Tasks for AWS Step Functions

AWS Step Functions is a web service that enables you to coordinate the components of distributed applications and microservices using visual workflows. You build applications from individual components that each perform a discrete function, or task, allowing you to scale and change applications quickly.

A Task state represents a single unit of work performed by a state machine. All work in your state machine is performed by tasks. This module contains a collection of classes that allow you to call various AWS services from your Step Functions state machine.

Be sure to familiarize yourself with the aws-stepfunctions module documentation first.

This module is part of the AWS Cloud Development Kit project.

Table Of Contents

Tasks for AWS Step Functions

Paths

Learn more about input and output processing in Step Functions here

Evaluate Expression

Use the EvaluateExpression to perform simple operations referencing state paths. The expression referenced in the task will be evaluated in a Lambda function (eval()). This allows you to not have to write Lambda code for simple operations.

Example: convert a wait time from milliseconds to seconds, concat this in a message and wait:

convert_to_seconds = tasks.EvaluateExpression(self, "Convert to seconds",
    expression="$.waitMilliseconds / 1000",
    result_path="$.waitSeconds"
)

create_message = tasks.EvaluateExpression(self, "Create message",
    # Note: this is a string inside a string.
    expression="`Now waiting ${$.waitSeconds} seconds...`",
    runtime=lambda_.Runtime.NODEJS_LATEST,
    result_path="$.message"
)

publish_message = tasks.SnsPublish(self, "Publish message",
    topic=sns.Topic(self, "cool-topic"),
    message=sfn.TaskInput.from_json_path_at("$.message"),
    result_path="$.sns"
)

wait = sfn.Wait(self, "Wait",
    time=sfn.WaitTime.seconds_path("$.waitSeconds")
)

sfn.StateMachine(self, "StateMachine",
    definition=convert_to_seconds.next(create_message).next(publish_message).next(wait)
)

The EvaluateExpression supports a runtime prop to specify the Lambda runtime to use to evaluate the expression. Currently, only runtimes of the Node.js family are supported.

API Gateway

Step Functions supports API Gateway through the service integration pattern.

HTTP APIs are designed for low-latency, cost-effective integrations with AWS services, including AWS Lambda, and HTTP endpoints. HTTP APIs support OIDC and OAuth 2.0 authorization, and come with built-in support for CORS and automatic deployments. Previous-generation REST APIs currently offer more features. More details can be found here.

Call REST API Endpoint

The CallApiGatewayRestApiEndpoint calls the REST API endpoint.

import aws_cdk.aws_apigateway as apigateway

rest_api = apigateway.RestApi(self, "MyRestApi")

invoke_task = tasks.CallApiGatewayRestApiEndpoint(self, "Call REST API",
    api=rest_api,
    stage_name="prod",
    method=tasks.HttpMethod.GET
)

Be aware that the header values must be arrays. When passing the Task Token in the headers field WAIT_FOR_TASK_TOKEN integration, use JsonPath.array() to wrap the token in an array:

import aws_cdk.aws_apigateway as apigateway
# api: apigateway.RestApi


tasks.CallApiGatewayRestApiEndpoint(self, "Endpoint",
    api=api,
    stage_name="Stage",
    method=tasks.HttpMethod.PUT,
    integration_pattern=sfn.IntegrationPattern.WAIT_FOR_TASK_TOKEN,
    headers=sfn.TaskInput.from_object({
        "TaskToken": sfn.JsonPath.array(sfn.JsonPath.task_token)
    })
)

Call HTTP API Endpoint

The CallApiGatewayHttpApiEndpoint calls the HTTP API endpoint.

import aws_cdk.aws_apigatewayv2 as apigatewayv2

http_api = apigatewayv2.HttpApi(self, "MyHttpApi")

invoke_task = tasks.CallApiGatewayHttpApiEndpoint(self, "Call HTTP API",
    api_id=http_api.api_id,
    api_stack=Stack.of(http_api),
    method=tasks.HttpMethod.GET
)

AWS SDK

Step Functions supports calling AWS service’s API actions through the service integration pattern.

You can use Step Functions’ AWS SDK integrations to call any of the over two hundred AWS services directly from your state machine, giving you access to over nine thousand API actions.

