Kubernetes Internal Load Balancing

The problem

Our Kubernetes (K8s) clusters were greenfield. There were no established platform patterns yet, and internal services still needed predictable load balancing. Some traffic patterns needed more control than a default Service type could offer. I volunteered for this work on one of the first clusters we stood up.

What I implemented

HAProxy-based load balancers for internal Kubernetes workloads that needed explicit routing, health checks, or traffic shaping. This was early infrastructure: before we had much Kubernetes implementation to copy, so the load-balancing layer had to be designed and proven from scratch.

What I took away

Load balancing is invisible until it is wrong. Getting health checks, backend selection, and observability right upfront saves painful debugging when a service looks up but half the pods never see traffic.