Containers

AWS CodeDeploy now supports linear and canary deployments for Amazon ECS

AWS CodeDeploy has extended blue/green deployment support for Elastic Container Service (Amazon ECS) to include canary and linear deployments for applications hosted on AWS Fargate or Amazon Compute Cloud (Amazon EC2).

Blue/green deployments are a safe deployment strategy provided by AWS CodeDeploy for minimizing interruptions caused by changing application versions. This is accomplished by creating your new application environment, referred to as green, alongside your current application that is serving your live traffic, referred to as blue. This allows for a period of time for monitoring and testing of the green environment before your live traffic is routed from blue to green and subsequently turning off the blue resources.

After the initial launch of blue/green deployments for Amazon ECS, many customers expressed interest in having control over the amount of traffic that was shifting over a period of time with their application updates.  Linear and canary deployments through CodeDeploy provide this by limiting the exposure of live traffic to the new application version to a percentage of total traffic in order to monitor performance before routing remaining traffic with confidence. You can also set up Amazon CloudWatch alarms and if an issue is detected CodeDeploy will automatically reverse the traffic routing back to your original version.  CodeDeploy uses ALB’s Weighted Target Groups to make this possible.

In this post, we will show you how to configure the new linear and canary deployments for Amazon ECS hosted on Fargate.

Here’s what we will do today:

  1. Create ECS service with CodeDeploy as the controller
  2. Create a new deployment group using the new linear configuration
  3. Trigger a CodeDeploy blue/green deployment using a preset linear deployment configuration for routing 10 percent of traffic every one minute.

The following pre-defined linear and canary deployment configurations are available as of today.  You can also create custom defined linear and canary deployments starting today as well.

Deployment Configuration Description
CodeDeployDefault.ECSCanary10Percent5Minutes Shifts 10 percent of traffic in the first increment. The remaining 90 percent is deployed 5 minutes later.
CodeDeployDefault.ECSCanary10Percent15Minutes Shifts 10 percent of traffic in the first increment. The remaining 90 percent is deployed 15 minutes later.
CodeDeployDefault.ECSLinear10PercentEvery1Minutes Shifts 10 percent of traffic every minute until all traffic is shifted.
CodeDeployDefault.ECSLinear10PercentEvery3Minutes Shifts 10 percent of traffic every three minutes until all traffic is shifted.

Let’s get started.

Assumptions

  • All activities and resources referenced within this blog are to be executed and created within and AWS account and within the same Region
  • Docker is installed on your local development system. Get Docker here
  • The latest AWS CLI installed, as we will use this to push our container images to Amazon ECR.

Implementation

Step 1:  Pull the GitHub repo that supports this blog

Pull the GitHub repository at github.com/aws-samples/aws-codedeploy-linear-canary-deployments-blog. This repository contains the CloudFormation template and the example Docker container to build for this post.  The following will pull the master branch to your machine.

git clone git@github.com:aws-samples/aws-codedeploy-linear-canary-deployments-blog.git

Step 2: Create an Amazon ECR repository

We will do a little upfront housekeeping that will allow us to create two versions of a container to illustrate updating between versions using a linear deployment.  Create a repository using the create-repository command:

aws ecr create-repository \
--repository-name ecs-sample-app \
--region us-east-2

The output includes the URI of the repository, with the following format:

111111111111.dkr.ecr.us-east-2.amazonaws.com/ecs-sample-app

Step 3: Build and push blue/green container images to the Amazon ECR repository

Using the pulled repo, move into the Docker directory and execute the following commands, transposing your account id where necessary:

docker build -t ecs-sample-app .
docker tag ecs-sample-app:latest 111111111111.dkr.ecr.us-east-2.amazonaws.com/ecs-sample-app:v1
aws ecr --region us-east-2 get-login --no-include-email --registry-ids 111111111111| bash
docker push 111111111111.dkr.ecr.us-east-2.amazonaws.com/ecs-sample-app:v1

You can now change branches and perform the same tasks above for a second image version (v2).  Use the following git command to check out the v2 branch of the repo:

git fetch && git checkout v2

Perform the same actions as above rebuilding the local container, tagging the container (switch v1 for v2), and push them to the repository.

You can now observe the containers in the repository.  Note the image URIs.

ECR repository listing

Image 1: ECR repository listing

Step 4: Creating infrastructure using AWS CloudFormation

Launching the following CloudFormation (below diagram), we will bring up our environment including a VPC and supporting resources.  Below is a diagram of the VPC architecture that will be deployed.

Open the CloudFormation management console and select With new resources (standard) in the Create stack pull-down menu. Make sure you’re in the same AWS region which you used in the previous steps. In the next window, you’ll select the Upload a template file radio button and upload the cloudformation/linear_ecs.yaml file in the git repository you cloned in the first step.

