Deploying Docker containers on ECS

Estimated reading time: 19 minutes

Overview

The Docker Compose CLI enables developers to use native Docker commands to run applications in Amazon EC2 Container Service (ECS) when building cloud-native applications.

The integration between Docker and Amazon ECS allows developers to use the Docker Compose CLI to:

  • Set up an AWS context in one Docker command, allowing you to switch from a local context to a cloud context and run applications quickly and easily
  • Simplify multi-container application development on Amazon ECS using Compose files

Prerequisites

To deploy Docker containers on ECS, you must meet the following requirements:

  1. Download and install Docker Desktop Stable version 2.3.0.5 or later, or Edge version 2.3.2.0 or later.

    Alternatively, install the Docker Compose CLI for Linux.

  2. Ensure you have an AWS account.

Docker not only runs multi-container applications locally, but also enables developers to seamlessly deploy Docker containers on Amazon ECS using a Compose file with the docker compose up command. The following sections contain instructions on how to deploy your Compose application on Amazon ECS.

Run an application on ECS

Requirements

AWS uses a fine-grained permission model, with specific role for each resource type and operation.

To ensure that Docker ECS integration is allowed to manage resources for your Compose application, you

have to ensure your AWS credentials grant access to following AWS IAM permissions:

  • cloudformation:*
  • ecs:ListAccountSettings
  • ecs:CreateCluster
  • ecs:CreateService
  • ec2:DescribeVpcs
  • ec2:DescribeSubnets
  • ec2:CreateSecurityGroup
  • ec2:DescribeSecurityGroups
  • ec2:DeleteSecurityGroup
  • iam:CreateRole
  • iam:AttachRolePolicy
  • iam:DetachRolePolicy
  • iam:DeleteRole
  • elasticloadbalancing:*
  • application-autoscaling:*
  • servicediscovery:*
  • logs:CreateLogGroup
  • logs:DescribeLogGroups
  • logs:FilterLogEvents
  • logs:DeleteLogGroup
  • route53:CreateHostedZone
  • route53:DeleteHostedZone
  • route53:GetHealthCheck
  • route53:GetHostedZone
  • route53:ListHostedZonesByName

GPU support, which relies on EC2 instances to run containers with attached GPU devices, require a few additional permissions:

  • ec2:DescribeVpcs
  • autoscaling:*
  • iam:CreateInstanceProfile
  • iam:AddRoleToInstanceProfile
  • iam:RemoveRoleFromInstanceProfile
  • iam:DeleteInstanceProfile

Create AWS context

Run the docker context create ecs myecscontext command to create an Amazon ECS Docker context named myecscontext. If you have already installed and configured the AWS CLI, the setup command lets you select an existing AWS profile to connect to Amazon. Otherwise, you can create a new profile by passing an AWS access key ID and a secret access key. Finally, you can configure your ECS context to retrieve AWS credentials by AWS_* environment variables, which is a common way to integrate with third-party tools and single-sign-on providers.

? Create a Docker context using:  [Use arrows to move, type to filter]
  An existing AWS profile
  AWS secret and token credentials
> AWS environment variables

After you have created an AWS context, you can list your Docker contexts by running the docker context ls command:

NAME                TYPE                DESCRIPTION                               DOCKER ENDPOINT               KUBERNETES ENDPOINT   ORCHESTRATOR
myecscontext        ecs                 credentials read from environment                                                             
default *           moby                Current DOCKER_HOST based configuration   unix:///var/run/docker.sock                         swarm

Run a Compose application

You can deploy and manage multi-container applications defined in Compose files to Amazon ECS using the docker compose command. To do this:

  • Ensure you are using your ECS context. You can do this either by specifying the --context myecscontext flag with your command, or by setting the current context using the command docker context use myecscontext.

  • Run docker compose up and docker compose down to start and then stop a full Compose application.

    By default, docker compose up uses the compose.yaml or docker-compose.yaml file in the current folder. You can specify the Compose file directly using the --file flag.

    You can also specify a name for the Compose application using the --project-name flag during deployment. If no name is specified, a name will be derived from the working directory.

Docker ECS integration converts the Compose application model into a set of AWS resources, described as a CloudFormation template. The actual mapping is described in technical documentation. You can review the generated template using docker compose convert command, and follow CloudFormation applying this model within AWS web console when you run docker compose up, in addition to CloudFormation events being displayed in your terminal.

  • You can view services created for the Compose application on Amazon ECS and their state using the docker compose ps command.

