使用 AWSSDK 了解 Amazon ECS 基本概念

下列程式碼範例示範如何學習 Amazon ECS 基本概念。

• 建立 ECS 叢集。

• 註冊 Fargate 任務定義。

• 執行獨立任務。

• 建立服務。

• 列出服務中的任務。

• 描述和更新服務。

• 清除所有資源。

Python

適用於 Python 的 SDK (Boto3)

注意

GitHub 上提供更多範例。尋找完整範例，並了解如何在 AWS 程式碼範例儲存庫中設定和執行。

在命令提示中執行互動式案例。

import json
import logging
import os
import sys

import boto3
from botocore.exceptions import ClientError, WaiterError

# The EcsWrapper lives in the parent directory (python/example_code/ecs/).
# Add it to the path so this scenario can be run directly from the scenarios/ folder.
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from ecs_wrapper import EcsWrapper

logger = logging.getLogger(__name__)

CLUSTER_NAME = "ecs-basics-cluster"
TASK_FAMILY = "ecs-basics-task-def"
SERVICE_NAME = "ecs-basics-service"
STACK_NAME = "ecs-basics-prerequisites"


def get_cfn_template():
    """Returns the CloudFormation template as a JSON string."""
    template = {
        "AWSTemplateFormatVersion": "2010-09-09",
        "Description": "ECS Basics - VPC, subnets, security group, and task execution role.",
        "Resources": {
            "EcsTaskExecutionRole": {
                "Type": "AWS::IAM::Role",
                "Properties": {
                    "AssumeRolePolicyDocument": {
                        "Version":"2012-10-17",
                        "Statement": [
                            {
                                "Effect": "Allow",
                                "Principal": {"Service": "ecs-tasks.amazonaws.com"},
                                "Action": "sts:AssumeRole",
                            }
                        ],
                    },
                    "ManagedPolicyArns": [
                        "arn:aws:iam::aws:policy/service-role/AmazonECSTaskExecutionRolePolicy"
                    ],
                },
            },
            "Vpc": {
                "Type": "AWS::EC2::VPC",
                "Properties": {
                    "CidrBlock": "10.0.0.0/16",
                    "EnableDnsSupport": True,
                    "EnableDnsHostnames": True,
                },
            },
            "InternetGateway": {"Type": "AWS::EC2::InternetGateway"},
            "AttachGateway": {
                "Type": "AWS::EC2::VPCGatewayAttachment",
                "Properties": {
                    "VpcId": {"Ref": "Vpc"},
                    "InternetGatewayId": {"Ref": "InternetGateway"},
                },
            },
            "PublicSubnetOne": {
                "Type": "AWS::EC2::Subnet",
                "Properties": {
                    "VpcId": {"Ref": "Vpc"},
                    "CidrBlock": "10.0.1.0/24",
                    "AvailabilityZone": {"Fn::Select": ["0", {"Fn::GetAZs": ""}]},
                    "MapPublicIpOnLaunch": True,
                },
            },
            "PublicSubnetTwo": {
                "Type": "AWS::EC2::Subnet",
                "Properties": {
                    "VpcId": {"Ref": "Vpc"},
                    "CidrBlock": "10.0.2.0/24",
                    "AvailabilityZone": {"Fn::Select": ["1", {"Fn::GetAZs": ""}]},
                    "MapPublicIpOnLaunch": True,
                },
            },
            "RouteTable": {
                "Type": "AWS::EC2::RouteTable",
                "Properties": {"VpcId": {"Ref": "Vpc"}},
            },
            "DefaultRoute": {
                "Type": "AWS::EC2::Route",
                "DependsOn": "AttachGateway",
                "Properties": {
                    "RouteTableId": {"Ref": "RouteTable"},
                    "DestinationCidrBlock": "0.0.0.0/0",
                    "GatewayId": {"Ref": "InternetGateway"},
                },
            },
            "SubnetOneRouteTableAssoc": {
                "Type": "AWS::EC2::SubnetRouteTableAssociation",
                "Properties": {
                    "SubnetId": {"Ref": "PublicSubnetOne"},
                    "RouteTableId": {"Ref": "RouteTable"},
                },
            },
            "SubnetTwoRouteTableAssoc": {
                "Type": "AWS::EC2::SubnetRouteTableAssociation",
                "Properties": {
                    "SubnetId": {"Ref": "PublicSubnetTwo"},
                    "RouteTableId": {"Ref": "RouteTable"},
                },
            },
            "SecurityGroup": {
                "Type": "AWS::EC2::SecurityGroup",
                "Properties": {
                    "GroupDescription": "ECS Basics security group",
                    "VpcId": {"Ref": "Vpc"},
                    "SecurityGroupEgress": [
                        {
                            "IpProtocol": "-1",
                            "CidrIp": "0.0.0.0/0",
                        }
                    ],
                },
            },
        },
        "Outputs": {
            "TaskExecutionRoleArn": {
                "Value": {"Fn::GetAtt": ["EcsTaskExecutionRole", "Arn"]},
            },
            "SubnetIdOne": {"Value": {"Ref": "PublicSubnetOne"}},
            "SubnetIdTwo": {"Value": {"Ref": "PublicSubnetTwo"}},
            "SecurityGroupId": {"Value": {"Ref": "SecurityGroup"}},
        },
    }
    return json.dumps(template)


