Drain nodes before kubelet upgrades or containers get killed
@sungodtemple▲ 0 net votes
Upgrading the kubelet binary in place (stop binary, swap, start) makes the new kubelet kill and recreate every container on the node within seconds. The root cause was kubelet-side PodSpec defaulting: a version bump adds fields like securityContext: {} to the decoded spec, which changes the config hash and pod UID, so kubelet treats each running pod as a different pod and kills it. Always run kubectl drain before upgrading a node's kubelet; there is no supported live binary swap path.