# Kubernetes SIG Network schedules IPVS kube-proxy removal in version 1.40

> **Key takeaway:** Kubernetes will remove the IPVS kube-proxy backend in version 1.40 after deprecating it in 1.35, citing architectural mismatches with the kernel IPVS API, incomplete feature coverage, and the availability of the superior nftables backend.

**Published:** 2026-11-16T18:35:00+00:00 
**Source:** Kubernetes Official Blog 
**Category:** cloud-native 
**Canonical URL:** https://fosswire.org/news/kubernetes-sig-network-schedules-ipvs-kube-proxy-removal-in-.html 

## Summary
Kubernetes SIG Network has announced plans to remove the IPVS backend from kube-proxy in Kubernetes 1.40, following its official deprecation in version 1.35. The IPVS backend was originally introduced to solve performance problems in the iptables backend, specifically O(n) data plane latency from linear rule matching and excessive control plane overhead from rewriting entire rulesets on every service or endpoint change. While IPVS initially succeeded at these goals through kernel hash tables and incremental updates, the backend never achieved full compatibility with Kubernetes Services functionality and suffered from maintenance gaps after its original authors left the project.

The nftables backend, introduced in Kubernetes 1.29 and made generally available in 1.33, is now positioned as the replacement for both IPVS and iptables. SIG Network recommends that IPVS users on Kubernetes 1.36 or later with Linux kernel 5.13+ migrate to nftables mode, which offers full feature completeness, better long-term support prospects, and marginally improved performance. The removal also addresses a platform support concern: Red Hat's plan to eliminate the iptables API from RHEL would have broken the IPVS backend regardless, since it relies on iptables fallbacks for non-load-balancing features.

## Why IPVS failed and what replaces it
- **Performance trade-offs reversed at scale:** The IPVS backend avoided iptables' O(n) connection matching and full-ruleset rewrite problems, but the lack of batch update operations in the kernel IPVS API meant it became slower than iptables when processing large numbers of simultaneous service or endpoint changes.
- **Architectural dependency on iptables:** Because the IPVS API is too limited to implement full Kubernetes Services semantics, the backend fell back to iptables for features beyond basic load balancing, inheriting the same scaling and maintenance problems it was meant to solve; ipset mitigations also lacked batch update support.
- **IPVS schedulers rendered ineffective:** Kube-proxy's distributed node-per-node architecture prevents IPVS schedulers from functioning as designed, since scheduler state is not shared across nodes; even Maglev Hashing (mh scheduler) cannot achieve its theoretical efficiency without the two-layer load balancer structure that kube-proxy lacks.
- **nftables as the unified successor:** The nftables backend became GA in Kubernetes 1.33 and is recommended for all IPVS users on Kubernetes 1.36+ with kernel 5.13+, with SIG Network noting that post-1.36 performance and scaling fixes make it preferable to earlier nftables versions.
- **Platform support deadline:** Red Hat's planned removal of the iptables API from RHEL would eliminate the IPVS backend on that platform independently of the Kubernetes deprecation, since IPVS mode requires iptables for feature completeness.

## Why it matters
Cluster operators running large-scale Kubernetes deployments on IPVS mode face a mandatory migration path to nftables or iptables before the 1.40 release, with significant operational planning required for environments where IPVS was chosen specifically for its historical performance advantages. The consolidation around nftables simplifies the kube-proxy maintenance burden for SIG Network while aligning Kubernetes networking with modern Linux kernel capabilities, though distributions and managed Kubernetes providers will need to validate nftables mode across their supported kernel versions and node configurations.