  • You can view logs from containers that are part of the Compose application using the docker compose logs command.

Rolling update

To update your application without interrupting production flow you can simply use docker compose up on the updated Compose project. Your ECS services are created with rolling update configuration. As you run docker compose up with a modified Compose file, the stack will be updated to reflect changes, and if required, some services will be replaced. This replacement process will follow the rolling-update configuration set by your services deploy.update_config configuration.

AWS ECS uses a percent-based model to define the number of containers to be run or shut down during a rolling update. The Docker Compose CLI computes rolling update configuration according to the parallelism and replicas fields. However, you might prefer to directly configure a rolling update using the extension fields x-aws-min_percent and x-aws-max_percent. The former sets the minimum percent of containers to run for service, and the latter sets the maximum percent of additional containers to start before previous versions are removed.

By default, the ECS rolling update is set to run twice the number of containers for a service (200%), and has the ability to shut down 100% containers during the update.

View application logs

The Docker Compose CLI configures AWS CloudWatch Logs service for your containers. By default you can see logs of your compose application the same way you check logs of local deployments:

# fetch logs for application in current working directory
docker compose logs

# specify compose project name
docker compose logs --project-name PROJECT

# specify compose file
docker compose logs --file /path/to/docker-compose.yaml

A log group is created for the application as docker-compose/<application_name>, and log streams are created for each service and container in your application as <application_name>/<service_name>/<container_ID>.

You can fine tune AWS CloudWatch Logs using extension field x-aws-logs_retention in your Compose file to set the number of retention days for log events. The default behavior is to keep logs forever.

You can also pass awslogs driver parameters to your container as standard Compose file logging.driver_opts elements.

Private Docker images

The Docker Compose CLI automatically configures authorization so you can pull private images from the Amazon ECR registry on the same AWS account. To pull private images from another registry, including Docker Hub, you’ll have to create a Username + Password (or a Username + Token) secret on the AWS Secrets Manager service.

For your convenience, the Docker Compose CLI offers the docker secret command, so you can manage secrets created on AWS SMS without having to install the AWS CLI.

docker secret create dockerhubAccessToken --username <dockerhubuser>  --password <dockerhubtoken>
arn:aws:secretsmanager:eu-west-3:12345:secret:DockerHubAccessToken

Once created, you can use this ARN in you Compose file using using x-aws-pull_credentials custom extension with the Docker image URI for your service.

services:
  worker:
    image: mycompany/privateimage
    x-aws-pull_credentials: "arn:aws:secretsmanager:eu-west-3:12345:secret:DockerHubAccessToken"

Note

If you set the Compose file version to 3.8 or later, you can use the same Compose file for local deployment using docker-compose. Custom ECS extensions will be ignored in this case.

Service discovery

Service-to-service communication is implemented transparently by default, so you can deploy your Compose applications with multiple interconnected services without changing the compose file between local and ECS deployment. Individual services can run with distinct constraints (memory, cpu) and replication rules.

Service names

Services are registered automatically by the Docker Compose CLI on AWS Cloud Map during application deployment. They are declared as fully qualified domain names of the form: <service>.<compose_project_name>.local.

Services can retrieve their dependencies using Compose service names (as they do when deploying locally with docker-compose), or optionally use the fully qualified names.

Dependent service startup time and DNS resolution

Services get concurrently scheduled on ECS when a Compose file is deployed. AWS Cloud Map introduces an initial delay for DNS service to be able to resolve your services domain names. Your code needs to support this delay by waiting for dependent services to be ready, or by adding a wait-script as the entrypoint to your Docker image, as documented in Control startup order. Note this need to wait for dependent services in your Compose application also exists when deploying locally with docker-compose, but the delay is typically shorter. Issues might become more visible when deploying to ECS if services do not wait for their dependencies to be available.

Alternatively, you can use the depends_on feature of the Compose file format. By doing this, dependent service will be created first, and application deployment will wait for it to be up and running before starting the creation of the dependent services.

Service isolation

Service isolation is implemented by the Security Groups rules, allowing services sharing a common Compose file “network” to communicate together using their Compose service names.

Volumes

ECS integration supports volume management based on Amazon Elastic File System (Amazon EFS). For a Compose file to declare a volume, ECS integration will define creation of an EFS file system within the CloudFormation template, with Retain policy so data won’t be deleted on application shut-down. If the same application (same project name) is deployed again, the file system will be re-attached to offer the same user experience developers are used to with docker-compose.

