From everything I've read, my LoadBalancer service should be random. The sessionAffinity is None and I assume it's using the default iptables random algorithm. Despite this, I've noticed that the actual affinity is sticky to the client IP. Only after some time has past will a different pod be used for a specific client. This is very annoying as all my posts hit the same pod. Each post is for a long running operation that gets queued per-pod. Ideally each pod would have about the same number of operations queued, but only a single pod is getting posts using the default configuration.
I know about the kube-proxy parameters to change to ipvs, but I don't know where in the AKS cluster lifecycle I can set the kube-proxy parameters.
How do I switch the AKS cluster to use the IPVS round-robin load balancer algorithm?
