We have one set of k8s clusters running an old version of k8s, v1.13.12. We also have newer clusters running v1.22.6. I’m noticing something that might be a behavior difference between them, and I’d like to see if it’s known where this did definitely change between those versions.
We have a situation where a deployment with a node selector gets rolled out to a cluster where none of the nodes have a label satisfying the node selector.
In the old k8s, it’s like k8s said “ok, sure, I’m done, as you didn’t ask me to run any pods”. The rollout succeeds, even though no pods were started.
It appears that a similar scenario run on the newer k8s actually fails rollout in that case, saying that no nodes were available.
Can anyone confirm that this would definitely changed between those two versions?