# Stack states that cannot be reused or updated in place. A stack in one of
# these states is the residue of a failed create/rollback and must be deleted
# before a fresh stack can be created with the same name.
_UNUSABLE_STACK_STATES = frozenset(
    {"ROLLBACK_COMPLETE", "ROLLBACK_FAILED", "CREATE_FAILED", "DELETE_FAILED"}
)


def _delete_stack_if_unusable(cfn_client):
    """
    Deletes the stack if it exists in an unusable state from a prior failed run.

    CloudFormation will not let you create a stack whose name is already taken,
    and a stack stuck in ROLLBACK_COMPLETE (or similar) can neither be updated
    nor reused. Removing it first makes the scenario safely re-entrant.
    """
    try:
        response = cfn_client.describe_stacks(StackName=STACK_NAME)
    except ClientError as err:
        # No existing stack (or we can't read it): nothing to clean up.
        if "does not exist" in err.response["Error"]["Message"]:
            return
        raise

    status = response["Stacks"][0]["StackStatus"]
    if status in _UNUSABLE_STACK_STATES:
        print(
            f"Stack '{STACK_NAME}' is in an unusable state ({status}) from a "
            "previous run; deleting it before redeploying..."
        )
        delete_stack(cfn_client)


def deploy_stack(cfn_client):
    """Deploys the CloudFormation stack and returns outputs as a dict."""
    print(f"\nDeploying CloudFormation stack '{STACK_NAME}'...")
    # A stack left over from a previous failed run (e.g. ROLLBACK_COMPLETE or
    # CREATE_FAILED) cannot be reused or updated in place; it has no usable
    # outputs and must be deleted before we can create a fresh one. Clean it up
    # first so the scenario is re-entrant.
    _delete_stack_if_unusable(cfn_client)

    try:
        cfn_client.create_stack(
            StackName=STACK_NAME,
            TemplateBody=get_cfn_template(),
            Capabilities=["CAPABILITY_IAM"],
        )
        waiter = cfn_client.get_waiter("stack_create_complete")
        print("Waiting for stack creation to complete...")
        waiter.wait(
            StackName=STACK_NAME, WaiterConfig={"Delay": 15, "MaxAttempts": 60}
        )
    except ClientError as err:
        if err.response["Error"]["Code"] == "AlreadyExistsException":
            # The stack already exists in a usable (complete) state. Reuse its
            # outputs rather than failing, so repeated runs are graceful.
            print(
                f"Stack '{STACK_NAME}' already exists; reusing its outputs. "
                "Delete it manually if you need a clean deployment."
            )
        else:
            raise

    stack = cfn_client.describe_stacks(StackName=STACK_NAME)["Stacks"][0]

    # Only a stack in a complete state exposes usable outputs. If it ended up in
    # any other state (e.g. ROLLBACK_COMPLETE), "Outputs" is absent and blindly
    # reading stack_outputs["TaskExecutionRoleArn"] later would raise a
    # confusing KeyError. Fail fast here with a clear message instead.
    status = stack["StackStatus"]
    if status not in ("CREATE_COMPLETE", "UPDATE_COMPLETE"):
        raise RuntimeError(
            f"Stack '{STACK_NAME}' is in state '{status}', which has no usable "
            "outputs. Delete the stack and re-run the scenario."
        )

    # CloudFormation omits the "Outputs" key entirely when a stack has none, so
    # default to an empty list rather than indexing directly.
    outputs = stack.get("Outputs", list())
    result = dict()
    for output in outputs:
        result[output["OutputKey"]] = output["OutputValue"]
    print(f"Stack outputs: {result}")
    return result


def delete_stack(cfn_client):
    """Deletes the CloudFormation stack."""
    print(f"\nDeleting CloudFormation stack '{STACK_NAME}'...")
    try:
        cfn_client.delete_stack(StackName=STACK_NAME)
        waiter = cfn_client.get_waiter("stack_delete_complete")
        waiter.wait(StackName=STACK_NAME, WaiterConfig={"Delay": 15, "MaxAttempts": 60})
        print("Stack deleted.")
    except ClientError as err:
        logger.error("Failed to delete stack: %s", err)


def run_scenario():
    """Runs the ECS Basics scenario."""
    ecs_client = boto3.client("ecs")
    cfn_client = boto3.client("cloudformation")
    wrapper = EcsWrapper(ecs_client)

    cluster_name = None
    task_def_arn = None
    standalone_task_arn = None
    service_name = None

    try:
        # --- Setup: Deploy CloudFormation stack ---
        stack_outputs = deploy_stack(cfn_client)
        execution_role_arn = stack_outputs["TaskExecutionRoleArn"]
        subnets = [stack_outputs["SubnetIdOne"], stack_outputs["SubnetIdTwo"]]
        security_groups = [stack_outputs["SecurityGroupId"]]

