Closed deads2k closed 7 years ago
cc @foxish @kow3ns @smarterclayton
I did not see this happening until recently.
The controller misses that the first pod transitioned to unready and proceeds to create the second pod. The test uses a client to hit the api server and will scale up the statefulset as soon as the broken pod is observed but the controller uses caches so it may observe the scaled up statefulset quicker than the updated pod.
I will update the test to ensure that we wait until the unreadiness is observed on StatefulSet.Status.Replicas field prior to the scaling operation
SGTM
https://k8s-gubernator.appspot.com/build/kubernetes-jenkins/pr-logs/pull/41821/pull-kubernetes-e2e-gce-gci/18753/
go run hack/e2e.go -v -test --test_args='--ginkgo.focus=StatefulSet\s\[k8s\.io\]\sBasic\sStatefulSet\sfunctionality\s\[StatefulSetBasic\]\sScaling\sshould\shappen\sin\spredictable\sorder\sand\shalt\sif\sany\sstateful\spod\sis\sunhealthy$'
@ncdc do you know the statefulset owners?