Details
-
Improvement
-
Resolution: Duplicate
-
Major
-
5.5.0
Description
Consider following scenario:
- Cluster has 3 or more server groups.
- Auto-failover max count = Y.
- Server group auto-failover is enabled.
- Server GroupA has nodes Node1, Node2 … NodeN.
- Y (max count) < N (number of nodes in GroupA).
- Nodes in GroupA, fail in cascading fashion. That is first Node1 fails, is auto-failed over, then Node2 fails and so on. Eventually, all nodes in GroupA fail.
Currently, in the above scenario, we will auto-failover Node1 … NodeY but not rest of the nodes because the max count has already been reached.
Recently, Perry & Anil had a suggestion to allow failover of rest of the nodes because it is safe to do so.
This improvement needs non-trivial code changes and will be considered post-Vulcan.