Blue/green versus canary deployments

WHAT THIS TESTS The interviewer checks whether you can match a release strategy to risk tolerance, cost, and the observability your platform provides.

A GOOD ANSWER COVERS Blue/green keeps two complete environments. Blue serves production while green holds the new release; once green is verified, you flip all traffic to it in a single switch, and rollback is simply flipping back. Canary releases the new version to a small slice of traffic, say one to five percent, then gradually increases the percentage while watching error rates and latency, promoting only if the metrics stay healthy. The core difference is granularity: blue/green is an all-at-once cutover, canary is incremental exposure.

COMMON WRONG ANSWERS Reversing the definitions. Saying canary requires two environments, or that blue/green exposes only a subset of users. Ignoring that canary fundamentally depends on good real-time metrics to decide whether to proceed.

LIKELY FOLLOW-UPS How do you handle database schema changes under each strategy. What signals trigger an automatic canary rollback. How does session affinity complicate either approach.

ONE CONCRETE EXAMPLE On a PaaS that supports traffic splitting, a canary deploy might send five percent of requests to the new revision, watch the error budget for ten minutes, then ramp to twenty-five, fifty, and one hundred percent. Blue/green on the same platform spins up a full second environment, runs smoke tests, then repoints the router. The trade-offs: blue/green gives the fastest, cleanest rollback and simple reasoning, but you pay for double capacity during the cutover and a bad release hits everyone the instant you flip. Canary limits blast radius to a small fraction of users and avoids duplicate full environments, but it requires reliable per-revision metrics, traffic-splitting support, and longer total rollout time. Choose blue/green when you value instant rollback and can afford the duplicate footprint; choose canary when you want to catch regressions on a small audience first.