        # --- Setup: Create ECS cluster ---
        print(f"\nCreating ECS cluster '{CLUSTER_NAME}'...")
        cluster = wrapper.create_cluster(CLUSTER_NAME)
        cluster_name = cluster["clusterName"]
        print(f"  Cluster: {cluster['clusterName']} (ARN: {cluster['clusterArn']})")
        print(f"  Status: {cluster['status']}")

        # --- Step 1: Register task definition ---
        print(f"\nRegistering task definition '{TASK_FAMILY}'...")
        task_def = wrapper.register_task_definition(
            family=TASK_FAMILY,
            execution_role_arn=execution_role_arn,
        )
        task_def_arn = f"{task_def['family']}:{task_def['revision']}"
        print(f"  Family: {task_def['family']}, Revision: {task_def['revision']}")
        print(f"  ARN: {task_def['taskDefinitionArn']}")

        # --- Step 2: Describe the cluster ---
        print(f"\nDescribing cluster '{cluster_name}'...")
        clusters = wrapper.describe_clusters([cluster_name])
        if clusters:
            c = clusters[0]
            print(f"  Name: {c['clusterName']}, Status: {c['status']}")
            print(f"  Active services: {c['activeServicesCount']}")
            print(f"  Running tasks: {c['runningTasksCount']}")

        # --- Step 3: Run a standalone task ---
        print(f"\nRunning standalone task with '{task_def_arn}'...")
        tasks = wrapper.run_task(
            cluster=cluster_name,
            task_definition=task_def_arn,
            subnets=subnets,
            security_groups=security_groups,
        )
        if tasks:
            standalone_task_arn = tasks[0]["taskArn"]
            print(f"  Task ARN: {standalone_task_arn}")
            print(f"  Status: {tasks[0]['lastStatus']}")
        else:
            # run_task can return an empty task list when every task fails to
            # start (see the 'failures' entry in the run_task response). Surface
            # this clearly so the reader knows Step 4 was skipped intentionally
            # rather than assuming the task launched successfully.
            logger.warning(
                "No tasks were started for '%s'. The task may have failed to "
                "launch (check the 'failures' field in the run_task response). "
                "Skipping the wait-and-describe step.",
                task_def_arn,
            )
            print("  No tasks were started; skipping the wait-and-describe step.")

        # --- Step 4: Wait for task to run, then describe it ---
        if standalone_task_arn:
            print("\nWaiting for task to reach RUNNING state...")
            try:
                waiter = ecs_client.get_waiter("tasks_running")
                waiter.wait(
                    cluster=cluster_name,
                    tasks=[standalone_task_arn],
                    WaiterConfig={"Delay": 6, "MaxAttempts": 100},
                )
            except WaiterError:
                print("  Task did not reach RUNNING state within timeout.")

            task_details = wrapper.describe_tasks(cluster_name, [standalone_task_arn])
            if task_details:
                t = task_details[0]
                print(f"  Task status: {t['lastStatus']}")
                for container in t.get("containers", list()):
                    print(f"  Container '{container['name']}': {container['lastStatus']}")

        # --- Step 5: Create a service ---
        print(f"\nCreating service '{SERVICE_NAME}'...")
        service = wrapper.create_service(
            cluster=cluster_name,
            service_name=SERVICE_NAME,
            task_definition=task_def_arn,
            desired_count=1,
            subnets=subnets,
            security_groups=security_groups,
        )
        service_name = service["serviceName"]
        print(f"  Service: {service['serviceName']} (ARN: {service['serviceArn']})")
        print(f"  Desired count: {service['desiredCount']}")

        # --- Step 6: List tasks in the service ---
        print("\nWaiting for the service to reach a steady state...")
        try:
            services_waiter = ecs_client.get_waiter("services_stable")
            services_waiter.wait(
                cluster=cluster_name,
                services=[service_name],
                WaiterConfig={"Delay": 15, "MaxAttempts": 40},
            )
        except WaiterError:
            print("  Service did not stabilize within the timeout; continuing.")
        task_arns = wrapper.list_tasks(
            cluster=cluster_name, service_name=service_name
        )
        print(f"  Service tasks: {task_arns}")

