← Lessons

quiz vs the machine

Gold1370

Machine Learning

Canary Model Rollout

Release a new model to a small slice of traffic and widen it only if it stays healthy.

4 min read · core · beat Gold to climb

What it is

A canary rollout releases a new model to a small fraction of live traffic first, such as one percent. You watch its health metrics, and only if they look good do you gradually raise its share until it takes all traffic.

The steps

  • Route a small slice of requests to the new model and the rest to the old one
  • Monitor error rate, latency, and key prediction metrics on the canary slice
  • If everything is healthy, increase the slice in stages
  • If something breaks, route all traffic back to the old model

Why it limits blast radius

The point is to bound the damage. If the new model is bad, only the small canary slice is affected, and you can roll back fast. This is far safer than flipping every user to a new model at once.

Canary versus other strategies

A canary differs from a shadow deployment because the canary actually serves users, so it can measure real impact. It differs from an A B test because its main goal is safety and gradual exposure, not a careful statistical comparison.

Key idea

A canary rollout exposes a new model to a small traffic slice first so failures stay small and easy to undo.

Check yourself

Answer to earn rating on the learn ladder.

1. What is the main goal of a canary rollout?

2. How does a canary differ from a shadow deployment?