
Kubernetes SIG Network has set a date for removing the IPVS backend from kube-proxy. The backend has been deprecated since Kubernetes 1.35, and kube-proxy already logs a warning when it runs in that mode. Removal is now planned for Kubernetes 1.40. If any of your clusters run kube-proxy in IPVS mode, plan the switch now. It is a small change, but it affects the path every Service packet takes, so it deserves a proper test.
Why IPVS is going away
IPVS mode was added as an experiment to beat the performance of the iptables backend. It did that, but the kernel’s IPVS API never fitted the Kubernetes Services model well. According to the project, the backend never handled every Service edge case correctly, and it became complex and hard to maintain. The nftables backend was introduced in Kubernetes 1.29 and became GA in 1.33. It was designed from the start to replace both iptables and IPVS in the long run.
| Kubernetes version | IPVS status |
|---|---|
| 1.35 | Deprecated, kube-proxy logs a warning |
| 1.36 and later | Recommended point to move to nftables |
| 1.40 | IPVS mode planned for removal |
The full plan is in KEP-5495.
Check which mode you run
On kubeadm-style clusters, the mode is set in the kube-proxy ConfigMap. To check what you have:
kubectl -n kube-system get configmap kube-proxy -o yaml | grep -E '^\s*mode:'
kubectl -n kube-system logs ds/kube-proxy | grep -i proxier
On a node, sudo ipvsadm -Ln lists virtual servers only when IPVS is active. An empty mode: usually means the default (iptables), not IPVS. Managed services such as EKS, AKS and GKE may set the mode through their own add-on configuration, so check your provider’s settings as well.
Option 1: move to nftables (recommended)
The project recommends nftables if you run Kubernetes 1.36 or later and your nodes use a kernel of 5.13 or newer. nftables went GA in 1.33, but later performance and scaling fixes were not backported to releases before 1.36, so upgrade the cluster first if needed.
# check node kernels
kubectl get nodes -o wide
# change mode: ipvs to mode: nftables
kubectl -n kube-system edit configmap kube-proxy
# restart kube-proxy
kubectl -n kube-system rollout restart ds kube-proxy
kubectl -n kube-system rollout status ds kube-proxy
# confirm
kubectl -n kube-system logs ds/kube-proxy | grep -i proxier
If you set IPVS-specific options in the kube-proxy configuration, map them to the matching nftables settings in the kube-proxy configuration reference. Do not just delete them.
Option 2: move to iptables
Use iptables if you cannot run nftables yet, for example on an older Kubernetes release or older node operating systems. The config change is the same, with mode: iptables. For technical reasons, though, the project recommends rebooting each node after the change rather than only restarting kube-proxy. The iptables backend has improved a lot since 2017, but it is still not as fast as IPVS on very large clusters.
Migration checklist
- List every cluster and record its kube-proxy mode and Kubernetes version.
- Check node kernel versions (5.13 or later for nftables).
- Find tooling that depends on IPVS, such as
ipvsadmscripts, monitoring of thekube-ipvs0interface, or IPVS scheduler tuning. - Switch a staging cluster first and test Services, NodePorts, external traffic policies and session affinity.
- Make the change in production during a quiet window, and reboot nodes one at a time if you move to iptables.
- Update your infrastructure-as-code (kubeadm configs, Helm values, Terraform) so new clusters do not come up in IPVS mode.
Doing this well before 1.40 means it never blocks a cluster upgrade.
How TechProvidence can help
We run Kubernetes and DevOps work for small teams: cluster upgrades, kube-proxy and CNI migrations, and infrastructure-as-code cleanup on AWS and self-hosted clusters. To have your clusters checked or migrated, contact us.
Source: Kubernetes Blog