A basic compose service using a volume can be declared like this:

services:
  nginx:
    image: nginx
    volumes:
      - mydata:/some/container/path
volumes:
  mydata:

With no specific volume options, the volume still must be declared in the volumessection for the compose file to be valid (in the above example the empty mydata: entry) If required, the initial file system can be customized using driver-opts:

volumes:
  my-data:
    driver_opts:
      # Filesystem configuration
      backup_policy: ENABLED
      lifecycle_policy: AFTER_14_DAYS
      performance_mode: maxIO
      throughput_mode: provisioned
      provisioned_throughput: 1024

File systems created by executing docker compose up on AWS can be listed using docker volume ls and removed with docker volume rm <filesystemID>.

An existing file system can also be used for users who already have data stored on EFS or want to use a file system created by another Compose stack.

volumes:
  my-data:
    external: true
    name: fs-123abcd

Accessing a volume from a container can introduce POSIX user ID permission issues, as Docker images can define arbitrary user ID / group ID for the process to run inside a container. However, the same uid:gid will have to match POSIX permissions on the file system. To work around the possible conflict, you can set the volume uid and gid to be used when accessing a volume:

volumes:
  my-data:
    driver_opts:
      # Access point configuration
      uid: 0
      gid: 0

Secrets

You can pass secrets to your ECS services using Docker model to bind sensitive data as files under /run/secrets. If your Compose file declares a secret as file, such a secret will be created as part of your application deployment on ECS. If you use an existing secret as external: true reference in your Compose file, use the ECS Secrets Manager full ARN as the secret name:

services:
  webapp:
    image: ...
    secrets:
      - foo

secrets:
  foo:
    name: "arn:aws:secretsmanager:eu-west-3:1234:secret:foo-ABC123"
    external: true

Secrets will be available at runtime for your service as a plain text file /run/secrets/foo.

The AWS Secrets Manager allows you to store sensitive data either as a plain text (like Docker secret does), or as a hierarchical JSON document. You can use the latter with Docker Compose CLI by using custom field x-aws-keys to define which entries in the JSON document to bind as a secret in your service container.

services:
  webapp:
    image: ...
    secrets:
      - foo

secrets:
  foo:
    name: "arn:aws:secretsmanager:eu-west-3:1234:secret:foo-ABC123"
    keys:
      - "bar"

By doing this, the secret for bar key will be available at runtime for your service as a plain text file /run/secrets/foo/bar. You can use the special value * to get all keys bound in your container.

Auto scaling

Scaling service static information (non auto-scaling) can be specified using the normal Compose syntax:

services:
  foo:
    deploy:
      replicas: 3

The Compose file model does not define any attributes to declare auto-scaling conditions. Therefore, we rely on x-aws-autoscaling custom extension to define the auto-scaling range, as well as cpu or memory to define target metric, expressed as resource usage percent.

services:
  foo:
    deploy:
      x-aws-autoscaling:
        min: 1
        max: 10 #required
        cpu: 75
        # mem: - mutualy exlusive with cpu

IAM roles

Your ECS Tasks are executed with a dedicated IAM role, granting access to AWS Managed policiesAmazonECSTaskExecutionRolePolicy and AmazonEC2ContainerRegistryReadOnly. In addition, if your service uses secrets, IAM Role gets additional permissions to read and decrypt secrets from the AWS Secret Manager.

You can grant additional managed policies to your service execution by using x-aws-policies inside a service definition:

services:
  foo:
    x-aws-policies:
      - "arn:aws:iam::aws:policy/AmazonS3FullAccess"

You can also write your own IAM Policy Document to fine tune the IAM role to be applied to your ECS service, and use x-aws-role inside a service definition to pass the yaml-formatted policy document.

services:
  foo:
    x-aws-role:
      Version: "2012-10-17"
      Statement:
        - Effect: "Allow"
          Action:
            - "some_aws_service"
          Resource:
            - "*"

Tuning the CloudFormation template

The Docker Compose CLI relies on Amazon CloudFormation to manage the application deployment. To get more control on the created resources, you can use docker compose convert to generate a CloudFormation stack file from your Compose file. This allows you to inspect resources it defines, or customize the template for your needs, and then apply the template to AWS using the AWS CLI, or the AWS web console.

Using existing AWS network resources

By default, the Docker Compose CLI creates an ECS cluster for your Compose application, a Security Group per network in your Compose file on your AWS account’s default VPC, and a LoadBalancer to route traffic to your services.

