Skip to content

Latest commit

 

History

History
566 lines (359 loc) · 23.4 KB

README.rst

File metadata and controls

566 lines (359 loc) · 23.4 KB
Raw

ECS Deploy

ecs-deploy simplifies deployments on Amazon ECS by providing a convenience CLI tool for complex actions, which are executed pretty often.

Key Features

  • support for complex task definitions (e.g. multiple containers & task role)
  • easily redeploy the current task definition (including docker pull of eventually updated images)
  • deploy new versions/tags or all containers or just a single container in your task definition
  • scale up or down by adjusting the desired count of running tasks
  • add or adjust containers environment variables
  • run one-off tasks from the CLI
  • automatically monitor deployments in New Relic

TL;DR

Deploy a new version of your service:

$ ecs deploy my-cluster my-service --tag 1.2.3

Redeploy the current version of a service:

$ ecs deploy my-cluster my-service

Scale up or down a service:

$ ecs scale my-cluster my-service 4

Updating a cron job:

$ ecs cron my-cluster my-task my-rule

Update a task definition (without running or deploying):

$ ecs update my-task

Installation

The project is available on PyPI. Simply run:

$ pip install ecs-deploy

Run via Docker

Instead of installing ecs-deploy locally, which requires a Python environment, you can run ecs-deploy via Docker. All versions starting from 1.7.1 are available on Docker Hub: https://cloud.docker.com/repository/docker/fabfuel/ecs-deploy

Running ecs-deploy via Docker is easy as:

docker run fabfuel/ecs-deploy:1.10.2

In this example, the stable version 1.10.2 is executed. Alternatively you can use Docker tags master or latest for the latest stable version or Docker tag develop for the newest development version of ecs-deploy.

Please be aware, that when running ecs-deploy via Docker, the configuration - as described below - does not apply. You have to provide credentials and the AWS region via the command as attributes or environment variables:

docker run fabfuel/ecs-deploy:1.10.2 ecs deploy my-cluster my-service --region eu-central-1 --access-key-id ABC --secret-access-key ABC

Configuration

As ecs-deploy is based on boto3 (the official AWS Python library), there are several ways to configure and store the authentication credentials. Please read the boto3 documentation for more details (http://boto3.readthedocs.org/en/latest/guide/configuration.html#configuration). The simplest way is by running:

$ aws configure

Alternatively you can pass the AWS credentials (via --access-key-id and --secret-access-key) or the AWS configuration profile (via --profile) as options when you run ecs.

AWS IAM

If you are using ecs-deploy with a role or user account that does not have full AWS access, such as in a deploy script, you will need to use or create an IAM policy with the correct set of permissions in order for your deploys to succeed. One option is to use the pre-specified AmazonECS_FullAccess (https://docs.aws.amazon.com/AmazonECS/latest/userguide/security-iam-awsmanpol.html#security-iam-awsmanpol-AmazonECS_FullAccess) policy. If you would prefer to create a role with a more minimal set of permissions, the following are required:

  • ecs:ListServices
  • ecs:UpdateService
  • ecs:ListTasks
  • ecs:RegisterTaskDefinition
  • ecs:DescribeServices
  • ecs:DescribeTasks
  • ecs:ListTaskDefinitions
  • ecs:DescribeTaskDefinition
  • ecs:DeregisterTaskDefinition

If using custom IAM permissions, you will also need to set the iam:PassRole policy for each IAM role. See here https://docs.aws.amazon.com/IAM/latest/UserGuide/id_roles_use_passrole.html for more information.

Note that not every permission is required for every action you can take in ecs-deploy. You may be able to adjust permissions based on your specific needs.

Actions

Currently the following actions are supported:

deploy

Redeploy a service either without any modifications or with a new image, environment variable, docker label, and/or command definition.

scale

Scale a service up or down and change the number of running tasks.

run

Run a one-off task based on an existing task-definition and optionally override command, environment variables and/or docker labels.

update

Update a task definition by creating a new revision to set a new image, environment variable, docker label, and/or command definition, etc.

cron (scheduled task)

Update a task definition and update a events rule (scheduled task) to use the new task definition.

Usage

For detailed information about the available actions, arguments and options, run:

$ ecs deploy --help
$ ecs scale --help
$ ecs run --help

Examples

All examples assume, that authentication has already been configured.

Deployment

Simple Redeploy

To redeploy a service without any modifications, but pulling the most recent image versions, run the following command. This will duplicate the current task definition and cause the service to redeploy all running tasks.:

$ ecs deploy my-cluster my-service

Deploy a new tag

To change the tag for all images in all containers in the task definition, run the following command:

$ ecs deploy my-cluster my-service -t 1.2.3

Deploy a new image

To change the image of a specific container, run the following command:

$ ecs deploy my-cluster my-service --image webserver nginx:1.11.8

This will modify the webserver container only and change its image to "nginx:1.11.8".

