Why a service mesh
A service mesh moves networking concerns — encryption between services, retries, timeouts, traffic splitting, per-request metrics and access rules — out of application code and into a layer of proxies the platform team runs. It is powerful and also one of the most over-adopted technologies in Kubernetes, so interviewers want both: how it works (sidecars, ambient mode, mTLS identities, Envoy) and when you would not use one. This topic covers the architecture, the data-plane models, Istio traffic management and security, Linkerd and Cilium as alternatives, and how to operate and debug a mesh.
The problems it solves
Once you have dozens of services calling each other, every team has to solve the same networking problems:
| Need | Without a mesh | With a mesh |
|---|---|---|
| Encryption and identity between services | Each app manages TLS certificates | Automatic mutual TLS with rotated workload certificates |
| Timeouts, retries, circuit breaking | A library per language, configured differently everywhere | Policy applied by the proxy, the same for every language |
| Canary and traffic splitting | Separate Deployments plus load balancer tricks | Weighted routes: 95% v1, 5% v2, or by header |
| Per-request metrics | Each team instruments differently | Uniform request rate, errors and latency for every call |
| Service-to-service authorisation | Network policies by IP, or app code | Rules by workload identity: "only checkout may call payments on POST /charge" |
The mesh gives the platform team one consistent place to enforce these, and app developers stop shipping networking code.
Architecture: data plane and control plane
- The data plane is the set of proxies that every request passes through. In Istio these are Envoy proxies (or the lighter ztunnel in ambient mode); in Linkerd, a small Rust proxy.
- The control plane (Istio's
istiod, Linkerd's control plane) watches Kubernetes for services, endpoints and mesh policies, issues certificates to each workload, and pushes configuration to the proxies. - Envoy receives configuration dynamically through the xDS APIs (listeners, routes, clusters, endpoints, secrets), so routing changes take effect in seconds without restarts.
If the control plane goes down, existing proxies keep working with their last configuration; what stops is new configuration, new certificates and new workloads joining. That is why istiod is run with several replicas.