Monitor Amazon ECS containers with ECS Exec - Amazon Elastic Container Service
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Monitor Amazon ECS containers with ECS Exec

With Amazon ECS Exec, you can directly interact with containers without needing to first interact with the host container operating system, open inbound ports, or manage SSH keys. You can use ECS Exec to run commands in or get a shell to a container running on an Amazon EC2 instance or on Amazon Fargate. This makes it easier to collect diagnostic information and quickly troubleshoot errors. For example, in a development context, you can use ECS Exec to easily interact with various process in your containers and troubleshoot your applications. And in production scenarios, you can use it to gain break-glass access to your containers to debug issues.

You can run commands in a running Linux or Windows container using ECS Exec from the Amazon ECS API, Amazon Command Line Interface (Amazon CLI), Amazon SDKs, or the Amazon Copilot CLI. For details on using ECS Exec, as well as a video walkthrough, using the Amazon Copilot CLI, see the Copilot GitHub documentation.

You can also use ECS Exec to maintain stricter access control policies. By selectively turning on this feature, you can control who can run commands and on which tasks they can run those commands. With a log of each command and their output, you can use ECS Exec to view which tasks were run and you can use CloudTrail to audit who accessed a container.

Considerations

For this topic, you should be familiar with the following aspects involved with using ECS Exec:

  • ECS Exec is not currently supported using the Amazon Web Services Management Console.

  • ECS Exec might not work as expected when running on operating systems not supported by Systems Manager. For information about the supported operating systems, see Operating system types in the Amazon Systems Manager User Guide.

  • ECS Exec is supported for tasks that run on the following infrastructure:

    • Linux containers on Amazon EC2 on any Amazon ECS-optimized AMI, including Bottlerocket

    • Linux and Windows containers on external instances (Amazon ECS Anywhere)

    • Linux and Windows containers on Amazon Fargate

    • Windows containers on Amazon EC2 on the following Windows Amazon ECS-optimized AMIs (with the container agent version 1.56 or later):

      • Amazon ECS-optimized Windows Server 2022 Full AMI

      • Amazon ECS-optimized Windows Server 2022 Core AMI

      • Amazon ECS-optimized Windows Server 2019 Full AMI

      • Amazon ECS-optimized Windows Server 2019 Core AMI

      • Amazon ECS-optimized Windows Server 20H2 Core AMI

  • If you configured an HTTP proxy for your task, set the NO_PROXY environment variable to "NO_PROXY=169.254.169.254,169.254.170.2" in order to bypass the proxy for EC2 instance metadata and IAM role traffic. If you don't configure the NO_PROXY environment variable, there can be failures when retrieving instance metadata or IAM role credentials from the metadata endpoint within the container. Setting the NO_PROXY environment variable as recommended filters the metadata and IAM traffic so that requests to 169.254.169.254 and 169.254.170.2 do not go through the HTTP proxy.

  • ECS Exec and Amazon VPC

    • If you are using interface Amazon VPC endpoints with Amazon ECS, you must create the interface Amazon VPC endpoints for the Systems Manager Session Manager (ssmmessages). For more information about Systems Manager VPC endpoints, see Use Amazon PrivateLink to set up a VPC endpoint for Session Manager in the Amazon Systems Manager User Guide.

    • If you are using interface Amazon VPC endpoints with Amazon ECS, and you are using Amazon KMS key for encryption, then you must create the interface Amazon VPC endpoint for Amazon KMS key. For more information, see Connecting to Amazon KMS key through a VPC endpoint in the Amazon Key Management Service Developer Guide.

    • When you have tasks that run on Amazon EC2 instances, use awsvpc networking mode. If you don't have internet access, such as not configured to use a NAT gateway), you must create the interface Amazon VPC endpoints for the Systems Manager Session Manager (ssmmessages). For more information about awsvpc network mode considerations, see Considerations. For more information about Systems Manager VPC endpoints, see Use Amazon PrivateLink to set up a VPC endpoint for Session Manager in the Amazon Systems Manager User Guide.

  • ECS Exec and SSM

    • When a user runs commands on a container using ECS Exec, these commands are run as the root user. The SSM agent and its child processes run as root even when you specify a user ID for the container.

    • The SSM agent requires that the container file system can be written to in order to create the required directories and files. Therefore, making the root file system read-only using the readonlyRootFilesystem task definition parameter, or any other method, isn't supported.