# my_bucket: s3.Bucket

get_object = tasks.CallAwsService(self, "GetObject",
    service="s3",
    action="getObject",
    parameters={
        "Bucket": my_bucket.bucket_name,
        "Key": sfn.JsonPath.string_at("$.key")
    },
    iam_resources=[my_bucket.arn_for_objects("*")]
)

Use camelCase for actions and PascalCase for parameter names.

The task automatically adds an IAM statement to the state machine role’s policy based on the service and action called. The resources for this statement must be specified in iamResources.

Use the iamAction prop to manually specify the IAM action name in the case where the IAM action name does not match with the API service/action name:

list_buckets = tasks.CallAwsService(self, "ListBuckets",
    service="s3",
    action="listBuckets",
    iam_resources=["*"],
    iam_action="s3:ListAllMyBuckets"
)

Use the additionalIamStatements prop to pass additional IAM statements that will be added to the state machine role’s policy. Use it in the case where the call requires more than a single statement to be executed:

detect_labels = tasks.CallAwsService(self, "DetectLabels",
    service="rekognition",
    action="detectLabels",
    iam_resources=["*"],
    additional_iam_statements=[
        iam.PolicyStatement(
            actions=["s3:getObject"],
            resources=["arn:aws:s3:::my-bucket/*"]
        )
    ]
)

Athena

Step Functions supports Athena through the service integration pattern.

StartQueryExecution

The StartQueryExecution API runs the SQL query statement.

start_query_execution_job = tasks.AthenaStartQueryExecution(self, "Start Athena Query",
    query_string=sfn.JsonPath.string_at("$.queryString"),
    query_execution_context=tasks.QueryExecutionContext(
        database_name="mydatabase"
    ),
    result_configuration=tasks.ResultConfiguration(
        encryption_configuration=tasks.EncryptionConfiguration(
            encryption_option=tasks.EncryptionOption.S3_MANAGED
        ),
        output_location=s3.Location(
            bucket_name="query-results-bucket",
            object_key="folder"
        )
    ),
    execution_parameters=["param1", "param2"]
)

GetQueryExecution

The GetQueryExecution API gets information about a single execution of a query.

get_query_execution_job = tasks.AthenaGetQueryExecution(self, "Get Query Execution",
    query_execution_id=sfn.JsonPath.string_at("$.QueryExecutionId")
)

GetQueryResults

The GetQueryResults API that streams the results of a single query execution specified by QueryExecutionId from S3.

get_query_results_job = tasks.AthenaGetQueryResults(self, "Get Query Results",
    query_execution_id=sfn.JsonPath.string_at("$.QueryExecutionId")
)

StopQueryExecution

The StopQueryExecution API that stops a query execution.

stop_query_execution_job = tasks.AthenaStopQueryExecution(self, "Stop Query Execution",
    query_execution_id=sfn.JsonPath.string_at("$.QueryExecutionId")
)

Batch

Step Functions supports Batch through the service integration pattern.

SubmitJob

The SubmitJob API submits an AWS Batch job from a job definition.

import aws_cdk.aws_batch as batch
# batch_job_definition: batch.EcsJobDefinition
# batch_queue: batch.JobQueue


task = tasks.BatchSubmitJob(self, "Submit Job",
    job_definition_arn=batch_job_definition.job_definition_arn,
    job_name="MyJob",
    job_queue_arn=batch_queue.job_queue_arn
)

Bedrock

Step Functions supports Bedrock through the service integration pattern.

InvokeModel

The InvokeModel API invokes the specified Bedrock model to run inference using the input provided. The format of the input body and the response body depend on the model selected.

import aws_cdk.aws_bedrock as bedrock


model = bedrock.FoundationModel.from_foundation_model_id(self, "Model", bedrock.FoundationModelIdentifier.AMAZON_TITAN_TEXT_G1_EXPRESS_V1)

task = tasks.BedrockInvokeModel(self, "Prompt Model",
    model=model,
    body=sfn.TaskInput.from_object({
        "input_text": "Generate a list of five first names.",
        "text_generation_config": {
            "max_token_count": 100,
            "temperature": 1
        }
    }),
    result_selector={
        "names": sfn.JsonPath.string_at("$.Body.results[0].outputText")
    }
)

CodeBuild

Step Functions supports CodeBuild through the service integration pattern.

