Skip to content
Tutorials Center
My learning

Topics

Lesson 2 of 36 ยท Getting Started

Istio Architecture: istiod and Envoy

A tour of Istio's architecture: istiod, Envoy sidecars, ingress and egress gateways, the ports involved, and how a config change reaches every proxy.

Istio has two halves. The data plane is the set of proxies that touch every request. The control plane is one Deployment, istiod, that tells those proxies what to do. Knowing which half does what makes troubleshooting much faster: a wrong route is usually a config problem, a slow request is usually a data plane problem.

Control plane

istiod One Deployment, usually 2+ replicas.
Config Watches Kubernetes resources.
CA Issues workload certificates.
xDS Streams config to proxies.
1 / 2

What istiod does

Before Istio 1.5 the control plane was several services (Pilot, Citadel, Galley). They were merged into the single binary istiod. It has three jobs:

  1. Service discovery and configuration. It watches the Kubernetes API for Services, Endpoints, Pods and Istio resources, and builds a model of the mesh.
  2. Certificate authority. It signs short-lived certificates for each workload so proxies can prove their identity.
  3. Injection and validation. It runs webhooks that add the sidecar to new pods and reject invalid Istio YAML.

istiod does not carry application traffic. If it is down, existing proxies keep working with the configuration they already have; you just cannot push changes or get new certificates once the old ones expire.

The Envoy proxy

Envoy is a high-performance L4/L7 proxy written in C++. In a pod it is started by a small program called pilot-agent, which also fetches certificates, health-checks the proxy, and merges metrics. The container is named istio-proxy.

Ports you will meet:

Port Purpose
15001 Outbound capture: traffic leaving the app is redirected here
15006 Inbound capture: traffic arriving at the pod is redirected here
15021 Health check used by Kubernetes readiness
15020 Merged Prometheus metrics from agent, Envoy and (optionally) the app
15090 Envoy's own Prometheus metrics
15012 istiod's gRPC xDS endpoint with mTLS
15014 istiod monitoring and debug
15017 istiod webhook endpoint

The inbound and outbound redirect is done with iptables rules installed when the pod starts, either by an init container (istio-init) or by the Istio CNI plugin, which avoids giving pods the NET_ADMIN capability.

Gateways

A gateway is just Envoy running as a standalone Deployment instead of as a sidecar. The ingress gateway receives traffic from outside the cluster. An egress gateway is an optional exit for traffic leaving the mesh. Both are configured with the same Istio resources as sidecars. Chapter 6 covers them in depth.

How a change reaches a proxy

You apply a VirtualService. The Kubernetes API stores it. istiod, watching the API, notices, recomputes configuration for every affected proxy, and pushes it over xDS, Envoy's streaming discovery API. Envoy applies it without restarting or dropping connections.

sequenceDiagram
    participant U as You (kubectl)
    participant K as Kubernetes API
    participant D as istiod
    participant E as Envoy sidecars
    U->>K: apply VirtualService
    K-->>D: watch event
    D->>D: recompute config
    D->>E: push over xDS (gRPC)
    E-->>D: ACK

Config flows from your YAML to every affected proxy in about a second in a healthy mesh.

A model of config push

This stand-in models istiod as a store of routing rules with subscribed proxies. It is not Istio's code; it shows why a push model lets you change behaviour without restarting anything.

Python Stand-in: a control plane pushing config

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Python Stand-in: control plane down, data plane keeps working

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Reading the pieces on a real cluster

On a cluster with Istio installed you can see these parts with ordinary Kubernetes commands (shown read-only here):

Terminal
kubectl get pods -n istio-system
kubectl get pod web-7d9c -n shop -o jsonpath='{.spec.containers[*].name}'
istioctl proxy-status

The first lists istiod and any gateways. The second shows an application container plus istio-proxy. The third lists each proxy with whether its listener, cluster, route and endpoint config is in sync with istiod; Installing Istio with istioctl shows how to get there.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. Which of these is NOT a job of istiod?

    Traffic flows through the proxies of the data plane, never through istiod.
  2. 2. What is an ingress gateway?

    Gateways are standalone Envoy Deployments configured with Istio resources.
  3. 3. What happens to existing traffic if istiod becomes unavailable?

    Config is cached in each proxy; you lose changes and certificate renewal, not live traffic.
  4. 4. Which API does istiod use to deliver config to Envoy?

    xDS is Envoy's discovery API; istiod streams updates over gRPC on port 15012.

Summary

  • Control plane: istiod (discovery, CA, webhooks). Data plane: Envoy sidecars and gateways.
  • Proxies capture traffic on ports 15001 (out) and 15006 (in) using iptables rules.
  • Config travels as xDS over gRPC; changes apply without restarts.
  • If istiod is down, traffic keeps flowing, but config and certificate changes stall.

Next: Installing Istio with istioctl