    • While starting SSM sessions outside of the execute-command action is possible, this results in the sessions not being logged and being counted against the session limit. We recommend limiting this access by denying the ssm:start-session action using an IAM policy. For more information, see Limiting access to the Start Session action.

  • The following features run as a sidecar container. Therefore, you must specify the container name to run the command on.

    • Runtime Monitoring

    • Service Connect

  • Users can run all of the commands that are available within the container context. The following actions might result in orphaned and zombie processes: terminating the main process of the container, terminating the command agent, and deleting dependencies. To cleanup zombie processes, we recommend adding the initProcessEnabled flag to your task definition.

  • ECS Exec uses some CPU and memory. You'll want to accommodate for that when specifying the CPU and memory resource allocations in your task definition.

  • You must be using Amazon CLI version 1.22.3 or later or Amazon CLI version 2.3.6 or later. For information about how to update the Amazon CLI, see Installing or updating the latest version of the Amazon CLI in the Amazon Command Line Interface User Guide Version 2.

  • You can have only one ECS Exec session per process ID (PID) namespace. If you are sharing a PID namespace in a task, you can only start ECS Exec sessions into one container.

  • The ECS Exec session has an idle timeout time of 20 minutes. This value can't be changed.

  • You can't turn on ECS Exec for existing tasks. It can only be turned on for new tasks.

  • You can't use ECS Exec when you use run-task to launch a task on a cluster that uses managed scaling with asynchronous placement (launch a task with no instance).

  • You can't run ECS Exec against Microsoft Nano Server containers.

Prerequisites

Before you start using ECS Exec, make sure that you have completed these actions:

  • Install and configure the Amazon CLI. For more information, see Get started with the Amazon CLI.

  • Install Session Manager plugin for the Amazon CLI. For more information, see Install the Session Manager plugin for the Amazon CLI.

  • You must use a task role with the appropriate permissions for ECS Exec. For more information, see Task IAM role.

  • ECS Exec has version requirements depending on whether your tasks are hosted on Amazon EC2 or Amazon Fargate:

    • If you're using Amazon EC2, you must use an Amazon ECS optimized AMI that was released after January 20th, 2021, with an agent version of 1.50.2 or greater. For more information, see Amazon ECS optimized AMIs.

    • If you're using Amazon Fargate, you must use platform version 1.4.0 or higher (Linux) or 1.0.0 (Windows). For more information, see Amazon Fargate platform versions.

Architecture

ECS Exec makes use of Amazon Systems Manager (SSM) Session Manager to establish a connection with the running container and uses Amazon Identity and Access Management (IAM) policies to control access to running commands in a running container. This is made possible by bind-mounting the necessary SSM agent binaries into the container. The Amazon ECS or Amazon Fargate agent is responsible for starting the SSM core agent inside the container alongside your application code. For more information, see Systems Manager Session Manager.

You can audit which user accessed the container using the ExecuteCommand event in Amazon CloudTrail and log each command (and their output) to Amazon S3 or Amazon CloudWatch Logs. To encrypt data between the local client and container with your own encryption key, you must provide the Amazon Key Management Service (Amazon KMS) key.

Using ECS Exec

Optional task definition changes

If you set the task definition parameter initProcessEnabled to true, this starts the init process inside the container. This removes any zombie SSM agent child processes found. The following provides an example.

{ "taskRoleArn": "ecsTaskRole", "networkMode": "awsvpc", "requiresCompatibilities": [ "EC2", "FARGATE" ], "executionRoleArn": "ecsTaskExecutionRole", "memory": ".5 gb", "cpu": ".25 vcpu", "containerDefinitions": [ { "name": "amazon-linux", "image": "amazonlinux:latest", "essential": true, "command": ["sleep","3600"], "linuxParameters": { "initProcessEnabled": true } } ], "family": "ecs-exec-task" }

Turning on ECS Exec for your tasks and services

You can turn on the ECS Exec feature for your services and standalone tasks by specifying the --enable-execute-command flag when using one of the following Amazon CLI commands: create-service, update-service, start-task, or run-task.

For example, if you run the following command, the ECS Exec feature is turned on for a newly created service that runs on Fargate. For more information about creating services, see create-service.

aws ecs create-service \ --cluster cluster-name \ --task-definition task-definition-name \ --enable-execute-command \ --service-name service-name \ --launch-type FARGATE \ --network-configuration "awsvpcConfiguration={subnets=[subnet-12344321],securityGroups=[sg-12344321],assignPublicIp=ENABLED}" \ --desired-count 1

After you turn on ECS Exec for a task, you can run the following command to confirm the task is ready to be used. If the lastStatus property of the ExecuteCommandAgent is listed as RUNNING and the enableExecuteCommand property is set to true, then your task is ready.

aws ecs describe-tasks \ --cluster cluster-name \ --tasks task-id

The following output snippet is an example of what you might see.

