Details
Description
Here's what I'm found in code while trying to address MB-5062 (second instance):
- on cluster with multiple buckets somebody starts failover
- our failover is sequential. Imagine orchestrator crashes after failing over node for one bucket and before failing over another bucket
- then orchestrator is restarted and user decides to add node back instead of continuing failover
- our code that detects if database files need to be deleted, now incorrectly does not deletes any db files on that partly failed over node, because one of buckets still resides on this node. Thus first bucket for which failover was completed is reusing old data files on that node.