Cloud & infrastructure

Kubernetes plans to remove kube-proxy IPVS backend in version 1.40

Kubernetes SIG Network has deprecated the IPVS backend for kube-proxy, citing architectural mismatches and maintenance burdens, with removal scheduled for version 1.40.

Illustration of network cables transitioning from chaotic to orderly streams in a server rack
Illustration generated for this article

The Kubernetes Special Interest Group (SIG) Network has officially announced the deprecation of the IPVS backend for kube-proxy, marking a significant shift in how the platform handles service proxying. This decision, finalized in late 2026, sets a clear timeline for the complete removal of the feature in Kubernetes version 1.40. Cluster administrators currently relying on IPVS must now prepare to migrate to alternative backends to maintain compatibility and support.

What happened

The IPVS backend was originally introduced as an experimental solution to address severe performance bottlenecks found in the older iptables backend. In early Kubernetes versions, iptables struggled with both data plane latency, where connection establishment slowed as the number of services grew, and control plane latency, where updating rules consumed excessive CPU and memory. IPVS solved these issues by using kernel hash tables for faster lookups and avoiding full ruleset rewrites on every update. For many years, it was the preferred choice for large-scale clusters requiring high throughput and low latency.

However, despite its initial performance advantages, the IPVS backend never achieved full feature parity with the Kubernetes Services API. The underlying kernel IPVS API proved to be a poor match for Kubernetes requirements, leaving certain edge cases unimplemented. Additionally, the codebase became increasingly difficult to maintain due to its architectural differences from other backends and a lack of batch update capabilities in the IPVS API itself. As a result, SIG Network decided to deprecate the backend in Kubernetes 1.35, with a final removal target set for version 1.40.

How it works

The core issue driving this deprecation is the mismatch between the IPVS kernel subsystem and the dynamic nature of Kubernetes services. While IPVS uses hash tables to solve the linear lookup problem of iptables, it lacks a batch update API. This means that while small changes are fast, large-scale updates can become slower than expected because each change must be processed individually. Furthermore, the IPVS backend often had to fall back to using iptables for features beyond basic load balancing, reintroducing the very scaling problems it was designed to solve.

Another critical limitation involves load balancing schedulers. Users initially favored IPVS for its ability to use sophisticated schedulers like Maglev Hashing. However, because kube-proxy runs independently on each node without shared state, these distributed schedulers cannot function effectively. Each node makes routing decisions in isolation, rendering advanced scheduling algorithms no more effective than simple random selection. This architectural constraint means the promised benefits of IPVS schedulers were largely illusory in a standard Kubernetes deployment.

Key details

  • The IPVS backend for kube-proxy is officially deprecated as of Kubernetes 1.35.
  • Complete removal of the IPVS backend is scheduled for Kubernetes version 1.40.
  • The nftables backend, which became generally available in version 1.33, is the recommended replacement for most users.
  • Users on older systems unable to use nftables should consider migrating to the modernized iptables backend.
  • The IPVS backend relies on iptables for some features, making it vulnerable to the same long-term support issues as pure iptables, including Red Hat’s plan to remove the iptables API from RHEL.
  • Migration to nftables requires Linux kernel 5.13 or later and Kubernetes 1.36 or newer for optimal performance and stability.

Why it matters

For engineering teams managing large Kubernetes clusters, this deprecation signals the end of an era for IPVS-based networking. Many organizations adopted IPVS specifically to handle scale, assuming it was the only viable option for high-performance service routing. The shift away from IPVS requires a reassessment of network configurations and potentially a upgrade path for underlying operating systems to support the nftables backend. Ignoring this timeline could lead to unsupported configurations and security risks once version 1.40 is released.

The move also highlights the maturity of the nftables backend, which now offers comparable or slightly better performance than IPVS without the architectural baggage. For developers and platform engineers, this simplifies the landscape by converging on a single, modern backend that aligns better with the Kubernetes API model. It reduces the maintenance burden on the open-source community and ensures that future networking features can be implemented consistently across all clusters.

What you can do

  • Audit your current kube-proxy configuration to determine if you are using mode: ipvs.
  • Check your node operating systems to ensure they run Linux kernel 5.13 or later, which is required for the recommended nftables backend.
  • If running Kubernetes 1.36 or later, switch your kube-proxy ConfigMap from mode: ipvs to mode: nftables and restart the pods.
  • For clusters on older Kubernetes versions or legacy OS distributions, evaluate the modernized iptables backend, which has seen significant performance improvements since 2017.
  • Test the migration in a staging environment to verify that service connectivity and load balancing behavior meet your application requirements.
  • Review KEP-5495 for the detailed technical timeline and specific migration steps provided by the Kubernetes project.
  • #kubernetes
  • #networking
  • #kube-proxy
  • #ipvs
  • #nftables

Tools from the Bytechap store

Keep reading

All stories