{ "tasks": [ { ... "containers": [ { ... "managedAgents": [ { "lastStartedAt": "2021-03-01T14:49:44.574000-06:00", "name": "ExecuteCommandAgent", "lastStatus": "RUNNING" } ] } ], ... "enableExecuteCommand": true, ... } ] }

Running commands using ECS Exec

After you have confirmed the ExecuteCommandAgent is running, you can open an interactive shell on your container using the following command. If your task contains multiple containers, you must specify the container name using the --container flag. Amazon ECS only supports initiating interactive sessions, so you must use the --interactive flag.

The following command will run an interactive /bin/sh command against a container named container-name for a task with an ID of task-id.

The task-id is the Amazon Resource Name (ARN) of the task.

aws ecs execute-command --cluster cluster-name \ --task task-id \ --container container-name \ --interactive \ --command "/bin/sh"

Logging using ECS Exec

Turning on logging in your tasks and services

Important

For more information about CloudWatch pricing, see CloudWatch Pricing. Amazon ECS also provides monitoring metrics that are provided at no additional cost. For more information, see Monitor Amazon ECS using CloudWatch.

Amazon ECS provides a default configuration for logging commands run using ECS Exec by sending logs to CloudWatch Logs using the awslogs log driver that's configured in your task definition. If you want to provide a custom configuration, the Amazon CLI supports a --configuration flag for both the create-cluster and update-cluster commands. It’s also important to know that the container image requires script and cat to be installed in order to have command logs uploaded correctly to Amazon S3 or CloudWatch Logs. For more information about creating clusters, see create-cluster.

Note

This configuration only handles the logging of the execute-command session. It doesn't affect logging of your application.

The following example creates a cluster and then logs the output to your CloudWatch Logs LogGroup named cloudwatch-log-group-name and your Amazon S3 bucket named s3-bucket-name.

You must use an Amazon KMS customer managed key to encrypt the log group when you set the CloudWatchEncryptionEnabled option to true. For information about how to encrypt the log group, see Encrypt log data in CloudWatch Logs using Amazon Key Management Service, in the Amazon CloudWatch Logs User Guide.

aws ecs create-cluster \ --cluster-name cluster-name \ --configuration executeCommandConfiguration="{ \ kmsKeyId=string, \ logging=OVERRIDE, \ logConfiguration={ \ cloudWatchLogGroupName=cloudwatch-log-group-name, \ cloudWatchEncryptionEnabled=true, \ s3BucketName=s3-bucket-name, \ s3EncryptionEnabled=true, \ s3KeyPrefix=demo \ } \ }"

The logging property determines the behavior of the logging capability of ECS Exec:

  • NONE: logging is turned off.

  • DEFAULT: logs are sent to the configured awslogs driver. If the driver isn't configured, then no log is saved.

  • OVERRIDE: logs are sent to the provided Amazon CloudWatch Logs LogGroup, Amazon S3 bucket, or both.

IAM permissions required for Amazon CloudWatch Logs or Amazon S3 Logging

To enable logging, the Amazon ECS task role that's referenced in your task definition needs to have additional permissions. These additional permissions can be added as a policy to the task role. They're different depending on if you direct your logs to Amazon CloudWatch Logs or Amazon S3.

Amazon CloudWatch Logs

The following example policy adds the required Amazon CloudWatch Logs permissions.

{ "Version": "2012-10-17", "Statement": [ { "Effect": "Allow", "Action": [ "logs:DescribeLogGroups" ], "Resource": "*" }, { "Effect": "Allow", "Action": [ "logs:CreateLogStream", "logs:DescribeLogStreams", "logs:PutLogEvents" ], "Resource": "arn:aws:logs:region:account-id:log-group:/aws/ecs/cloudwatch-log-group-name:*" } ] }
Amazon S3

The following example policy adds the required Amazon S3 permissions.

