On EKS the Traefik Helm values matter more than the chart name: enable the kubernetesCRD provider for the namespaces that actually own IngressRoute objects, pin an ingressClass as default only if nothing else claims it, and let the Service annotation create a single internet-facing NLB that forwards 80/443 into the Traefik pods. Terminate TLS with ACM on the NLB or on Traefik’s websecure entrypoint consistently — mixing both without a clear SNI story produces intermittent handshake failures that look like random 502s. Keep dashboard and API behind a Host rule that never shares the public app Host, and publish Prometheus metrics so you can see router/service churn after each deploy.
Ops prove-out after install: kubectl get svc should show the NLB hostname, then curl the public Host and expect the upstream status, not Traefik’s default backend. If client source IPs look wrong behind a nested NLB or Global Accelerator path, check proxy protocol / client IP preservation on the NLB target group before blaming Traefik middleware. When routers vanish for one namespace only, verify the provider namespace allow-list and that the IngressRoute’s entryPoints match websecure — most “Traefik broke after Helm upgrade” incidents are provider scope or TLS ownership mismatches, not the IngressRoute DSL itself.