Skip to content

Choosing an ingress-nginx replacement

ingress-nginx was retired in March 2026. Every cluster that ran it needs a successor, and there are several good ones. This page helps you decide, and it is honest about when wxingress is not the best fit.

The real question: rewrite or not?

What a switch costs is mostly not the controller. It is the Ingresses: every nginx.ingress.kubernetes.io/* annotation, every default your applications silently depend on (timeouts, buffering, retries, body size limits, HSTS), every Helm chart of a vendor that writes those Ingresses for you. The successors differ in how much of that you have to rewrite and test again.

Path What you rewrite Fits when
wxingress nothing, except Lua in snippets (to njs) you want the switch to be invisible to the applications, now
Another Ingress controller (Traefik, HAProxy, F5 NGINX Ingress Controller, cloud provider controllers) the annotations, into that controller's own, and possibly the defaults you already use that controller elsewhere, or need its specific features
Gateway API (Envoy Gateway, Istio, Cilium, NGINX Gateway Fabric, …) every Ingress, into Gateways and routes (tools such as ingress2gateway help) you plan the move to the Gateway API anyway and can afford to touch every route

Some controllers read a subset of ingress-nginx's annotations. Check which ones your Ingresses use (the migration guide has the command) against what the controller you consider supports, and test the defaults, not just the annotations.

What makes wxingress different

  • The compatibility is measured. 97 Ingresses from production clusters and popular Helm charts, probed 2562 ways against ingress-nginx and rke2-ingress-nginx: 2511 identical answers, and each of the 51 differences has a documented reason. The scripts are in the repository; run them against your own Ingresses. See Testing and benchmarks.
  • ingress-nginx's defaults, not just its annotations. An Ingress without annotations behaves as it did under ingress-nginx: the 5 s connect timeout, buffering off, three retries, HSTS, the 308 redirect, the WebSocket headers.
  • The operational surface stays. The IngressClass nginx, the admission webhook's checks, the Prometheus metric names your dashboards and alerts use, the TCP/UDP services ConfigMap, the default backend.
  • A switch you can undo. wxingress and ingress-nginx serve the same Ingresses at the same time, so traffic moves node by node or address by address, and back. There is a tested runbook for RKE2.
  • It is faster. No Lua in the request path: about 4 times the HTTP requests per second of ingress-nginx 1.15 on the same node, half the CPU per HTTPS request, half the memory under load.
  • The Gateway API is included. When you are ready to move routes to the Gateway API, the same controller serves Gateways and HTTPRoutes next to your Ingresses, and the ingress-nginx annotations even work on routes. You can migrate to the Gateway API later, one route at a time, instead of all at once. (The Gateway API side is tested against the conformance suite but, unlike the Ingress side, not yet in our production.)

When wxingress is not the best choice

  • You depend heavily on Lua. Small Lua snippets translate to njs well. If your platform runs large amounts of custom Lua (OpenResty plugins, custom balancers), a rewrite is real work, and an OpenResty-based gateway may suit you better.
  • You want a service mesh anyway. If you are introducing Istio or Cilium for other reasons, their Gateway API implementations may be the more consistent choice.
  • You need a vendor's commercial product. wxingress is community-driven open source with commercial support from WizardTales; if your organisation requires a specific vendor, follow that requirement.

Try it without risk

Install wxingress next to ingress-nginx, without taking over the IngressClass or any traffic, and compare the answers of both for every host of your cluster. The migration guide shows how; nothing changes for your users until you move traffic.