StartBuild

StartBuild starts a CodeBuild Project by Project Name.

import aws_cdk.aws_codebuild as codebuild


codebuild_project = codebuild.Project(self, "Project",
    project_name="MyTestProject",
    build_spec=codebuild.BuildSpec.from_object({
        "version": "0.2",
        "phases": {
            "build": {
                "commands": ["echo "Hello, CodeBuild!""
                ]
            }
        }
    })
)

task = tasks.CodeBuildStartBuild(self, "Task",
    project=codebuild_project,
    integration_pattern=sfn.IntegrationPattern.RUN_JOB,
    environment_variables_override={
        "ZONE": codebuild.BuildEnvironmentVariable(
            type=codebuild.BuildEnvironmentVariableType.PLAINTEXT,
            value=sfn.JsonPath.string_at("$.envVariables.zone")
        )
    }
)

StartBuildBatch

StartBuildBatch starts a batch CodeBuild for a project by project name. It is necessary to enable the batch build feature in the CodeBuild project.

import aws_cdk.aws_codebuild as codebuild


project = codebuild.Project(self, "Project",
    project_name="MyTestProject",
    build_spec=codebuild.BuildSpec.from_object_to_yaml({
        "version": 0.2,
        "batch": {
            "build-list": [{
                "identifier": "id",
                "buildspec": "version: 0.2
phases:
  build:
    commands:
      - echo "Hello, from small!""
            }
            ]
        }
    })
)
project.enable_batch_builds()

task = tasks.CodeBuildStartBuildBatch(self, "buildBatchTask",
    project=project,
    integration_pattern=sfn.IntegrationPattern.REQUEST_RESPONSE,
    environment_variables_override={
        "test": codebuild.BuildEnvironmentVariable(
            type=codebuild.BuildEnvironmentVariableType.PLAINTEXT,
            value="testValue"
        )
    }
)

Note: enableBatchBuilds() will do nothing for imported projects. If you are using an imported project, you must ensure that the project is already configured for batch builds.

DynamoDB

You can call DynamoDB APIs from a Task state. Read more about calling DynamoDB APIs here

GetItem

The GetItem operation returns a set of attributes for the item with the given primary key.

# my_table: dynamodb.Table

tasks.DynamoGetItem(self, "Get Item",
    key={"message_id": tasks.DynamoAttributeValue.from_string("message-007")},
    table=my_table
)

PutItem

The PutItem operation creates a new item, or replaces an old item with a new item.

# my_table: dynamodb.Table

tasks.DynamoPutItem(self, "PutItem",
    item={
        "MessageId": tasks.DynamoAttributeValue.from_string("message-007"),
        "Text": tasks.DynamoAttributeValue.from_string(sfn.JsonPath.string_at("$.bar")),
        "TotalCount": tasks.DynamoAttributeValue.from_number(10)
    },
    table=my_table
)

DeleteItem

The DeleteItem operation deletes a single item in a table by primary key.

# my_table: dynamodb.Table

tasks.DynamoDeleteItem(self, "DeleteItem",
    key={"MessageId": tasks.DynamoAttributeValue.from_string("message-007")},
    table=my_table,
    result_path=sfn.JsonPath.DISCARD
)

UpdateItem

The UpdateItem operation edits an existing item’s attributes, or adds a new item to the table if it does not already exist.

# my_table: dynamodb.Table

tasks.DynamoUpdateItem(self, "UpdateItem",
    key={
        "MessageId": tasks.DynamoAttributeValue.from_string("message-007")
    },
    table=my_table,
    expression_attribute_values={
        ":val": tasks.DynamoAttributeValue.number_from_string(sfn.JsonPath.string_at("$.Item.TotalCount.N")),
        ":rand": tasks.DynamoAttributeValue.from_number(20)
    },
    update_expression="SET TotalCount = :val + :rand"
)

ECS

Step Functions supports ECS/Fargate through the service integration pattern.

RunTask

RunTask starts a new task using the specified task definition.

EC2

The EC2 launch type allows you to run your containerized applications on a cluster of Amazon EC2 instances that you manage.

