Jump to conversation
Unresolved conversations (4)
@ehashman ehashman Jun 11, 2021
This is sort of the crux of the issue -- previously, this logic only toggled on termination, but it's possible for pods to need this logic on restart (e.g. daemonsets, node crash).
pkg/kubelet/kubelet_pods.go
@ehashman ehashman Jun 11, 2021
I think I _finally_ got it. This section here was preventing pods that didn't have a DeletionTimestamp set for some reason from getting an unknown termination set. So, remove this check, and increment restart count by one if this is not set below.
pkg/kubelet/kubelet_pods.go
@ehashman ehashman May 22, 2021
This should handle the daemonset pods with init containers.
Outdated
pkg/kubelet/kubelet_pods.go
@ehashman ehashman May 20, 2021
I think this will ensure the old status still gets set while avoiding the bugs fixed by https://github.com/kubernetes/kubernetes/pull/54593 and https://github.com/kubernetes/kubernetes/pull/51644
Outdated
pkg/kubelet/kubelet_pods.go
Resolved conversations (1)
@ehashman ehashman May 7, 2021
I *think* this should fix it. Previously, this logic only kicked in when the current state is unknown, which caused default to always go to Waiting, even if there was an old status. I think that's the bug.
Outdated
pkg/kubelet/kubelet_pods.go