If a route is configured with spec.port.to pointing to a specific service port, but the alternate backend service(s) specify a different port, the load balancing strategy defaults to roundrobin instead of the expected random.

Conversely, if you have a route and the spec.to service has a port that doesn't match spec.port.to, but the alternate backend has a valid port, it sets it to roundrobin instead of the expected random as well.

The cause is that createServiceAliasConfig function currently does not filter out service units that do not have matching ports when calculating activeServiceUnits.

This can potentially cause higher-than-normal HaProxy memory usage in these specific scenarios.

4.12-4.16    

    100%

    1. Create route with a single alternate backend service that has a different port than spec.port.to service
    2. Observe HaProxy config file contains "roundrobin" for this backend instead of the expected "random"    

    "roundrobin" algorithm for HaProxy backend

     "random" algorithm for HaProxy backend

    Miciah found the bug here: https://github.com/openshift/router/pull/576#issuecomment-2065129648