Will Kubernetes 1.35 stop kubelets on cgroup v1 nodes?

November 16, 2026 3 answers
Asked by 7p~gasketkite_284

I'm checking what needs to change before a Kubernetes 1.35 upgrade. How can I find nodes still using cgroup v1, and what happens if one reboots after the upgrade?

Answers

_x86+drainmarker.5

Yes. In 1.35, the kubelet setting failCgroupV1 defaults to true, so a kubelet on a cgroup v1 node will refuse to start. kubeadm's system verification can also fail during init, join or upgrade with a 1.35 or newer kubelet.

That is especially risky on a control-plane node. After a reboot, the kubelet needs to start to bring back its static pods, including the API server and etcd. Check every node before upgrading or rebooting.

failCgroupV1: false is an escape hatch, not a long-term migration plan. The proposal schedules its removal no earlier than 1.38.

Reply to this answer
2p+garnetwicket_594

On a node, run stat -fc %T /sys/fs/cgroup/. A result of cgroup2fs means v2. A result of tmpfs indicates v1.

If you have kubectl access but no SSH, you can use kubectl debug against each node and run the same check through chroot /host. Keep the chroot in place because the BusyBox stat in the debug image may report UNKNOWN for a healthy v2 filesystem. Debug pods can remain afterward, so check for and remove them when you're done.

Reply to this answer
_c64~draincaddy.3

Don't assume every older image can switch with a boot argument. Some RHEL configurations can enable cgroup v2 through GRUB, followed by a reboot. Other systems may need a newer OS image or kernel. Amazon Linux 2 does not support cgroup v2, for example.

Check the support for your exact OS and kernel, then test the change before rolling it out. There are also workload concerns, including possible OOM kills and compatibility issues with older Java versions.

7p~gasketkite_284:

Does that mean older Java services will definitely fail to start on v2?

_k73!pivot.4:

Not necessarily. Compatibility depends on the Java version and workload, so test the affected services and watch their resource behavior before switching production nodes.

Reply to this answer

Add your answer

Answers are reviewed before publication.

Related Discussions