When a task that uses the EC2 launch type is launched, Amazon ECS must determine where to place the task based on the requirements specified in the task definition, such as CPU and memory. Similarly, when you scale down the task count, Amazon ECS must determine which tasks to terminate. You can apply task placement strategies and constraints to customize how Amazon ECS places and terminates tasks. Learn more about task placement

The latest ACTIVE revision of the passed task definition is used for running the task.

The following example runs a job from a task definition on EC2

vpc = ec2.Vpc.from_lookup(self, "Vpc",
    is_default=True
)

cluster = ecs.Cluster(self, "Ec2Cluster", vpc=vpc)
cluster.add_capacity("DefaultAutoScalingGroup",
    instance_type=ec2.InstanceType("t2.micro"),
    vpc_subnets=ec2.SubnetSelection(subnet_type=ec2.SubnetType.PUBLIC)
)

task_definition = ecs.TaskDefinition(self, "TD",
    compatibility=ecs.Compatibility.EC2
)

task_definition.add_container("TheContainer",
    image=ecs.ContainerImage.from_registry("foo/bar"),
    memory_limit_mi_b=256
)

run_task = tasks.EcsRunTask(self, "Run",
    integration_pattern=sfn.IntegrationPattern.RUN_JOB,
    cluster=cluster,
    task_definition=task_definition,
    launch_target=tasks.EcsEc2LaunchTarget(
        placement_strategies=[
            ecs.PlacementStrategy.spread_across_instances(),
            ecs.PlacementStrategy.packed_by_cpu(),
            ecs.PlacementStrategy.randomly()
        ],
        placement_constraints=[
            ecs.PlacementConstraint.member_of("blieptuut")
        ]
    ),
    propagated_tag_source=ecs.PropagatedTagSource.TASK_DEFINITION
)

Fargate

AWS Fargate is a serverless compute engine for containers that works with Amazon Elastic Container Service (ECS). Fargate makes it easy for you to focus on building your applications. Fargate removes the need to provision and manage servers, lets you specify and pay for resources per application, and improves security through application isolation by design. Learn more about Fargate

The Fargate launch type allows you to run your containerized applications without the need to provision and manage the backend infrastructure. Just register your task definition and Fargate launches the container for you. The latest ACTIVE revision of the passed task definition is used for running the task. Learn more about Fargate Versioning

The following example runs a job from a task definition on Fargate

vpc = ec2.Vpc.from_lookup(self, "Vpc",
    is_default=True
)

cluster = ecs.Cluster(self, "FargateCluster", vpc=vpc)

task_definition = ecs.TaskDefinition(self, "TD",
    memory_mi_b="512",
    cpu="256",
    compatibility=ecs.Compatibility.FARGATE
)

container_definition = task_definition.add_container("TheContainer",
    image=ecs.ContainerImage.from_registry("foo/bar"),
    memory_limit_mi_b=256
)

run_task = tasks.EcsRunTask(self, "RunFargate",
    integration_pattern=sfn.IntegrationPattern.RUN_JOB,
    cluster=cluster,
    task_definition=task_definition,
    assign_public_ip=True,
    container_overrides=[tasks.ContainerOverride(
        container_definition=container_definition,
        environment=[tasks.TaskEnvironmentVariable(name="SOME_KEY", value=sfn.JsonPath.string_at("$.SomeKey"))]
    )],
    launch_target=tasks.EcsFargateLaunchTarget(),
    propagated_tag_source=ecs.PropagatedTagSource.TASK_DEFINITION
)

ECS enable Exec

By setting the property enableExecuteCommand to true, you can enable the ECS Exec feature for the task for either Fargate or EC2 launch types.

vpc = ec2.Vpc.from_lookup(self, "Vpc",
    is_default=True
)
cluster = ecs.Cluster(self, "ECSCluster", vpc=vpc)

task_definition = ecs.TaskDefinition(self, "TD",
    compatibility=ecs.Compatibility.EC2
)

task_definition.add_container("TheContainer",
    image=ecs.ContainerImage.from_registry("foo/bar"),
    memory_limit_mi_b=256
)

run_task = tasks.EcsRunTask(self, "Run",
    integration_pattern=sfn.IntegrationPattern.RUN_JOB,
    cluster=cluster,
    task_definition=task_definition,
    launch_target=tasks.EcsEc2LaunchTarget(),
    enable_execute_command=True
)

EMR

Step Functions supports Amazon EMR through the service integration pattern. The service integration APIs correspond to Amazon EMR APIs but differ in the parameters that are used.

