← Back to knowledge center

After Ingress-NGINX retirement: migrating to Gateway API

Ingress-NGINX support ended in March 2026. Existing installations continue to route traffic, but no longer receive fixes or security updates. Migration is a behavioural engineering task, not a manifest rename.

V
Virtek Engineering BoardSystems integration and operations

Discover actual use

Inspect every cluster, namespace, Helm release and GitOps repository. Inventory IngressClass objects, controllers, external load balancers, certificates and DNS. Export annotations separately: rewrites, authentication, CORS, rate limits, canaries, snippets, timeouts and body limits often define the real service.

Record an owner, criticality, domain, backend, TLS, external address, source-IP requirement and SLO for every route. An ownerless Ingress is a migration risk, not an automatic conversion candidate.

Select a Gateway API implementation

Gateway API defines shared resources, while the dataplane and extensions remain implementation-specific. Compare route support, policies, cloud load balancers, WAF, service mesh, observability and vendor lifecycle.

Verify supported profiles and CRD versions. Let the platform team manage GatewayClass, the infrastructure boundary own Gateway and application teams own Routes. This separates authority more cleanly than a common annotation set.

Translate behaviour, not YAML

Ingress2Gateway can produce a starting point but cannot prove equivalence for controller-specific annotations. Build a mapping table with the old rule, new resource or policy, expected behaviour, test and exception owner.

Test path precedence, regex and rewrites, headers, client IP, WebSocket and gRPC, sticky sessions, mTLS, request limits, redirects and error handling. Do not copy snippets blindly; convert them into policies, filters or application behaviour.

Migrate in parallel

Run the new controller beside the old one with a separate address. Execute synthetic and integration tests, then shift DNS with a controlled TTL or distribute a traffic percentage through a load balancer. Compare responses, latency, TLS, backend metrics and security logs.

Define stop criteria, rapid rollback and an observation window. Move critical routes in groups so a shared-policy error does not affect every service at once.

Retire safely

Confirm that no traffic reaches the old address, remove public endpoints and obsolete permissions, and retain configuration and test evidence. Update templates, CI policy and the platform catalogue so new Ingress objects do not return.

Success means more than reachable applications: authority is clear, the implementation is maintained, tests are repeatable and the Gateway API upgrade path is owned.

Need an architecture
for your workload?

We will review inputs, risks and constraints, then propose a reasoned solution.

Talk to an engineer↗︎