Canary and linear deployments

Automating Deployments on AWS

Dunieski Otano

Amazon Web Services Solutions Architect

One bad release, every user

• Payment-logic change deploys instantly to 100% of traffic
• Bug surfaces under live load; every checkout fails
• A 10% canary would have caught it in five minutes
• Cost of skipping the canary: real money lost

What is a canary deployment

• Send a small slice of traffic to the new version first
• Watch metrics for a fixed wait period
• If healthy, shift the rest all at once
• Common pattern: 10% for 5 minutes, then 100%

What is a linear deployment

• Shift traffic in equal increments
• Wait between each step
• Example: 10% every 1 minute for 10 minutes
• Smoother than canary; takes longer overall

Canary vs. linear, side by side

• Canary: one slice, then everything
• Linear: many small steps
• Canary catches issues fast; linear ramps gradually
• Both pause if a rollback is triggered

CodeDeploy configurations for Lambda

• CodeDeployDefault.LambdaCanary10Percent5Minutes: 10% for 5 min, then 100%
• CodeDeployDefault.LambdaLinear10PercentEvery1Minute: +10% per minute
• CodeDeployDefault.LambdaAllAtOnce: full traffic immediately
• ECS has matching ECS-prefixed configurations

Automatic rollback with CloudWatch alarms

• Attach CloudWatch alarms to the deployment group
• Common alarms: errors, duration, 5xx
• If the alarm enters the ALARM state, deployment stops
• CodeDeploy rolls back to the previous version

Picking the right configuration

• Routine low-risk patch → AllAtOnce
• High-risk production change → Canary
• Want a smooth steady ramp → Linear
• Always pair with alarms for auto-rollback

Let's practice!

Automating Deployments on AWS