Read more about the differences when using these service integrations.

Create Cluster

Creates and starts running a cluster (job flow). Corresponds to the runJobFlow API in EMR.

cluster_role = iam.Role(self, "ClusterRole",
    assumed_by=iam.ServicePrincipal("ec2.amazonaws.com")
)

service_role = iam.Role(self, "ServiceRole",
    assumed_by=iam.ServicePrincipal("elasticmapreduce.amazonaws.com")
)

auto_scaling_role = iam.Role(self, "AutoScalingRole",
    assumed_by=iam.ServicePrincipal("elasticmapreduce.amazonaws.com")
)

auto_scaling_role.assume_role_policy.add_statements(
    iam.PolicyStatement(
        effect=iam.Effect.ALLOW,
        principals=[
            iam.ServicePrincipal("application-autoscaling.amazonaws.com")
        ],
        actions=["sts:AssumeRole"
        ]
    ))

tasks.EmrCreateCluster(self, "Create Cluster",
    instances=tasks.EmrCreateCluster.InstancesConfigProperty(),
    cluster_role=cluster_role,
    name=sfn.TaskInput.from_json_path_at("$.ClusterName").value,
    service_role=service_role,
    auto_scaling_role=auto_scaling_role
)

You can use the launch specification for On-Demand and Spot instances in the fleet.

tasks.EmrCreateCluster(self, "OnDemandSpecification",
    instances=tasks.EmrCreateCluster.InstancesConfigProperty(
        instance_fleets=[tasks.EmrCreateCluster.InstanceFleetConfigProperty(
            instance_fleet_type=tasks.EmrCreateCluster.InstanceRoleType.MASTER,
            launch_specifications=tasks.EmrCreateCluster.InstanceFleetProvisioningSpecificationsProperty(
                on_demand_specification=tasks.EmrCreateCluster.OnDemandProvisioningSpecificationProperty(
                    allocation_strategy=tasks.EmrCreateCluster.OnDemandAllocationStrategy.LOWEST_PRICE
                )
            )
        )]
    ),
    name="OnDemandCluster",
    integration_pattern=sfn.IntegrationPattern.RUN_JOB
)

tasks.EmrCreateCluster(self, "SpotSpecification",
    instances=tasks.EmrCreateCluster.InstancesConfigProperty(
        instance_fleets=[tasks.EmrCreateCluster.InstanceFleetConfigProperty(
            instance_fleet_type=tasks.EmrCreateCluster.InstanceRoleType.MASTER,
            launch_specifications=tasks.EmrCreateCluster.InstanceFleetProvisioningSpecificationsProperty(
                spot_specification=tasks.EmrCreateCluster.SpotProvisioningSpecificationProperty(
                    allocation_strategy=tasks.EmrCreateCluster.SpotAllocationStrategy.CAPACITY_OPTIMIZED,
                    timeout_action=tasks.EmrCreateCluster.SpotTimeoutAction.TERMINATE_CLUSTER,
                    timeout=Duration.minutes(5)
                )
            )
        )]
    ),
    name="SpotCluster",
    integration_pattern=sfn.IntegrationPattern.RUN_JOB
)

If you want to run multiple steps in parallel, you can specify the stepConcurrencyLevel property. The concurrency range is between 1 and 256 inclusive, where the default concurrency of 1 means no step concurrency is allowed. stepConcurrencyLevel requires the EMR release label to be 5.28.0 or above.

tasks.EmrCreateCluster(self, "Create Cluster",
    instances=tasks.EmrCreateCluster.InstancesConfigProperty(),
    name=sfn.TaskInput.from_json_path_at("$.ClusterName").value,
    step_concurrency_level=10
)

Termination Protection

Locks a cluster (job flow) so the EC2 instances in the cluster cannot be terminated by user intervention, an API call, or a job-flow error.

