The Ingress Router pods are experiencing instability under high load. Specifically, whenever the configured maximum connection limit (maxconn) is saturated, the liveness probe fails and pods terminate and restart instead of gracefully throttling traffic.

latest

With an existing route, generate a load over maxconn, we used https://github.com/mparram/test-backend?tab=readme-ov-file#run-in-openshift    

    1. Install and configure https://github.com/mparram/test-backend?tab=readme-ov-file#run-in-openshift
    2. Set the maxconn to 2000 (optional)
    3. Scale the replicas enough to reach maxconn on haproxy (could be checked looking at haproxy_frontend_current_sessions metric)
    4. Review the haproxy behavior
    

The Kubelet restarts the HAProxy container following a liveness probe failure.

Haproxy gracefully throttling traffic.