Kubernetes and cgroup v1: check your nodes before upgrading

What happened

The New Stack has published a piece in its Road to KubeCon series with the headline that Kubernetes on cgroup v1 is dead, and looks at what comes next. The page text we could read was thin, so we are not going to quote versions or dates. The direction is clear, though: Kubernetes is moving on from the older cgroup v1 interface to cgroup v2 for resource control on nodes.

Why it matters

Cgroups are how the Linux kernel limits and accounts for CPU and memory per container. Kubernetes relies on them for every request and limit you set. If your nodes still run cgroup v1, a future cluster upgrade may stop working or behave differently, and you would find out at the worst time: during an upgrade.

Teams without a platform team are most exposed. Node images are often chosen once and then left alone. Old self-managed nodes, long-lived custom AMIs, and older base images are the usual places where v1 hides. Managed clusters tend to have newer defaults, but that is not guaranteed for node pools you created years ago.

The second risk is software inside the containers. Monitoring agents, older Java runtimes, and tools that read cgroup files directly may assume the v1 layout. They can report wrong memory figures or fail to start on v2.

What to do this week

• Find out which cgroup version each node pool uses. On a node, running stat -fc %T /sys/fs/cgroup prints cgroup2fs for v2 and tmpfs for v1.
• List your node images and operating system versions, including any pool you have not touched for a year or more.
• Read the release notes and deprecation notes for the Kubernetes version you plan to move to, and confirm what they say about cgroup v1 for your setup.
• Test one non-production node pool on cgroup v2. Watch memory limits, out-of-memory kills, and your monitoring agents’ numbers.
• Check the container runtime and any agents or older language runtimes in your images for cgroup v2 support.

None of this needs a big project. A short inventory and one test pool will tell you whether the move is a non-event or needs planning.

Need a hand?

If you would like help checking your nodes, planning the move to cgroup v2, or sequencing a Kubernetes upgrade, that is exactly what our Kubernetes consulting covers. The first call is free and it is with an engineer, not a salesperson.

Source: Kubernetes on cgroup v1 is dead. Here’s what comes next.

Related service: Kubernetes Consulting