Rolling, blue-green or canary?

Dimension	Rolling deployments	Blue-green deployments	Canary releases
In one line	Replace instances a batch at a time	Build a second environment, then switch all traffic to it	Give the new version a small share of traffic, then widen it
Extra capacity	None to one batch, depending on surge settings	A full second copy during the release (ECS warns it may double resource use)	A few extra instances, or none if the platform splits traffic between versions
Rollback speed	Slow: another rollout, batch by batch	Seconds: switch the router back, while the old side is still running	Fast: set the canary’s weight to 0
Versions live at once	Two, mixed, for the length of the rollout	One serving traffic; both running until the old side is scaled down	Two, side by side, for the whole rollout
Blast radius of a bad release	Each batch that passes health checks; a bug that health checks miss reaches everyone	Everyone, from the moment of the switch until you switch back	Only the canary’s share, if you catch it before widening
What it needs	Several instances, health checks or readiness probes, backward-compatible changes	Room for two environments, a switchable router or DNS, schema changes split into expand and contract	Weighted routing, good metrics compared against the stable version, enough traffic for a signal
Best for	Everyday releases of stateless services; the default strategy of a Kubernetes Deployment	Releases that need instant rollback or can’t run mixed for long	High-traffic services where a regression must reach as few users as possible
Built into	Kubernetes Deployments, ECS, EC2 Auto Scaling instance refresh, VM Scale Sets	ECS, CodeDeploy, App Service slots, Argo Rollouts	Lambda aliases, API Gateway, Cloud Run, CodeDeploy, ECS, Argo Rollouts, Flagger

Rule of thumb

Start with rolling deployments; they are the default for a reason and cost nothing extra. Add blue-green when a slow rollback would hurt more than paying for a second environment, and canaries when you have the traffic and the metrics to judge a new version on a slice of users. They combine: Amazon ECS canary deployments shift traffic between a blue and a green revision, and many teams canary first and then roll the rest. All three keep two versions alive for a while, so database changes still need expand and contract.

Rolling, blue-green or canary?

Rule of thumb

Sources