
For single partitioned indexes with replica, failover and recovery of nodes might result in index rebuilds.
Steps of reproduction
- Create 4 1-partitioned indexes with 1 replica over 3 nodes.
- Now failover and add-back one of the nodes, the expectation is that the same partitions to remain on the node that went down and came back.
- But these steps result in node-partition reassignments during recovery rebalance
For making similar situations work, planner uses the stable plan from the manager instance's cache. But the stable plan computation wasn't realizing that the plan isn't stable as it was with a single partition and replica after a failover.
Improved the IsStablePlan logic to account the above situation as well.
Build couchbase-server-7.2.0-1823 contains cbgt commit b38da2e with commit message:
MB-53348- 1 partition/1 replica index gets rebuild over failover-recovery