Corresponds to the setTerminationProtection API in EMR.

tasks.EmrSetClusterTerminationProtection(self, "Task",
    cluster_id="ClusterId",
    termination_protected=False
)

Terminate Cluster

Shuts down a cluster (job flow). Corresponds to the terminateJobFlows API in EMR.

tasks.EmrTerminateCluster(self, "Task",
    cluster_id="ClusterId"
)

Add Step

Adds a new step to a running cluster. Corresponds to the addJobFlowSteps API in EMR.

tasks.EmrAddStep(self, "Task",
    cluster_id="ClusterId",
    name="StepName",
    jar="Jar",
    action_on_failure=tasks.ActionOnFailure.CONTINUE
)

To specify a custom runtime role use the executionRoleArn property.

Note: The EMR cluster must be created with a security configuration and the runtime role must have a specific trust policy. See this blog post for more details.

import aws_cdk.aws_emr as emr


cfn_security_configuration = emr.CfnSecurityConfiguration(self, "EmrSecurityConfiguration",
    name="AddStepRuntimeRoleSecConfig",
    security_configuration=JSON.parse("""
            {
              "AuthorizationConfiguration": {
                  "IAMConfiguration": {
                      "EnableApplicationScopedIAMRole": true,
                      "ApplicationScopedIAMRoleConfiguration":
                          {
                              "PropagateSourceIdentity": true
                          }
                  },
                  "LakeFormationConfiguration": {
                      "AuthorizedSessionTagValue": "Amazon EMR"
                  }
              }
            }""")
)

task = tasks.EmrCreateCluster(self, "Create Cluster",
    instances=tasks.EmrCreateCluster.InstancesConfigProperty(),
    name=sfn.TaskInput.from_json_path_at("$.ClusterName").value,
    security_configuration=cfn_security_configuration.name
)

execution_role = iam.Role(self, "Role",
    assumed_by=iam.ArnPrincipal(task.cluster_role.role_arn)
)

execution_role.assume_role_policy.add_statements(
    iam.PolicyStatement(
        effect=iam.Effect.ALLOW,
        principals=[task.cluster_role
        ],
        actions=["sts:SetSourceIdentity"
        ]
    ),
    iam.PolicyStatement(
        effect=iam.Effect.ALLOW,
        principals=[task.cluster_role
        ],
        actions=["sts:TagSession"
        ],
        conditions={
            "StringEquals": {
                "aws:RequestTag/LakeFormationAuthorizedCaller": "Amazon EMR"
            }
        }
    ))

tasks.EmrAddStep(self, "Task",
    cluster_id="ClusterId",
    execution_role_arn=execution_role.role_arn,
    name="StepName",
    jar="Jar",
    action_on_failure=tasks.ActionOnFailure.CONTINUE
)

Cancel Step

Cancels a pending step in a running cluster. Corresponds to the cancelSteps API in EMR.

tasks.EmrCancelStep(self, "Task",
    cluster_id="ClusterId",
    step_id="StepId"
)

Modify Instance Fleet

Modifies the target On-Demand and target Spot capacities for the instance fleet with the specified InstanceFleetName.

Corresponds to the modifyInstanceFleet API in EMR.

tasks.EmrModifyInstanceFleetByName(self, "Task",
    cluster_id="ClusterId",
    instance_fleet_name="InstanceFleetName",
    target_on_demand_capacity=2,
    target_spot_capacity=0
)

Modify Instance Group

Modifies the number of nodes and configuration settings of an instance group.

Corresponds to the modifyInstanceGroups API in EMR.

tasks.EmrModifyInstanceGroupByName(self, "Task",
    cluster_id="ClusterId",
    instance_group_name=sfn.JsonPath.string_at("$.InstanceGroupName"),
    instance_group=tasks.EmrModifyInstanceGroupByName.InstanceGroupModifyConfigProperty(
        instance_count=1
    )
)

EMR on EKS

Step Functions supports Amazon EMR on EKS through the service integration pattern. The service integration APIs correspond to Amazon EMR on EKS APIs, but differ in the parameters that are used.

Read more about the differences when using these service integrations.

Setting up the EKS cluster is required.

