Current behaviour

APIcast first send a request to "system.master.". The DNS server would send immediately NXDOMAIN and hence it then queries "system-master.3scale.svc.cluster.local". The same happens for backend-listener.

However some customers have slow response times due to the fact that their DNS queries are being sent to an external DNS and they fail (servfail) because it doesn't exist. 

The workaround is to update the BACKEND_ENDPOINT_OVERRIDE and THREESCALE_PORTAL_ENDPOINT with http://backend-listener.<NAMESPACE NAME>.svc.cluster.local:3000 and http://system-master.<NAMESPACE NAME>.svc.cluster.local:3000 endpoints respectively instead of http://backend-listener:3000 and http://system-master:3000. See full solution here. 

Expected behaviour
By default APIcast should use http://backend-listener.<NAMESPACE NAME>.svc.cluster.local:3000 and http://system-master.<NAMESPACE NAME>.svc.cluster.local:3000 to avoid delays when external DNS servers are being used.