Provide a value for the ImageUrl parameter, leave all other values defaulted.  This value corresponds to the first container image (v1) URI from Step 3.  Accept the defaults to proceed through next screens and IAM acknowledgment to create stack.  Once the stack is deployed (CREATE_COMPLETE), navigate to the outputs section and gather the outputs for reference.  We will use these moving forward.  The output will look something like this:

CloudFormation output listing

Image 3: CloudFormation output listing

Step 5: Creating an Amazon ECS service for blue/green deployments

Now that we have the underlying infrastructure, we are ready to create an Amazon ECS service with blue/green deployment support from CodeDeploy.  We will use the CLI to create this.

Within the pulled git repo, navigate to the json_files directory and edit the create_service.json to replace the following values with the corresponding values from your CloudFormation outputs.

Here is a transposed example version of this given the above outputs:

{
	"cluster": "ecs-blog-ECSCluster-1BC08Z8YA1A5X",
	"serviceName": "ecs-blog-svc",
	"taskDefinition": "arn:aws:ecs:us-east-2:111111111111:task-definition/ecs-blog-svc:1",
	"loadBalancers": [
		{
			"targetGroupArn": "arn:aws:elasticloadbalancing:us-east-2:111111111111:targetgroup/ecs-b-Targe-NP1KC1KYAEK1/01cf50be3fa1524c",
			"containerName": "ecs-blog-svc",
			"containerPort": 80
		}
	],
	"launchType": "FARGATE",
	"schedulingStrategy": "REPLICA",
	"deploymentController": {
		"type": "CODE_DEPLOY"
	},
	"platformVersion": "LATEST",
	"networkConfiguration": {
		"awsvpcConfiguration": {
			"assignPublicIp": "ENABLED",
			"securityGroups": ["sg-08fdb4f721268b4f0"],
			"subnets": ["subnet-0d348d95b7121ae37", "subnet-0075725bb118146dd"]
		}
	},
	"desiredCount": 2
}

Once you have transposed your output values, execute the following from within the json_files directory in your repository:

aws ecs create-service \
--cli-input-json file://create_service.json \
--region us-east-2

The output includes the ARN of the service, with the following format.

arn:aws:ecs:us-east-2:111111111111:service/ecs-blog-svc

Step 6: Create the CodeDeploy Resources

Use the following steps to create your CodeDeploy application and the CodeDeploy deployment group.

To create CodeDeploy resources

  1. Use the create-application command to create a CodeDeploy application, specifying the ECS compute platform:
aws deploy create-application \
--application-name ecs-blog-app \
--compute-platform ECS \
--region us-east-2

The output includes the application ID, with the following format:

{
	"applicationId": "09ac62f6-a64d-4ffe-a890-207c8af7494d"
}
  1. Use the create-deployment-group command to create a CodeDeploy deployment group.

Within the pulled git repo, navigate to the json_file, edit the code_deployment_group.json to replace the following values with the corresponding values from your CloudFormation outputs.

Here is a transposed example version of this given the above outputs:

{
	"applicationName": "ecs-blog-app",
	"deploymentGroupName": "ecs-blog-app-dg",
	"deploymentConfigName": "CodeDeployDefault.ECSLinear10PercentEvery1Minutes",
	"serviceRoleArn": "arn:aws:iam::111111111111:role/ecs-blog-EcsRoleForCodeDeploy-176BC3DEKYQ0E",
	"deploymentStyle": {
		"deploymentType": "BLUE_GREEN",
		"deploymentOption": "WITH_TRAFFIC_CONTROL"
	},
	"blueGreenDeploymentConfiguration": {
		"terminateBlueInstancesOnDeploymentSuccess": {
			"action": "TERMINATE",
			"terminationWaitTimeInMinutes": 5
		},
		"deploymentReadyOption": {
			"actionOnTimeout": "CONTINUE_DEPLOYMENT"
		}
	},
	"loadBalancerInfo": {
		"targetGroupPairInfoList": [{
			"targetGroups": [{
					"name": "ecs-b-Targe-NP1KC1KYAEK1"
				},
				{
					"name": "ecs-b-Targe-1PFVGY72CEI09"
				}
			],
			"prodTrafficRoute": {
				"listenerArns": [
					"arn:aws:elasticloadbalancing:us-east-2:111111111111:listener/app/ecs-b-Publi-14GAO5UFNWDMZ/24e50a3e4a4b205c/8f817f85ec2fe816"
				]
			},
			"testTrafficRoute": {
				"listenerArns": [
					"arn:aws:elasticloadbalancing:us-east-2:111111111111:listener/app/ecs-b-Publi-14GAO5UFNWDMZ/24e50a3e4a4b205c/b511569563eb2261"
				]
			}
		}]
	},
	"ecsServices": [{
		"serviceName": "ecs-blog-svc",
		"clusterName": "ecs-blog-ECSCluster-1BC08Z8YA1A5X"
	}]
}