Create Virtual Cluster

The CreateVirtualCluster API creates a single virtual cluster that’s mapped to a single Kubernetes namespace.

The EKS cluster containing the Kubernetes namespace where the virtual cluster will be mapped can be passed in from the task input.

tasks.EmrContainersCreateVirtualCluster(self, "Create a Virtual Cluster",
    eks_cluster=tasks.EksClusterInput.from_task_input(sfn.TaskInput.from_text("clusterId"))
)

The EKS cluster can also be passed in directly.

import aws_cdk.aws_eks as eks

# eks_cluster: eks.Cluster


tasks.EmrContainersCreateVirtualCluster(self, "Create a Virtual Cluster",
    eks_cluster=tasks.EksClusterInput.from_cluster(eks_cluster)
)

By default, the Kubernetes namespace that a virtual cluster maps to is “default”, but a specific namespace within an EKS cluster can be selected.

tasks.EmrContainersCreateVirtualCluster(self, "Create a Virtual Cluster",
    eks_cluster=tasks.EksClusterInput.from_task_input(sfn.TaskInput.from_text("clusterId")),
    eks_namespace="specified-namespace"
)

Delete Virtual Cluster

The DeleteVirtualCluster API deletes a virtual cluster.

tasks.EmrContainersDeleteVirtualCluster(self, "Delete a Virtual Cluster",
    virtual_cluster_id=sfn.TaskInput.from_json_path_at("$.virtualCluster")
)

Start Job Run

The StartJobRun API starts a job run. A job is a unit of work that you submit to Amazon EMR on EKS for execution. The work performed by the job can be defined by a Spark jar, PySpark script, or SparkSQL query. A job run is an execution of the job on the virtual cluster.

Required setup:

If not done already, follow the steps to setup EMR on EKS and create an EKS Cluster.
Enable Cluster access
Enable IAM Role access

The following actions must be performed if the virtual cluster ID is supplied from the task input. Otherwise, if it is supplied statically in the state machine definition, these actions will be done automatically.

Create an IAM role
Update the Role Trust Policy of the Job Execution Role.

The job can be configured with spark submit parameters:

tasks.EmrContainersStartJobRun(self, "EMR Containers Start Job Run",
    virtual_cluster=tasks.VirtualClusterInput.from_virtual_cluster_id("de92jdei2910fwedz"),
    release_label=tasks.ReleaseLabel.EMR_6_2_0,
    job_driver=tasks.JobDriver(
        spark_submit_job_driver=tasks.SparkSubmitJobDriver(
            entry_point=sfn.TaskInput.from_text("local:///usr/lib/spark/examples/src/main/python/pi.py"),
            spark_submit_parameters="--conf spark.executor.instances=2 --conf spark.executor.memory=2G --conf spark.executor.cores=2 --conf spark.driver.cores=1"
        )
    )
)

Configuring the job can also be done via application configuration:

tasks.EmrContainersStartJobRun(self, "EMR Containers Start Job Run",
    virtual_cluster=tasks.VirtualClusterInput.from_virtual_cluster_id("de92jdei2910fwedz"),
    release_label=tasks.ReleaseLabel.EMR_6_2_0,
    job_name="EMR-Containers-Job",
    job_driver=tasks.JobDriver(
        spark_submit_job_driver=tasks.SparkSubmitJobDriver(
            entry_point=sfn.TaskInput.from_text("local:///usr/lib/spark/examples/src/main/python/pi.py")
        )
    ),
    application_config=[tasks.ApplicationConfiguration(
        classification=tasks.Classification.SPARK_DEFAULTS,
        properties={
            "spark.executor.instances": "1",
            "spark.executor.memory": "512M"
        }
    )]
)

Job monitoring can be enabled if monitoring.logging is set true. This automatically generates an S3 bucket and CloudWatch logs.

tasks.EmrContainersStartJobRun(self, "EMR Containers Start Job Run",
    virtual_cluster=tasks.VirtualClusterInput.from_virtual_cluster_id("de92jdei2910fwedz"),
    release_label=tasks.ReleaseLabel.EMR_6_2_0,
    job_driver=tasks.JobDriver(
        spark_submit_job_driver=tasks.SparkSubmitJobDriver(
            entry_point=sfn.TaskInput.from_text("local:///usr/lib/spark/examples/src/main/python/pi.py"),
            spark_submit_parameters="--conf spark.executor.instances=2 --conf spark.executor.memory=2G --conf spark.executor.cores=2 --conf spark.driver.cores=1"
        )
    ),
    monitoring=tasks.Monitoring(
        logging=True
    )
)

