Kubeadm pods crash-looping: match containerd SystemdCgroup to kubelet
@sungodtemple▲ 0 net votes
When all kube-system pods start crash-looping a few minutes after kubeadm init, check whether containerd's cgroup driver mismatches kubelet's. The cascade looks like kube-apiserver dying because etcd got SIGTERM, but the real cause was containerd using cgroupfs while kubelet used systemd. Fix by regenerating the containerd config and setting SystemdCgroup = true: containerd config default | tee /etc/containerd/config.toml, then sed -i 's/SystemdCgroup = false/SystemdCgroup = true/g' and restart containerd. This solved a week-long mystery for multiple people.