Create the CodeDeploy deployment group:

aws deploy create-deployment-group \
--cli-input-json file://code_deployment_group.json \
--region us-east-2

The output includes the deployment group ID, with the following format:

{
	"deploymentGroupId": "3c7d3ddf-0c8e-463e-8914-526496f0cad5"
}

We have now successfully setup linear deployment configurations for an Amazon ECS service running on AWS Fargate.  Time to exercise a linear deployment.  Let’s look at what we’ve created on the Amazon ECS Service and CodeDeploy Side

Navigate to the ECS cluster console.  From here click on the cluster, in this case ecs-blog-ECSCluster-11U2U784RFRG1.  Observe the service we’ve created via the CLI:

ECS cluster console

Image 4: ECS cluster console

Navigate to the CodeDeploy applications console.  Here we can observe the application and deployment group we have created.  Click on the application ecs-blog-app and then through to the deployment group ecs-blog-app-dg.  We can observe the deployment group configuration. Note the deployment configuration.

CodeDeploy deployment group console

Image 5: CodeDeploy deployment group console

Let’s look at the container that is running on Amazon ECS, by visiting the ExternalURL for the Application Load Balancer.  This can be copied from the outputs in Step 4.  If we navigate to that URL, we can observe the NGINX home page with a blue information section.

NGINX landing page

Image 6: NGINX landing page

Let’s exercise the deployment configuration we created.  In order to do this, we need to create a revision of the Task Definition being used, and instruct the ECS Service to use the new task set.  We will do this using the second container version (v2) URI, that we noted in Step 3.

Step 7: Update Task Definition revision using green container image

Navigate to the ECS Task Definitions console.  Here we will check the box next to the Task Definition ecs-blog-svc, then click Create new revision button.  This will bring us to the edit screen for creating a new revision.  We will navigate to the Container Definitions section of the screen and update the container definition by clicking on the Container Name:

ECS task definitions - container definition edit

Image 7: ECS task definitions – container definition edit

This will bring us to edit container page. Simply replace the Image with the secondary image (v2) URI from Step 3.

ECS task definitions - edit container

Image 8: ECS task definitions – edit container

Click Update and then Create.  We have created a new revision.

Successfully updated task definition container

Image 9: Successfully updated task definition container

Step 8: Update ECS service to use new Task Definition revision and trigger a CodeDeploy linear deployment

Navigate to the ECS cluster console.  From here click on the cluster, in this case ecs-blog-ECSCluster-1BC08Z8YA1A5X.  Click the check box next to the service, ecs-blog-svc, then click the Update button.

ECS cluster console - service

Image 10: ECS cluster console – service

In the edit screen for the service, select the last revision (2) under the Task Definition area.

Service update task definition revision

Image 11: Service update task definition revision

Click Next Step through all the following steps, accepting the default until the last wizard screen, click Update Service.

Successfully updated Service

Image 12: Successfully updated Service

Step 9: Observe linear deployment

Navigate to the CodeDeploy console.

Initially the deployment will create a replacement task set.  You can see the Traffic shifting process to the right, at 100% to the original task set.

As the shifting progresses, you can observe the percentage shift.

Navigate in a separate window to the External URL of the Application Load Balancer.  Observe the change, enable the refresh, the information box on the NGINX home will shift from blue to green.

After the shifting period is complete, the deployment will terminate the original task set.  This period of wait is configurable.  In our case we have set this to 5 minutes, the default is 60 minutes.

Conclusion

In this post, I showed you how to create a Fargate ECS service with blue/green linear deployments powered by CodeDeploy.  Take this example and attempt the other variants of linear/canary deployment configurations, customize one for your needs as well.

For information about how to automate this end-to-end using a continuous delivery pipeline in CodePipeline and ECR, check out this blog post on building a continuous delivery pipeline for your container images with Amazon ECR as source.

I hope that the information in this post helps you get started implementing this for your own applications!

Darren Ball

Darren Ball

Darren Ball is a Solutions Architect at AWS. His current focus is on helping Greenfied customers with their cloud migration journey, as well as building solutions at the crossroads of all things cloud and on premise with an emphasis on DevOps and containers.

Kia Fallahi

Kia Fallahi

Kiamarz Fallahi is a Senior Technical Product Manager on the Code Suite team at AWS. His primary focus is to provide delightful solutions to enhance the productivity of software developers using AWS developer tools.