The logic of DetermineReplicas() in the cluster-ingress-operator has not been updated since SNO was created. However, since then we began supporting compact clusters with a single worker (i.e. 3 masters are schedulable and there is 1 additional worker node) (MGMT-8987).

If a cluster is created with this topology on day 1, it gets the InfrastructureTopology SingleReplica (OCPBUGS-64793). As a result, only a single Ingress pod is created, even though there are 4 nodes that it could be validly scheduled to.

A 3+1 cluster without mastersSchedulable is not a supported topology, but it may be possible to create a cluster with this topology since no validation prevents it and we will end up running only a single Ingress instance. Notably, this will remain the case even if the cluster is scaled out to multiple worker nodes - Ingress will remain non-HA.

The cluster-ingress-operator should be modified to only use single-replica mode on SNO:

if topology == configv1.SingleReplicaTopologyMode && infraConfig.Status.ControlPlaneTopology == configv1.SingleReplicaTopologyMode {
    return 1
}

Arguably we should also set the DefaultPlacement to ControlPlane by default on all compact clusters as well? (That would also have the effect of eliminating this issue at least for users who follow the default.)