I realized, through the official mongodb guides for replicas on top of Kubernetes, that you have to announce a new service for each replica member.
So, for the basic set of replicas, you declare 3 services, one primary, two secondary - service__primary, service__secondary_1, service_secondary_2.
Each of these roaming services then only proxies / balances the load into one container, which starts the mongod process.
source share