Closed johscheuer closed 2 years ago
Another approach could be to check in this loop: https://github.com/FoundationDB/fdb-kubernetes-operator/blob/main/controllers/bounce_processes.go#L60-L66 if we have processes that are not version compatible and add them directly to the addresses
slices. As a safeguard wee could also check the the cluster is not in an upgrade mode anymore (expected version != running version).
We solved this in another issue.
What would you like to be added/changed?
It can happen that a process is not bounced during an upgrade e.g. when the reboot RPC is failing between the operator and the fdbserver process. In order to mitigate this issue we could try to recreate all process groups that have an old version after the initial bounce.