Canary Analysis
ELI5 — The Vibe Check
Named after canaries in coal mines — you send a small version of your new deployment into production first to see if it dies. Route 5% of traffic to the new version, monitor it closely, and if everything looks good, gradually increase. If the canary keels over, you pull it out before anyone else gets hurt.
Real Talk
An automated deployment validation technique that routes a small percentage of production traffic to a new version and compares its metrics (error rates, latency, resource usage) against the baseline. Tools like Kayenta or Flagger automate the statistical comparison and promotion/rollback decisions.
When You'll Hear This
"Canary analysis caught a memory leak — the new version's p99 latency was 3x worse than baseline." / "Our canary pipeline promotes automatically if error rates stay within 0.1% of the baseline after 30 minutes."
Related Terms
A/B Testing
Show half your users a blue button and the other half a green button, then see which one gets more clicks. That's A/B testing — science experiments on your
Blue-Green Deployment
Blue-Green Deployment is like having two identical restaurants — one is open (blue), and you're setting up the other (green) with the new menu.
Feature Flag Service
Feature flags are like light switches for your code. You deploy new features turned OFF, then flip the switch for 1% of users, then 10%, then everyone. If