Deploy several new images

The -i or --image option can also be passed several times:

$ ecs deploy my-cluster my-service -i webserver nginx:1.9 -i application my-app:1.2.3

This will change the webserver's container image to "nginx:1.9" and the application's image to "my-app:1.2.3".

Deploy a custom task definition

To deploy any task definition (independent of which is currently used in the service), you can use the --task parameter. The value can be:

A fully qualified task ARN:

$ ecs deploy my-cluster my-service --task arn:aws:ecs:eu-central-1:123456789012:task-definition/my-task:20

A task family name with revision:

$ ecs deploy my-cluster my-service --task my-task:20

Or just a task family name. It this case, the most recent revision is used:

$ ecs deploy my-cluster my-service --task my-task

Important

ecs will still create a new task definition, which then is used in the service. This is done, to retain consistent behaviour and to ensure the ECS agent e.g. pulls all images. But the newly created task definition will be based on the given task, not the currently used task.

Set an environment variable

To add a new or adjust an existing environment variable of a specific container, run the following command:

$ ecs deploy my-cluster my-service -e webserver SOME_VARIABLE SOME_VALUE

This will modify the webserver container definition and add or overwrite the environment variable SOME_VARIABLE with the value "SOME_VALUE". This way you can add new or adjust already existing environment variables.

Adjust multiple environment variables

You can add or change multiple environment variables at once, by adding the -e (or --env) options several times:

$ ecs deploy my-cluster my-service -e webserver SOME_VARIABLE SOME_VALUE -e webserver OTHER_VARIABLE OTHER_VALUE -e app APP_VARIABLE APP_VALUE

This will modify the definition of two containers. The webserver's environment variable SOME_VARIABLE will be set to "SOME_VALUE" and the variable OTHER_VARIABLE to "OTHER_VALUE". The app's environment variable APP_VARIABLE will be set to "APP_VALUE".

Set environment variables exclusively, remove all other pre-existing environment variables

To reset all existing environment variables of a task definition, use the flag --exclusive-env

$ ecs deploy my-cluster my-service -e webserver SOME_VARIABLE SOME_VALUE --exclusive-env

This will remove all other existing environment variables of all containers of the task definition, except for the variable SOME_VARIABLE with the value "SOME_VALUE" in the webserver container.

Set a secret environment variable from the AWS Parameter Store

Important

This option was introduced by AWS in ECS Agent v1.22.0. Make sure your ECS agent version is >= 1.22.0 or else your task will not deploy.

To add a new or adjust an existing secret of a specific container, run the following command:

$ ecs deploy my-cluster my-service -s webserver SOME_SECRET KEY_OF_SECRET_IN_PARAMETER_STORE

You can also specify the full arn of the parameter:

$ ecs deploy my-cluster my-service -s webserver SOME_SECRET arn:aws:ssm:<aws region>:<aws account id>:parameter/KEY_OF_SECRET_IN_PARAMETER_STORE

This will modify the webserver container definition and add or overwrite the environment variable SOME_SECRET with the value of the KEY_OF_SECRET_IN_PARAMETER_STORE in the AWS Parameter Store of the AWS Systems Manager.

Set secrets exclusively, remove all other pre-existing secret environment variables

To reset all existing secrets (secret environment variables) of a task definition, use the flag --exclusive-secrets

$ ecs deploy my-cluster my-service -s webserver NEW_SECRET KEY_OF_SECRET_IN_PARAMETER_STORE --exclusive-secret

This will remove all other existing secret environment variables of all containers of the task definition, except for the new secret variable NEW_SECRET with the value coming from the AWS Parameter Store with the name "KEY_OF_SECRET_IN_PARAMETER_STORE" in the webserver container.

Set environment via .env files

Instead of setting environment variables separately, you can pass a .env file per container to set the whole environment at once. You can either point to a local file or a file stored on S3, via:

$ ecs deploy my-cluster my-service --env-file my-app env/my-app.env

$ ecs deploy my-cluster my-service --s3-env-file my-app arn:aws:s3:::my-ecs-environment/my-app.env

Set secrets via .env files

Instead of setting secrets separately, you can pass a .env file per container to set all secrets at once.

This will expect an env file format, but any values will be set as the valueFrom parameter in the secrets config. This value can be either the path or the full ARN of a secret in the AWS Parameter Store. For example, with a secrets.env file like the following:

` SOME_SECRET=arn:aws:ssm:<aws region>:<aws account id>:parameter/KEY_OF_SECRET_IN_PARAMETER_STORE `

$ ecs deploy my-cluster my-service --secret-env-file webserver env/secrets.env

This will modify the webserver container definition and add or overwrite the environment variable SOME_SECRET with the value of the KEY_OF_SECRET_IN_PARAMETER_STORE in the AWS Parameter Store of the AWS Systems Manager.

