Open
Description
From kubernetes/kubernetes#96038 (comment)
It looks like we do delete deployment, configmap and secret at roughly the same time. This is incorrect in general, as there can be large delay between deployment deletion and pod deletion (e.g. if we make our tests even faster and the kube-controller-manager has a O(minutes) backlog of pods to be deleted).
During that delay, the running pod should be able to restart for whatever reason (e.g. kernel panic, machine failure).
We should modify our test to delete configmaps and secrets after we waited for all pods to be deleted.