Skip to content
CertDirectory.io

Limited Time: 40% OFF all Linux Foundation Certifications. Save now

Back to the blog
Cloud-native news5 min read

Kubernetes sets a removal date for kube-proxy IPVS mode

Sajeeva Nambadawa Vithanage19 Nov 2026 · Senior Software Site Reliability Engineer

Kubernetes SIG Network has set a date for the end of kube-proxy's ipvs mode. In a post on the Kubernetes blog on 16 November 2026, the SIG said it plans to remove the backend in Kubernetes 1.40, and it asked everyone still on ipvs to plan a move to nftables or iptables.

If you run clusters, this is a migration to put on next year's plan. If you're studying for the CKA or the new CKNE, it's a useful reminder that "how a Service actually reaches a Pod" is not frozen knowledge. The answer is changing, and the exams follow the docs.

Key points

  • ipvs mode has been deprecated since Kubernetes 1.35 and still runs, with a warning.
  • The official docs say it will be disabled by default in 1.40 (you can turn it back on with the KubeProxyIPVS feature gate) and fully removed in 1.43.
  • The recommended replacement is nftables mode (GA since 1.33), on Kubernetes 1.36 or later and Linux kernel 5.13 or later.
  • If you can't use nftables yet, iptables mode is the fallback, and it is far faster than it was in 2017.
  • The default mode in Kubernetes 1.37 is still iptables; a future release will switch the default to nftables.

What changed

kube-proxy runs on every node and turns Services into packet-forwarding rules. On Linux it has three backends: iptables, ipvs and nftables.

ipvs was added to fix real scaling problems in the old iptables backend: slow rule matching in big clusters, and kube-proxy rewriting the whole ruleset on every change. It did fix them. But the SIG's post explains that "the kernel IPVS API has turned out to be a bad match for the Kubernetes Services API", and the backend never handled every Service edge case correctly.

It also still leans on iptables for features that IPVS can't express. That matters because Red Hat plans to remove the iptables API from RHEL, which would take the ipvs backend with it on those systems.

The timeline lives in KEP-5495 and in the kube-proxy reference docs:

Release What happens to ipvs mode
1.35 Deprecated: kube-proxy logs a warning, docs updated
1.37 KubeProxyIPVS feature gate added, on by default
1.40 Gate off by default: ipvs mode won't start unless you re-enable it
1.43 Code removed

So "removed in 1.40" in the blog post means "gone unless you opt back in". Treat 1.40 as your real deadline, not 1.43.

IPVS schedulers aren't a reason to stay

Some teams chose ipvs for its load-balancing schedulers instead of the random choice iptables and nftables make. The post is frank here: because each node's kube-proxy keeps its own scheduler state, most schedulers don't do better than random selection in a Kubernetes cluster. If that was your reason for ipvs, it's probably not a reason to stay.

How to check and switch

First, find out which mode you're on. In a kubeadm cluster the setting lives in the kube-proxy ConfigMap:

kubectl -n kube-system get configmap kube-proxy -o yaml | grep -n "mode:"

An empty mode: "" means the default (iptables in 1.37), not ipvs.

To move to nftables on a kubeadm cluster, change the mode and restart the kube-proxy Pods, as the post describes:

mode: nftables   # in the kube-proxy ConfigMap, under config.conf
kubectl -n kube-system rollout restart daemonset kube-proxy

If you override other ipvs.* settings, move them to their nftables equivalents too. Then read the "Migrating from iptables mode to nftables" notes in the docs: nftables mode only listens for NodePorts on the node's primary addresses by default, and it doesn't add firewall rules for you.

Switching to iptables instead? The SIG recommends rebooting each node after the change, not just restarting kube-proxy.

Pro tip: set the mode explicitly in every cluster, even if you're happy with the default. The docs recommend it so a future default change to nftables doesn't surprise you during an upgrade.

What this means if you're studying for the CKA or CKNE

The CKA exam currently runs on Kubernetes v1.37, where iptables is still the default. The curriculum doesn't list proxy modes by name, but Services & Networking (20%) and Troubleshooting (30%) both assume you can follow traffic from a Service to its endpoints. Knowing that kube-proxy owns that job, and where its config lives, saves time when a Service "should work" and doesn't.

For the Certified Kubernetes Network Engineer (CKNE), launched by CNCF on 10 November 2026, this is closer to the core. Its Service Networking and DNS domain (25%) includes "Understanding kube-proxy and CNI Alternatives", and Core Infrastructure and CNI lists Linux tools such as iptables for packet-level issues. Practise reading iptables and nft rulesets on a lab cluster, not only YAML.

I'd build a small kubeadm cluster, switch it between modes, and watch what changes on the node. It's an afternoon well spent for either exam. If you're preparing for the CKA, our discount page has the current offers.

FAQ

Will my cluster break when I upgrade to 1.40? Only if kube-proxy is set to ipvs mode. From 1.40 it won't run in that mode unless you enable the KubeProxyIPVS feature gate, which is a stopgap until the code is removed in 1.43.

Is nftables mode ready for production? Yes. It has been GA since Kubernetes 1.33. The SIG recommends it on 1.36 or later, because some performance fixes weren't backported to earlier releases.

What if my nodes run an older kernel? nftables mode needs Linux kernel 5.13 or later. On older nodes, switch to iptables mode for now and plan the OS upgrade.

Does this affect Windows nodes? No. On Windows, kube-proxy only has the kernelspace mode.

Do I need to learn IPVS for the CKA? The curriculum doesn't name it. Focus on Services, endpoints, DNS, Gateway API and troubleshooting connectivity, and know that kube-proxy is the component that programs Service rules on each node.

Check your clusters' kube-proxy mode this week, while 1.40 is still a long way off.

Sources