Bookmarks

    Nothing saved yet. Use the bookmark on any article. Bookmarks are kept on this device only: they stay private and work without an account.

    Settings

    Appearance

    Reading mode

    Font size

    Content width

    Line spacing

    Keyboard shortcuts

    /
    Search
    j k
    Next / previous story
    ← →
    Previous / next section (columns)
    b
    Bookmarks
    s
    Settings
    r
    Columns / scroll
    t
    Light / dark

    Saved in this browser only.

    Ping2Prompt
    Subscribe to the Weekly Brief Sign in
    A row of blue server cubes receding into the distance, each marked with the white Kubernetes ship's wheel logo.
    Photo by Growtika on Unsplash

    Kubernetes 1.40 will remove the kube-proxy ipvs backend; plan your move

    Kubernetes

    Kubernetes 1.40 removes kube-proxy's ipvs backend

    Kubernetes SIG Network plans to remove the ipvs backend of kube-proxy in Kubernetes 1.40. It was deprecated as of 1.35 and stays available, with a deprecation warning, until it is removed.

    This is a deprecation notice, not a vulnerability fix. No CVE is involved. All ipvs users should plan to migrate to the nftables or iptables backends.

    Why SIG Network is dropping ipvs

    SIG Network gives several reasons:

    • ipvs is complex, not fully feature complete, and architecturally different from the other two backends, so it costs more to maintain.
    • The kernel IPVS API is a bad match for the Services API, and ipvs never handled all Service edge cases correctly.
    • The IPVS API has no batch updates, so ipvs can be slower than iptables when many changes arrive at once.
    • Most IPVS schedulers are not useful in kube-proxy, because kube-proxy processes on different nodes do not share scheduler state.
    • The original authors left the Kubernetes ecosystem, and finding new maintainers took a while.

    ipvs was built to fix real iptables problems: one rule per service IP with linear matching, and a full ruleset rewrite on every change. In very large clusters that rewrite could use 100% of one CPU. The nftables backend, introduced in 1.29 and GA in 1.33, was planned to eventually replace both older backends.

    ipvs users, including those on RHEL, need a migration plan

    You are affected if kube-proxy runs in ipvs mode on any of your clusters. RHEL nodes have a second deadline: Red Hat plans to remove the iptables API entirely, and the ipvs backend uses iptables, so it will also stop working there.

    Switch to nftables on 1.36 and kernel 5.13, else iptables

    Kubernetes recommends nftables if you run Kubernetes 1.36 or later on nodes with Linux kernel 5.13 or later. Some nftables performance and scaling fixes made after 1.33 were not backported to releases before 1.36. The project calls nftables fully supported and feature-complete.

    The configuration keys below come from the Kubernetes blog post listed under Sources; check them against your own release before changing anything.

    • Change the kube-proxy configuration to nftables mode and restart kube-proxy.
    • On kubeadm clusters, change mode: ipvs to mode: nftables in the kube-proxy ConfigMap, then restart the kube-proxy pods.
    • Change any other ipvs-specific kube-proxy options to their nftables equivalents.

    If nftables is not possible because of an older Kubernetes or an older Linux distribution, use the iptables backend, which has improved greatly since ipvs appeared in 2017. For iptables, SIG Network recommends rebooting each node after the configuration change instead of only restarting kube-proxy. SIG Network says iptables in 2026 is still not as good as ipvs but much better than in 2017, so many who found it unusable would not hit problems now.

    What the post leaves open about timing and speed

    KEP-5495 holds the full timeline and plan for the deprecation. The performance claim, that nftables is very slightly better than ipvs, rests on the project's own connection-latency graph, not an independent benchmark.

    Sources

    Image: Photo by Growtika on Unsplash


    This story was produced with AI assistance from the sources listed above and fact-checked against them before publication. Read our editorial methodology and the corrections log. Spotted an error? Email [email protected].

    Continue

    Share on Bluesky · Mastodon · Email ·

    Discussion