Architectural · Deployment · DWG NO. 056

Blue-Green Deployment

Run two identical production environments and switch traffic between them all at once, for releases with zero downtime and instant rollback.

TypeDeployment

ScopeWhole environment

ComplexityMedium

Common inZero-downtime releases, environments where instant rollback matters

01 · The problem

Deploying a new version in place risks downtime and makes rollback slow

Replacing the running version of a service in place means a window where it's unavailable or serving a half-updated mix of instances. If the new version has a serious problem, rolling back means another in-place deploy — under pressure, with the same risk the original deploy had.

02 · Structure

Two full environments, one router deciding which is live

"Blue" is the currently live environment; "green" is an idle, fully deployed copy running the new version. Once green is verified healthy, the router switches all traffic to it instantly. Blue stays running, idle, ready to receive traffic back immediately if a problem appears.

03 · Reference implementation

A router switching all traffic to the new environment at once

`switchTraffic` is the entire release — an instant cutover, and just as instant to reverse by switching back.

blue-green-router.js

const environments = {
  blue: { version: "v1", handle: () => "response from v1" },
  green: { version: "v2", handle: () => "response from v2" },
};

let live = "blue";

function route(request) {
  return environments[live].handle();
}

function switchTraffic(target) {
  console.log(`router: verifying ${target} is healthy...`);
  live = target; // instant cutover — no partial rollout, no in-place replace
  console.log(`router: all traffic now routed to ${target} (${environments[target].version})`);
}

console.log(route());       // "response from v1"
switchTraffic("green");
console.log(route());       // "response from v2"

// Rollback, if green has a problem, is exactly as instant:
switchTraffic("blue");

04 · When to use it

Trade-offs

Zero-downtime cutover — The switch is effectively instantaneous from the client's point of view — no mixed-version window.
Rollback is just switching back — Blue stays fully running and ready, so reverting a bad release is as fast as the original deploy.
Double the infrastructure, briefly — Running two full production environments simultaneously costs more, if only for the release window — and database schema changes need care since both versions may briefly need to work against the same data.