System Design Article

Service Mesh, Sidecar & Service Discovery

Difficulty: Hard

Once you have more than a handful of services, the cross-cutting concerns (mTLS, retries, circuit breaking, load balancing, traffic shifting, observability) start to dominate. Doing them in every service in every language is a maintenance nightmare. The sidecar pattern moves these concerns into a co-located proxy that runs next to your service, and a service mesh is the control plane that programs every sidecar in your fleet from one place. This lesson covers how a mesh actually works (data plane vs control plane, Envoy as the de-facto data plane, Istio and Linkerd as control planes), how service discovery underpins it, and the very real cost (latency tax, complexity, on-call burden) so you know when a mesh helps and when it is over-engineering.