Canary Deployment

Noun · Development

Definitions

  1. A deployment strategy that gradually rolls out changes to a small percentage of users before deploying to the entire fleet. Named after canaries in coal mines — if the canary (small deployment) dies (shows errors), you stop before exposing everyone.

    In plain English: Releasing new code to a small percentage of users first to see if it causes problems before rolling it out to everyone — testing in the real world with a safety net.

  2. The percentage of traffic routed to the canary typically starts at 1-5% and increases gradually if metrics look healthy. Key metrics to monitor: error rates, latency percentiles (p50, p95, p99), and business metrics like conversion rates.

    Example: 'The canary got 2% of traffic for an hour. Error rate spiked to 5% vs. the baseline 0.1%, so we automatically rolled back before any user noticed.'

    Source: operational detail

Origin Story

The deployment strategy named after coal mine birds

**Canary deployment** takes its name from the practice of bringing canaries into coal mines to detect toxic gases. If the canary died, miners knew to evacuate. In software, you route a small percentage of traffic to the new version and monitor for problems.

Unlike blue-green deployments (all-or-nothing switch), canary releases are gradual. You might start with 1% of traffic, then 5%, then 25%, then 100%, watching error rates and latency at each stage. If anything looks wrong, you roll back before most users are affected.

Google, Facebook, and Netflix pioneered canary deployments at scale. The technique requires good observability -- you need to detect that the 1% canary population has a 0.5% higher error rate, which is a tiny signal in noisy production data.

Coined by: Industry practice (named after coal mine canaries)

Context: 2000s, adopted widely by major tech companies

Fun fact: Real canaries were used in British coal mines from 1911 until 1986, when they were replaced by electronic gas detectors. The Royal Mining Museum reportedly received complaints from animal rights groups about the software industry's continued use of the term.

Related Terms