The kube-proxy ipvs backend, first added as a performance experiment, is scheduled for removal in Kubernetes 1.40 after being deprecated in 1.35. Engineers who rely on this mode need to plan a migration to either the nftables or iptables backends to keep their clusters functional and performant.
Background: why the IPVS backend existed
Early versions of kube-proxy used iptables rules that grew linearly with the number of services. This caused two scaling issues: data‑plane latency, because each new connection traversed an O(n) rule set, and control‑plane latency, because any service or endpoint change triggered a full rewrite of the ruleset, consuming a full CPU core in large clusters. The ipvs backend leveraged the kernel IPVS subsystem’s hash tables to provide constant‑time lookups and avoided full rule‑set rewrites by updating individual entries via the IPVS API.
Why the IPVS backend is being retired
Despite its performance gains, the ipvs implementation never achieved full parity with iptables. Maintaining the code proved difficult after the original authors moved on, and the backend inherited several limitations:
- It lacks batch‑update capabilities, making large‑scale changes slower than
iptablesin some scenarios. - To support advanced Service features, it falls back to
iptablesandipset, re‑introducing the scaling problems it was meant to avoid. - Red Hat’s plan to drop the iptables API from RHEL means the
ipvsbackend would lose its underlying dependency on that platform.
Operational implications and migration options
Practitioners should evaluate their current kube-proxy configuration. If --proxy-mode=ipvs is set, the cluster will emit a deprecation warning and will fail to start once v1.40 is adopted. Two supported alternatives exist:
- nftables backend: Introduced experimentally in v1.29 and promoted to GA in v1.33, it offers a modern rule engine that replaces both
iptablesandipvs. It eliminates the need for a separate iptables fallback and aligns with the kernel’s nftables subsystem. - iptables backend: The traditional mode remains fully supported. It continues to provide correct Service semantics, though large clusters may still experience the original scaling constraints.
Migration steps generally include updating the kube-proxy command line or ConfigMap to set --proxy-mode=nftables (or iptables), validating Service behavior in a staging environment, and monitoring CPU and latency metrics during the transition.
Related CloudNinjas coverage: Kubernetes.
What This Means For Practitioners
Teams should treat the upcoming removal as a hard deadline for any production workloads still using ipvs. The practical actions are:
- Audit clusters for the
ipvsflag and document where it is used. - Test the
nftablesbackend in a non‑production cluster to confirm compatibility with existing Service configurations. - If
nftablesis not viable, verify that theiptablesbackend meets performance expectations before the upgrade. - Plan the kube-proxy configuration change to coincide with the next minor version upgrade that includes v1.40.
By completing these steps before the v1.40 release, engineers can avoid unexpected service outages and retain the performance characteristics they expect from their Kubernetes networking stack.


