In-place and blue/green deployments

Automating Deployments on AWS

Dunieski Otano

Amazon Web Services Solutions Architect

The 2 AM outage

• All-at-once deploy goes wrong
• Every server runs the broken version
• Customers see errors for an hour
• Rollback takes another full deploy

Why deployment strategy matters

• A bad release can take a service down
• All-at-once = fastest, riskiest
• Gradual rollout = slower, safer
• The strategy controls blast radius if it fails

In-place deployments

• Deploy onto the same instances that serve traffic
• One batch at a time: half, then half; or one by one
• Health-check each batch before moving on
• Cheap; no extra infrastructure needed

Blue/green deployments

• Blue = current production fleet
• Green = brand-new fleet with the new version
• Load balancer shifts traffic from blue to green
• Old fleet stays available for fast rollback

Lambda and ECS are blue/green by design

• Lambda: traffic shifts via alias version pointer
• ECS: traffic shifts via replacement task set
• No EC2-style fleet duplication required
• You still pick a deployment configuration

Choosing in-place vs. blue/green

• Tight budget, low-risk patch → in-place
• Need zero downtime or instant rollback → blue/green
• Single legacy server → in-place is the only option
• Lambda or ECS → blue/green is built in

Rollback in CodeDeploy

• Rollback = redeploy the previous successful revision
• Trigger automatically on deployment failure
• Trigger automatically on CloudWatch alarm
• Blue/green rollback: flip the load balancer back

Let's practice!

Automating Deployments on AWS