{ "Version": "2012-10-17", "Statement": [ { "Effect": "Allow", "Action": [ "s3:GetBucketLocation" ], "Resource": "*" }, { "Effect": "Allow", "Action": [ "s3:GetEncryptionConfiguration" ], "Resource": "arn:aws:s3:::s3-bucket-name" }, { "Effect": "Allow", "Action": [ "s3:PutObject" ], "Resource": "arn:aws:s3:::s3-bucket-name/*" } ] }

IAM permissions required for encryption using your own Amazon KMS key (KMS key)

By default, the data transferred between your local client and the container uses TLS 1.2 encryption that Amazon provides. To further encrypt data using your own KMS key, you must create a KMS key and add the kms:Decrypt permission to your task IAM role. This permission is used by your container to decrypt the data. For more information about creating a KMS key, see Creating keys.

You add the following inline policy to your task IAM role which requires the Amazon KMS permissions. For more information, see ECS Exec permissions.

{ "Version": "2012-10-17", "Statement": [ { "Effect": "Allow", "Action": [ "kms:Decrypt" ], "Resource": "kms-key-arn" } ] }

For the data to be encrypted using your own KMS key, the user or group using the execute-command action must be granted the kms:GenerateDataKey permission.

The following example policy for your user or group contains the required permission to use your own KMS key. You must specify the Amazon Resource Name (ARN) of your KMS key.

{ "Version": "2012-10-17", "Statement": [ { "Effect": "Allow", "Action": [ "kms:GenerateDataKey" ], "Resource": "kms-key-arn" } ] }

Using IAM policies to limit access to ECS Exec

You limit user access to the execute-command API action by using one or more of the following IAM policy condition keys:

  • aws:ResourceTag/clusterTagKey

  • ecs:ResourceTag/clusterTagKey

  • aws:ResourceTag/taskTagKey

  • ecs:ResourceTag/taskTagKey

  • ecs:container-name

  • ecs:cluster

  • ecs:task

  • ecs:enable-execute-command

With the following example IAM policy, users can run commands in containers that are running within tasks with a tag that has an environment key and development value and in a cluster that's named cluster-name.

{ "Version": "2012-10-17", "Statement": [ { "Effect": "Allow", "Action": [ "ecs:ExecuteCommand", "ecs:DescribeTasks" ], "Resource": [ "arn:aws:ecs:region:aws-account-id:task/cluster-name/*", "arn:aws:ecs:region:aws-account-id:cluster/*" ], "Condition": { "StringEquals": { "ecs:ResourceTag/environment": "development" } } } ] }

With the following IAM policy example, users can't use the execute-command API when the container name is production-app.

{ "Version": "2012-10-17", "Statement": [ { "Effect": "Deny", "Action": [ "ecs:ExecuteCommand" ], "Resource": "*", "Condition": { "StringEquals": { "ecs:container-name": "production-app" } } } ] }

With the following IAM policy, users can only launch tasks when ECS Exec is turned off.

{ "Version": "2012-10-17", "Statement": [ { "Effect": "Allow", "Action": [ "ecs:RunTask", "ecs:StartTask", "ecs:CreateService", "ecs:UpdateService" ], "Resource": "*", "Condition": { "StringEquals": { "ecs:enable-execute-command": "false" } } } ] }
Note

Because the execute-command API action contains only task and cluster resources in a request, only cluster and task tags are evaluated.

For more information about IAM policy condition keys, see Actions, resources, and condition keys for Amazon Elastic Container Service in the Service Authorization Reference.

Limiting access to the Start Session action

While starting SSM sessions on your container outside of ECS Exec is possible, this could potentially result in the sessions not being logged. Sessions started outside of ECS Exec also count against the session quota. We recommend limiting this access by denying the ssm:start-session action directly for your Amazon ECS tasks using an IAM policy. You can deny access to all Amazon ECS tasks or to specific tasks based on the tags used.

The following is an example IAM policy that denies access to the ssm:start-session action for tasks in all Regions with a specified cluster name. You can optionally include a wildcard with the cluster-name.

{ "Version": "2012-10-17", "Statement": [ { "Effect": "Deny", "Action": "ssm:StartSession", "Resource": [ "arn:aws:ecs:region:aws-account-id:task/cluster-name/*", "arn:aws:ecs:region:aws-account-id:cluster/*" ] } ] }

The following is an example IAM policy that denies access to the ssm:start-session action on resources in all Regions tagged with tag key Task-Tag-Key and tag value Exec-Task.

{ "Version": "2012-10-17", "Statement": [ { "Effect": "Deny", "Action": "ssm:StartSession", "Resource": "arn:aws:ecs:*:*:task/*", "Condition": { "StringEquals": { "aws:ResourceTag/Task-Tag-Key": "Exec-Task" } } } ] }

For help with any issues you may run into when using Amazon ECS Exec, see Troubleshooting issues with Exec.