AWS for DevOps — Deployment Strategies

6.4 — Deployment Strategies

Choosing the right deployment strategy is the difference between a seamless release and a 3 a.m. incident. AWS CodeDeploy and ECS support several strategies out of the box.

Strategy Comparison

Strategy	Downtime	Rollback Speed	Risk	Best For
In-place	Brief	Slow (re-deploy old version)	Medium	Dev/staging, cost-sensitive
Rolling	Zero	Medium	Low–Medium	Fleet of EC2 instances
Blue/Green	Zero	Instant (switch back)	Low	Production, ECS, Lambda
Canary	Zero	Fast (shift traffic back)	Very Low	High-traffic production
All-at-once	Brief	Slow	High	Non-critical environments

In-Place Deployment

Each instance is updated one at a time (or in batches). The old version is stopped, new version installed.

  Instance A   Instance B   Instance C
  ┌──v1──┐    ┌──v1──┐    ┌──v1──┐
  │      │    │      │    │      │
  └──────┘    └──────┘    └──────┘
      │
      ▼  (take offline, update)
  ┌──v2──┐    ┌──v1──┐    ┌──v1──┐
  │  ✓   │    │      │    │      │
  └──────┘    └──────┘    └──────┘
                  │
                  ▼  (take offline, update)
  ┌──v2──┐    ┌──v2──┐    ┌──v1──┐
  │  ✓   │    │  ✓   │    │      │
  └──────┘    └──────┘    └──────┘

$ aws deploy create-deployment-group \
    --application-name my-app \
    --deployment-group-name staging \
    --deployment-config-name CodeDeployDefault.OneAtATime \
    --deployment-style deploymentType=IN_PLACE,deploymentOption=WITHOUT_TRAFFIC_CONTROL

Blue/Green Deployment

A complete parallel environment is created. Traffic shifts from old (blue) to new (green) after validation.

  Load Balancer
       │
  ┌────┴────┐
  │         │
  ▼         ▼
 Blue      Green
 (v1)      (v2)
 ┌───┐    ┌───┐
 │ A │    │ D │   ◀── new instances provisioned
 │ B │    │ E │
 │ C │    │ F │
 └───┘    └───┘

  Step 1: Green instances created, v2 deployed
  Step 2: Health checks pass on Green
  Step 3: ALB shifts traffic Blue → Green
  Step 4: Blue instances terminated (or kept for rollback)

CodeDeploy (EC2 Blue/Green):

$ aws deploy create-deployment-group \
    --application-name my-app \
    --deployment-group-name production \
    --deployment-style deploymentType=BLUE_GREEN,deploymentOption=WITH_TRAFFIC_CONTROL \
    --blue-green-deployment-configuration \
        terminateBlueInstancesOnDeploymentSuccess={action=TERMINATE,terminationWaitTimeInMinutes=60},\
        deploymentReadyOption={actionOnTimeout=CONTINUE_DEPLOYMENT,waitTimeInMinutes=0},\
        greenFleetProvisioningOption={action=COPY_AUTO_SCALING_GROUP} \
    --load-balancer-info targetGroupInfoList=[{name=my-app-tg}]

ECS Blue/Green (built-in):

$ aws ecs create-service \
    --cluster production \
    --service-name my-app \
    --task-definition my-app:3 \
    --deployment-controller type=CODE_DEPLOY \
    --load-balancers targetGroupArn=arn:aws:elasticloadbalancing:...:targetgroup/blue-tg/abc,containerName=my-app,containerPort=8080

Canary Deployment

Traffic is shifted in two increments: a small percentage first (canary), then the rest after validation.

  100% traffic
       │
  ┌────┴────┐
  │    90%  │ 10%
  ▼         ▼
 v1        v2  (canary)
 ┌───┐    ┌───┐
 │ A │    │ D │   ◀── small slice first
 │ B │    │   │
 │ C │    │   │
 └───┘    └───┘

  After X minutes (if healthy):

       0%  │ 100%
            ▼
           v2
          ┌───┐
          │ D │   ◀── full traffic shift
          │ E │
          │ F │
          └───┘

CodeDeploy Canary Configs:

Configuration	Behaviour
CodeDeployDefault.ECSCanary10Percent5Minutes	10% for 5 min, then 100%
CodeDeployDefault.ECSCanary10Percent15Minutes	10% for 15 min, then 100%
CodeDeployDefault.LambdaCanary10Percent5Minutes	Same pattern for Lambda

Custom canary config:

$ aws deploy create-deployment-config \
    --deployment-config-name Canary25Percent10Minutes \
    --compute-platform ECS \
    --traffic-routing-config \
        type=TimeBasedCanary,\
        timeBasedCanary={canaryPercentage=25,canaryInterval=10}

Linear Deployment

Traffic shifts in equal increments at regular intervals.

Configuration	Behaviour
CodeDeployDefault.ECSLinear10PercentEvery1Minute	+10% every minute
CodeDeployDefault.ECSLinear10PercentEvery3Minutes	+10% every 3 minutes
CodeDeployDefault.LambdaLinear10PercentEvery1Minute	Same for Lambda

Choosing a Strategy — Decision Flow

  Can you afford downtime?
       │
    Yes │         No
       │          │
       ▼          ▼
  All-at-once   Do you need instant rollback?
  or In-place        │
                  Yes │         No
                     │          │
                     ▼          ▼
               Blue/Green    Rolling
                     │
                     ▼
              Is this high-traffic production?
                     │
                  Yes │         No
                     │          │
                     ▼          ▼
                  Canary    Blue/Green

Rollback Comparison

Strategy	Rollback Method	Time to Rollback
In-place	Re-deploy previous revision	Minutes (full deploy cycle)
Rolling	Re-deploy previous revision	Minutes
Blue/Green	Switch traffic back to Blue	Seconds
Canary	Stop traffic shift, route all to v1	Seconds

Common Mistakes

Mistake	Why It Hurts	Fix
Using All-at-once in production	Full outage if deploy fails	Use Blue/Green or Canary
Blue/Green without health checks	Bad version receives 100% traffic	Configure ALB health checks
Terminating Blue too fast	Can’t rollback if issues found later	Keep Blue for 30-60 minutes
No CloudWatch alarms on canary	Canary proceeds even if errors spike	Set deployment alarms

Quick Reference

Task	Command
In-place deploy	--deployment-style deploymentType=IN_PLACE
Blue/Green deploy	--deployment-style deploymentType=BLUE_GREEN
Create canary config	aws deploy create-deployment-config --traffic-routing-config type=TimeBasedCanary,...
Create linear config	aws deploy create-deployment-config --traffic-routing-config type=TimeBasedLinear,...
Stop deployment	aws deploy stop-deployment --deployment-id ID

---

Next → GitHub as Source