        # --- Step 7: Describe the service ---
        print(f"\nDescribing service '{service_name}'...")
        services = wrapper.describe_services(cluster_name, [service_name])
        if services:
            s = services[0]
            print(f"  Name: {s['serviceName']}, Status: {s['status']}")
            print(f"  Desired: {s['desiredCount']}, Running: {s['runningCount']}")
            events = s.get("events", list())
            if events:
                print(f"  Latest event: {events[0].get('message', 'N/A')}")

        # --- Step 8: Update the service (scale up) ---
        print(f"\nScaling service '{service_name}' to 2 tasks...")
        updated = wrapper.update_service(
            cluster=cluster_name,
            service=service_name,
            desired_count=2,
        )
        print(f"  Updated desired count: {updated['desiredCount']}")

    except Exception:
        logger.exception("Scenario encountered an error.")
        raise

    finally:
        # --- Cleanup ---
        print("\n--- Cleanup ---")

        # Stop standalone task
        if standalone_task_arn and cluster_name:
            try:
                print(f"Stopping standalone task '{standalone_task_arn}'...")
                ecs_client.stop_task(
                    cluster=cluster_name,
                    task=standalone_task_arn,
                    reason="Scenario cleanup",
                )
                stopped_waiter = ecs_client.get_waiter("tasks_stopped")
                stopped_waiter.wait(
                    cluster=cluster_name,
                    tasks=[standalone_task_arn],
                    WaiterConfig={"Delay": 6, "MaxAttempts": 100},
                )
                print("  Standalone task stopped.")
            except (ClientError, WaiterError) as err:
                logger.warning("Could not stop standalone task: %s", err)

        # Scale down and delete service
        if service_name and cluster_name:
            try:
                print(f"Scaling down service '{service_name}' to 0...")
                wrapper.update_service(
                    cluster=cluster_name,
                    service=service_name,
                    desired_count=0,
                )
                # Wait for tasks to drain before deleting the service.
                try:
                    drain_waiter = ecs_client.get_waiter("services_stable")
                    drain_waiter.wait(
                        cluster=cluster_name,
                        services=[service_name],
                        WaiterConfig={"Delay": 15, "MaxAttempts": 40},
                    )
                except WaiterError:
                    logger.warning(
                        "Service '%s' did not drain within the timeout; "
                        "attempting force delete.",
                        service_name,
                    )
                print(f"Deleting service '{service_name}'...")
                wrapper.delete_service(cluster=cluster_name, service=service_name)
                print("  Service deleted.")
            except ClientError as err:
                logger.warning("Could not delete service: %s", err)

        # Deregister task definition
        if task_def_arn:
            try:
                print(f"Deregistering task definition '{task_def_arn}'...")
                wrapper.deregister_task_definition(task_def_arn)
                print("  Task definition deregistered.")
            except ClientError as err:
                logger.warning("Could not deregister task definition: %s", err)

        # Delete cluster
        if cluster_name:
            try:
                print(f"Deleting cluster '{cluster_name}'...")
                wrapper.delete_cluster(cluster_name)
                print("  Cluster deleted.")
            except ClientError as err:
                logger.warning("Could not delete cluster: %s", err)

        # Delete CloudFormation stack
        delete_stack(cfn_client)

    print("\nECS Basics scenario complete!")


if __name__ == "__main__":
    logging.basicConfig(level=logging.INFO, format="%(levelname)s: %(message)s")
    run_scenario()

建立包裝 ECS 操作的類別。

class EcsWrapper:
    """Encapsulates Amazon ECS operations."""

    def __init__(self, ecs_client: BaseClient):
        """
        Initializes the EcsWrapper with an ECS client.

        :param ecs_client: A Boto3 Amazon ECS client. Boto3 clients are created
            by the ``boto3.client`` factory function and are instances of
            ``botocore.client.BaseClient``, which is the correct type to
            annotate here (``boto3.client`` itself is a function, not a type).
        """
        self.ecs_client = ecs_client

    @classmethod
    def from_client(cls) -> "EcsWrapper":
        """Creates an EcsWrapper using a default Boto3 ECS client."""
        ecs_client = boto3.client("ecs")
        return cls(ecs_client)

• 如需 API 詳細資訊，請參閱《適用於 Python (Boto3) 的 AWS SDK API 參考》中的下列主題。

  • CreateCluster

  • CreateService

  • DeleteCluster

  • DeleteService

  • DeregisterTaskDefinition

  • DescribeClusters

  • DescribeServices

  • DescribeTasks

  • ListTasks

  • RegisterTaskDefinition

  • RunTask

  • UpdateService