Kubernetes on cgroup v1 is dead. Here’s what comes next.
The New Stack Bill Doerrfeld
Kubernetes is moving off cgroup v1, and v1.35 already blocks new kubelets on it by default. That shift ties straight into AI, edge, and better resource control.
Based on reporting by The New Stack, Bill Doerrfeld — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Kubernetes is running into a familiar kind of cleanup job: the old Linux resource interface is being pushed out, and the people building around AI and edge workloads are already adjusting. In this week’s Road to KubeCon roundup from The New Stack, cgroup v2 takes center stage, alongside node swap, edge compute, and the growing pressure on networking and observability.
Paco Xu of DaoCloud makes the case bluntly. cgroup v1 has too many limits, he says, while cgroup v2 offers a single hierarchy and a cleaner base for resource isolation. Depending on the Kubernetes version and setup, that also opens the door to features like memory QoS updates, container-aware OOM handling, and rootless support. Since Kubernetes v1.35, kubelet refuses to start on cgroup v1 nodes by default. Xu’s advice is simple: move every Linux node to v2 before upgrading.
That resource shift matters because the workloads are changing underneath it. Ocean Xie and Yuan Wang point to agentic AI as a memory-hungry problem: nodes often hit RAM limits before they run out of CPU, and the idle memory left behind makes cluster density harder to justify. Their answer is node swap, which reached general availability in Kubernetes v1.34. In their tests, pairing swap with fast NVMe SSDs produced density gains of up to three times in some scenarios.
Edge AI is putting its own pressure on infrastructure. HPE’s Aaron Lamond argues that organizations keep trying to force unoptimized servers into jobs they were never built for, especially when intelligence gets more distributed. He says purpose-built compute needs to account for hardware, software, security, and operations together, and points to ProLiant edge servers as suited for resource-constrained environments that still need high-grade security.
The same theme shows up in the co-located events around KubeCon + CloudNativeCon North America 2026 in Salt Lake City. Kubernetes on Edge Day is meant to close the gap between data-center habits and edge reality, with observability and security in focus. Cilium’s own event is leaning into AI, GPU networking, and security too, which is what happens when Kubernetes keeps getting asked to do more than the people who wrote it probably imagined.
My take — AI-written commentary, not fact-checked reporting
The cleanest read here is that Kubernetes is finally forcing operators to grow up about the kernel layer instead of treating it like plumbing nobody has to think about. cgroup v1 limps along until it doesn’t, and then the bill arrives with interest. The fun part is that AI workloads are speeding up the cleanup, which is very on-brand for infrastructure: nothing motivates a migration like running out of memory in public.
Read more about this at: The New Stack