Set a docker label

To add a new or adjust an existing docker labels of a specific container, run the following command:

$ ecs deploy my-cluster my-service -d webserver somelabel somevalue

This will modify the webserver container definition and add or overwrite the docker label "somelabel" with the value "somevalue". This way you can add new or adjust already existing docker labels.

Adjust multiple docker labels

You can add or change multiple docker labels at once, by adding the -d (or --docker-label) options several times:

$ ecs deploy my-cluster my-service -d webserver somelabel somevalue -d webserver otherlabel othervalue -d app applabel appvalue

This will modify the definition of two containers. The webserver's docker label "somelabel" will be set to "somevalue" and the label "otherlabel" to "othervalue". The app's docker label "applabel" will be set to "appvalue".

Set docker labels exclusively, remove all other pre-existing docker labels

To reset all existing docker labels of a task definition, use the flag --exclusive-docker-labels

$ ecs deploy my-cluster my-service -d webserver somelabel somevalue --exclusive-docker-labels

This will remove all other existing docker labels of all containers of the task definition, except for the label "somelabel" with the value "somevalue" in the webserver container.

Modify a command

To change the command of a specific container, run the following command:

$ ecs deploy my-cluster my-service --command webserver "nginx"

This will modify the webserver container and change its command to "nginx". If you have a command that requires arguments as well, then you can simply specify it like this as you would normally do:

$ ecs deploy my-cluster my-service --command webserver "ngnix -c /etc/ngnix/ngnix.conf"

This works fine as long as any of the arguments do not contain any spaces. In case arguments to the command itself contain spaces, then you can use the JSON format:

$ ecs deploy my-cluster my-service --command webserver '["sh", "-c", "while true; do echo Time files like an arrow $(date); sleep 1; done;"]'

More about this can be looked up in documentation. https://docs.aws.amazon.com/AmazonECS/latest/developerguide/task_definition_parameters.html#container_definitions

Set a task role

To change or set the role, the service's task should run as, use the following command:

$ ecs deploy my-cluster my-service -r arn:aws:iam::123456789012:role/MySpecialEcsTaskRole

This will set the task role to "MySpecialEcsTaskRole".

Set CPU and memory reservation

  • Set the cpu value for a task: --task-cpu 0.
  • Set the cpu value for a task container: --cpu <container_name> 0.
  • Set the memory value (hard limit) for a task: --task-memory 256.
  • Set the memory value (hard limit) for a task container: --memory <container_name> 256.
  • Set the memoryreservation value (soft limit) for a task definition: --memoryreservation <container_name> 256.

Set privileged or essential flags

  • Set the privileged value for a task definition: --privileged <container_name> True|False.
  • Set the essential value for a task definition: --essential <container_name> True|False.

Set logging configuration

Set the logConfiguration values for a task definition:

--log <container_name> awslogs awslogs-group <log_group_name>
--log <container_name> awslogs awslogs-region <region>
--log <container_name> awslogs awslogs-stream-prefix <stream_prefix>

Set port mapping

  • Set the port mappings values for a task definition: --port <container_name> <container_port> <host_port>.
    • Supports --exclusive-ports.
    • The protocol is fixed to tcp.

Set volumes & mount points

  • Set the volumes values for a task definition --volume <volume_name> /host/path.
    • <volume_name> can then be used with --mount.
  • Set the mount points values for a task definition: --mount <container_name> <volume_name> /container/path.
    • Supports --exclusive-mounts.
    • <volume_name> is the one set by --volume.
  • Set the ulimits values for a task definition: --ulimit <container_name> memlock 67108864 67108864.
    • Supports --exclusive-ulimits.
  • Set the systemControls values for a task definition: --system-control <container_name> net.core.somaxconn 511.
    • Supports --exclusive-system-controls.
  • Set the healthCheck values for a task definition: --health-check <container_name> <command> <interval> <timeout> <retries> <start_period>.

Set Health Checks

  • Example --health-check webserver "curl -f http://localhost/alive/" 30 5 3 0

Placeholder Container

  • Add placeholder containers: --add-container <container_name>.
  • To comply with the minimum requirements for a task definition, a placeholder container is set like this:
    • The container name is <container_name>.
    • The container image is PLACEHOLDER.
    • The container soft limit is 128.
  • The idea is to set sensible values with the deployment.

It is possible to add and define a new container with the same deployment:

--add-container redis --image redis redis:6 --port redis 6379 6379

Remove containers

  • Containers can be removed: --remove-container <container_name>.
    • Leaves the original containers, if all containers would be removed.

All but the container flags can be used with ecs deploy and ecs cron. The container flags are used with ecs deploy only.

