Mastering Zero‑Downtime Deployments: Blue/Green, Canary, Rolling & More

1. Blue/Green Deployment

Blue/Green deployment is a widely used zero‑downtime technique that relies on two identical environments: one active (green) running the current version and one inactive (blue) where the new version is deployed and tested. After successful testing, traffic is switched to the blue environment, allowing a quick rollback to green if problems arise.

Advantages : the inactive environment can be updated without affecting production, risk is low, and rollback is fast because the old version remains untouched.

Drawbacks : requires extra infrastructure, handling of in‑flight requests, possible database synchronization issues, and higher resource costs.

Suitable scenarios : when you can afford a duplicate environment, need a clean cut‑over, and want to avoid downtime during upgrades.

2. A/B Testing

A/B testing is a scientific method for comparing two variants of a feature or UI by routing a portion of traffic to each version and measuring performance metrics. Unlike Blue/Green, its goal is to evaluate user impact rather than to safely release code, and it can be combined with Blue/Green deployments.

3. Canary (Gray) Release

Canary (or gray) release gradually introduces a new version to a small subset of users while the previous version remains fully operational. The process typically includes preparing artifacts, removing the canary server from the load balancer, deploying the new version, running automated tests, re‑adding the server, and, if successful, rolling out to the rest of the fleet.

Suitable scenarios : when you want to keep the old version running, need incremental rollout, or wish to combine with A/B testing to evaluate multiple options.

4. Rolling Update

Rolling update updates instances in batches (e.g., 20% at a time) until the entire cluster runs the new version. It saves resources compared to Blue/Green because only one set of instances is needed, but it lacks a guaranteed stable environment for testing and makes rollback more complex.

Typical drawbacks include uncertainty about the health of the partially updated cluster, difficulty rolling back mid‑process, and challenges when autoscaling changes the number of instances during deployment.

5. Red‑Black Deployment

Red‑Black (used by Netflix on AWS) creates a new server group with the new version via an Auto Scaling Group and Launch Configuration. If tests fail, the new group is discarded; if they pass, the Elastic Load Balancer is pointed to the new group and the old group is terminated. This approach provides zero‑downtime with immutable infrastructure.

Benefits include always‑online service and no need to keep duplicate environments running simultaneously.

Article source: twt企业IT社区 (copyright belongs to the original author).