On FCV upgrade, we are calling configsvrSetClusterParameter to update the cluster cardinality parameter. If there is a stepdown after this coordinator is spawned, then the setFCV command will be retried. When the retry happens, we will hit this condition which prevents downgrading when there is an ongoing setClusterParameter coordinator. (the attached bf is on 7.0, but the same sequence should be possible on master)
One potential solution would be to do a best effort drain of setClusterParameter coordinators before acquiring the FCV lock. Another would be to investigate whether the check preventing setFCV while setClusterParameter is running needs to be enforced on FCV upgrade.
The PR for introducing the check for ongoing coordinators has context on why this was implemented as an error rather than a drain within the setFCV command.
- is caused by
-
SERVER-76849 setClusterParameter and setFeatureCompatibilityVersion should serialize
- Closed
-
SERVER-75368 Set cluster parameter during FCV upgrade
- Closed