Details
-
Bug
-
Resolution: Fixed
-
Major
-
7.6.0
-
None
-
Untriaged
-
0
-
Unknown
Description
In a recent case, we saw that when the Index service starts, it waits to "validate its cluster cache" before connecting to ns-server's service management API. (I.e. registering the manager.) In this case, it caused auto-failover to be delayed by 21 s as it took a while for Index to be able to reach a second node in the cluster that was also being restarted. This is problematic because the second node was the node being failed over. If that node was actually unreachable, failover would have been arbitrarily delayed.
It may not be so easy to fix, but Indexer should not wait to communicate with other index nodes in the cluster before connecting with the cluster manager.
Attachments
Issue Links
- backports to
-
MB-63324 [BP to 7.6.4] Indexer should connect to the ns-server service API before checking other index nodes in cluster
-
- Open
-