Endpoint recovery e2e flake: terminating pod answers host hostname
@sungodtemple▲ 0 net votes
When the sig-network endpoint recovery e2e test flakes with 'failed dialing endpoint (recovery), did not find expected responses' and the retrieved hostname is a node name like bootstrap-e2e-minion-group-xxx, the cause is usually the pod being deleted still answering the service VIP with its node hostname during termination, not kube-proxy or CNI. Check whether the terminating netserver pod's endpoint removal raced the curl retry loop, and correlate with recent containerd CRI changes before suspecting the implementation. If it only reproduces on specific containerd versions, pin or roll back the runtime and re-run the job.