Ignore capacity issues

If your cluster is undersized or the service's deployment options are not optimally set, the cluster might be incapable to run blue-green-deployments. In this case, you might see errors like these:

ERROR: (service my-service) was unable to place a task because no container instance met all of its requirements. The closest matching (container-instance 123456-1234-1234-1234-1234567890) is already using a port required by your task. For more information, see the Troubleshooting section of the Amazon ECS Developer Guide.

There might also be warnings about insufficient memory or CPU.

To ignore these warnings, you can run the deployment with the flag --ignore-warnings:

$ ecs deploy my-cluster my-service --ignore-warnings

In that case, the warning is printed, but the script continues and waits for a successful deployment until it times out.

Deployment timeout

The deploy and scale actions allow defining a timeout (in seconds) via the --timeout parameter. This instructs ecs-deploy to wait for ECS to finish the deployment for the given number of seconds.

To run a deployment without waiting for the successful or failed result at all, set --timeout to the value of -1.

Multi-Account Setup

If you manage different environments of your system in multiple differnt AWS accounts, you can now easily assume a deployment role in the target account in which your ECS cluster is running. You only need to provide --account with the AWS account id and --assume-role with the name of the role you want to assume in the target account. ecs-deploy automatically assumes this role and deploys inside your target account:

Example:

$ ecs deploy my-cluster my-service --account 1234567890 --assume-role ecsDeployRole

Scaling

Scale a service

To change the number of running tasks and scale a service up and down, run this command:

$ ecs scale my-cluster my-service 4

Running a Task

Run a one-off task

To run a one-off task, based on an existing task-definition, run this command:

$ ecs run my-cluster my-task

You can define just the task family (e.g. my-task) or you can run a specific revision of the task-definition (e.g. my-task:123). And optionally you can add or adjust environment variables like this:

$ ecs run my-cluster my-task:123 -e my-container MY_VARIABLE "my value"

Run a task with a custom command

You can override the command definition via option -c or --command followed by the container name and the command in a natural syntax, e.g. no conversion to comma-separation required:

$ ecs run my-cluster my-task -c my-container "python some-script.py param1 param2"

The JSON syntax explained above regarding modifying a command is also applicable here.

Run a task in a Fargate Cluster

If you want to run a one-off task in a Fargate cluster, additional configuration is required, to instruct AWS e.g. which subnets or security groups to use. The required parameters for this are:

  • launchtype
  • securitygroup
  • subnet
  • public-ip

Example:

$ ecs run my-fargate-cluster my-task --launchtype=FARGATE --securitygroup sg-01234567890123456 --subnet subnet-01234567890123456 --public-ip

You can pass multiple subnet as well as multiple securitygroup values. the public-ip flag determines, if the task receives a public IP address or not. Please see ecs run --help for more details.

Monitoring

With ECS deploy you can track your deployments automatically. Currently only New Relic is supported:

New Relic

To record a deployment in New Relic, you can provide the the API Key (Attention: this is a specific REST API Key, not the license key) and the application id in two ways:

Via cli options:

$ ecs deploy my-cluster my-service --newrelic-apikey ABCDEFGHIJKLMN --newrelic-appid 1234567890

Or implicitly via environment variables NEW_RELIC_API_KEY and NEW_RELIC_APP_ID

$ export NEW_RELIC_API_KEY=ABCDEFGHIJKLMN
$ export NEW_RELIC_APP_ID=1234567890
$ ecs deploy my-cluster my-service

Optionally you can provide additional information for the deployment:

  • --comment "New feature X" - comment to the deployment
  • --user john.doe - the name of the user who deployed with
  • --newrelic-revision 1.0.0 - explicitly set the revision to use for the deployment

Note: If neither --tag nor --newrelic-revision are provided, the deployment will not be recorded.

Troubleshooting

If the service configuration in ECS is not optimally set, you might be seeing timeout or other errors during the deployment.

Timeout

The timeout error means, that AWS ECS takes longer for the full deployment cycle then ecs-deploy is told to wait. The deployment itself still might finish successfully, if there are no other problems with the deployed containers.

You can increase the time (in seconds) to wait for finishing the deployment via the --timeout parameter. This time includes the full cycle of stopping all old containers and (re)starting all new containers. Different stacks require different timeout values, the default is 300 seconds.

The overall deployment time depends on different things:

  • the type of the application. For example node.js containers tend to take a long time to get stopped. But nginx containers tend to stop immediately, etc.
  • are old and new containers able to run in parallel (e.g. using dynamic ports)?
  • the deployment options and strategy (Maximum percent > 100)?
  • the desired count of running tasks, compared to
  • the number of ECS instances in the cluster

Alternative Implementation

There are some other libraries/tools available on GitHub, which also handle the deployment of containers in AWS ECS. If you prefer another language over Python, have a look at these projects:

Shell
ecs-deploy - https://github.com/silinternational/ecs-deploy
Ruby
broadside - https://github.com/lumoslabs/broadside