OpenShift 4 - Intermittent 503 from all routes or timeouts when calling to backends directly when `allow-from-openshift-ingress` network policy is applied

Issue

• Intermittent 503 errors from backend routes when curled from clients to pods in namespaces where allow-from-openshift-ingress is applied.
• Reviewing the access logs from router-default pods in the Openshift-ingress namespace indicates that the pods are constantly being marked DOWN/unavailable then coming back up.
• Curling to a target backend directly from a host node indicates that intermittently the packet will time out/drop, resulting in a TCP connection timeout.