OCP regression where EgressIP addresses respond to ALB health probes for ingress NodePorts, causing intermittent external connection failures

Issue

After upgrading an Azure Red Hat OpenShift (ARO) cluster to OpenShift Container Platform (OCP) 4.18.13, nodes assigned EgressIP addresses begin responding to Azure Load Balancer (ALB) health probes associated with ingress NodePort services.

As a result:

• The Azure Load Balancer incorrectly marks these EgressIP addresses as healthy backends.
• The ALB then forwards ingress traffic to these EgressIPs.
• Since no ingress router pods are listening on EgressIPs, the connections are immediately refused (TCP RST).

• Users experience intermittent “connection refused” errors when accessing routes exposed via the default ingress controller.

  Important:
  This behavior was not present in OCP 4.17.25, where only the primary node IPs responded to health probes and served ingress traffic.