Details
-
Improvement
-
Resolution: Fixed
-
Major
-
7.2.0
Description
When failing over services such as n1ql, the SDK will still route traffic to n1ql node that is about to be failed over. Even after we have decided to failover the node and failover processing is currently ongoing.
The SDK derive the service nodes based on GCCCP payload which is determined by,
1. the service_map which get updated at the start of failover for each service, in failover:failover_service(as a part of failover_service_nodes here)
2. the cluster membership which is update at the end of failover in failover:deactivate_nodes as here
We should investigate if the update to service map can be performed at the start of failover, so that the SDK are notified ahead of time to not route requests to the soon to be failed over node.
Attachments
Issue Links
- relates to
-
MB-57932 Use ns-server service map to detect start of index node failover in GsiClient
- Closed