With the following basic compose file, the Docker Compose CLI will automatically create these ECS constructs including the load balancer to route traffic to the exposed port 80.

services:
  nginx:
    image: nginx
    ports:
      - "80:80"

If your AWS account does not have permissions to create such resources, or if you want to manage these yourself, you can use the following custom Compose extensions:

  • Use x-aws-cluster as a top-level element in your Compose file to set the ARN of an ECS cluster when deploying a Compose application. Otherwise, a cluster will be created for the Compose project.

  • Use x-aws-vpc as a top-level element in your Compose file to set the ARN of a VPC when deploying a Compose application.

  • Use x-aws-loadbalancer as a top-level element in your Compose file to set the ARN of an existing LoadBalancer.

The latter can be used for those who want to customize application exposure, typically to use an existing domain name for your application:

  1. Use the AWS web console or CLI to get your VPC and Subnets IDs. You can retrieve the default VPC ID and attached subnets using this AWS CLI commands:
$ aws ec2 describe-vpcs --filters Name=isDefault,Values=true --query 'Vpcs[0].VpcId' 

"vpc-123456"
$ aws ec2 describe-subnets --filters Name=vpc-id,Values=vpc-123456 --query 'Subnets[*].SubnetId'

[
    "subnet-1234abcd", 
    "subnet-6789ef00", 
]
  1. Use the AWS CLI to create your load balancer. The AWS Web Console can also be used but will require adding at least one listener, which we don’t need here.
$ aws elbv2 create-load-balancer --name myloadbalancer --type application --subnets "subnet-1234abcd" "subnet-6789ef00"

{
    "LoadBalancers": [
        {
            "IpAddressType": "ipv4", 
            "VpcId": "vpc-123456", 
            "LoadBalancerArn": "arn:aws:elasticloadbalancing:us-east-1:1234567890:loadbalancer/app/myloadbalancer/123abcd456", 
            "DNSName": "myloadbalancer-123456.us-east-1.elb.amazonaws.com", 
...            
  1. To assign your application an existing domain name, you can configure your DNS with a CNAME entry pointing to just-created loadbalancer’s DNSName reported as you created the loadbalancer.

  2. Use Loadbalancer ARN to set x-aws-loadbalancer in your compose file, and deploy your application using docker compose up command.

Please note Docker ECS integration won’t be aware of this domain name, so docker compose ps command will report URLs with loadbalancer DNSName, not your own domain.

You also can use external: true inside a network definition in your Compose file for Docker Compose CLI to not create a Security Group, and set name with the ID of an existing SecurityGroup you want to use for network connectivity between services:

networks:
  back_tier:
    external: true
    name: "sg-1234acbd"

Local simulation

When you deploy your application on ECS, you may also rely on the additional AWS services. In such cases, your code must embed the AWS SDK and retrieve API credentials at runtime. AWS offers a credentials discovery mechanism which is fully implemented by the SDK, and relies on accessing a metadata service on a fixed IP address.

Once you adopt this approach, running your application locally for testing or debug purposes can be difficult. Therefore, we have introduced an option on context creation to set the ecs-local context to maintain application portability between local workstation and the AWS cloud provider.

$ docker context create ecs --local-simulation ecsLocal
Successfully created ecs-local context "ecsLocal"

When you select a local simulation context, running the docker compose up command doesn’t deploy your application on ECS. Therefore, you must run it locally, automatically adjusting your Compose application so it includes the ECS local endpoints. This allows the AWS SDK used by application code to access a local mock container as “AWS metadata API” and retrieve credentials from your own local .aws/credentials config file.

Install the Docker Compose CLI on Linux

The Docker Compose CLI adds support for running and managing containers on ECS.

Install Prerequisites

Docker 19.03 or later

Install script

You can install the new CLI using the install script:

curl -L https://raw.githubusercontent.com/docker/compose-cli/main/scripts/install/install_linux.sh | sh

FAQ

What does the error this tool requires the "new ARN resource ID format" mean?

This error message means that your account requires the new ARN resource ID format for ECS. To learn more, see Migrating your Amazon ECS deployment to the new ARN and resource ID format.

Feedback

Thank you for trying out the Docker Compose CLI. Your feedback is very important to us. Let us know your feedback by creating an issue in the Compose CLI GitHub repository.

Docker, AWS, ECS, Integration, context, Compose, cli, deploy, containers, cloud