A canary deployment (named after the "canary in a coal mine") is a deployment strategy where a new version of a service is released to a small percentage of users or servers first. The team monitors the canary for errors, latency increases, or other anomalies before gradually rolling the change out to the entire fleet.
A typical canary process starts by deploying to 1-5% of traffic, then monitoring key metrics (error rate, latency, business metrics) for a defined observation period. If the metrics look healthy, traffic is gradually increased (10%, 25%, 50%, 100%). If problems are detected, the canary is rolled back without impacting most users.
Canary deployments require good observability to compare canary metrics against the baseline. They also benefit from external monitoring — a canary that looks fine in internal metrics might still cause issues detectable by synthetic monitoring from external locations. Hyperping can monitor canary endpoints to verify external availability during rollouts.