Otherwise, providing monitoring for jobs with existing log groups and log buckets is also available.

import aws_cdk.aws_logs as logs


log_group = logs.LogGroup(self, "Log Group")
log_bucket = s3.Bucket(self, "S3 Bucket")

tasks.EmrContainersStartJobRun(self, "EMR Containers Start Job Run",
    virtual_cluster=tasks.VirtualClusterInput.from_virtual_cluster_id("de92jdei2910fwedz"),
    release_label=tasks.ReleaseLabel.EMR_6_2_0,
    job_driver=tasks.JobDriver(
        spark_submit_job_driver=tasks.SparkSubmitJobDriver(
            entry_point=sfn.TaskInput.from_text("local:///usr/lib/spark/examples/src/main/python/pi.py"),
            spark_submit_parameters="--conf spark.executor.instances=2 --conf spark.executor.memory=2G --conf spark.executor.cores=2 --conf spark.driver.cores=1"
        )
    ),
    monitoring=tasks.Monitoring(
        log_group=log_group,
        log_bucket=log_bucket
    )
)

Users can provide their own existing Job Execution Role.

tasks.EmrContainersStartJobRun(self, "EMR Containers Start Job Run",
    virtual_cluster=tasks.VirtualClusterInput.from_task_input(sfn.TaskInput.from_json_path_at("$.VirtualClusterId")),
    release_label=tasks.ReleaseLabel.EMR_6_2_0,
    job_name="EMR-Containers-Job",
    execution_role=iam.Role.from_role_arn(self, "Job-Execution-Role", "arn:aws:iam::xxxxxxxxxxxx:role/JobExecutionRole"),
    job_driver=tasks.JobDriver(
        spark_submit_job_driver=tasks.SparkSubmitJobDriver(
            entry_point=sfn.TaskInput.from_text("local:///usr/lib/spark/examples/src/main/python/pi.py"),
            spark_submit_parameters="--conf spark.executor.instances=2 --conf spark.executor.memory=2G --conf spark.executor.cores=2 --conf spark.driver.cores=1"
        )
    )
)

EKS

Step Functions supports Amazon EKS through the service integration pattern. The service integration APIs correspond to Amazon EKS APIs.

Read more about the differences when using these service integrations.

Call

Read and write Kubernetes resource objects via a Kubernetes API endpoint. Corresponds to the call API in Step Functions Connector.

The following code snippet includes a Task state that uses eks:call to list the pods.

import aws_cdk.aws_eks as eks


my_eks_cluster = eks.Cluster(self, "my sample cluster",
    version=eks.KubernetesVersion.V1_18,
    cluster_name="myEksCluster"
)

tasks.EksCall(self, "Call a EKS Endpoint",
    cluster=my_eks_cluster,
    http_method=tasks.HttpMethods.GET,
    http_path="/api/v1/namespaces/default/pods"
)

EventBridge

Step Functions supports Amazon EventBridge through the service integration pattern. The service integration APIs correspond to Amazon EventBridge APIs.

Read more about the differences when using these service integrations.

Put Events

Send events to an EventBridge bus. Corresponds to the put-events API in Step Functions Connector.

The following code snippet includes a Task state that uses events:putevents to send an event to the default bus.

import aws_cdk.aws_events as events


my_event_bus = events.EventBus(self, "EventBus",
    event_bus_name="MyEventBus1"
)

tasks.EventBridgePutEvents(self, "Send an event to EventBridge",
    entries=[tasks.EventBridgePutEventsEntry(
        detail=sfn.TaskInput.from_object({
            "Message": "Hello from Step Functions!"
        }),
        event_bus=my_event_bus,
        detail_type="MessageFromStepFunctions",
        source="step.functions"
    )]
)

Glue

Step Functions supports AWS Glue through the service integration pattern.

StartJobRun

You can call the StartJobRun API from a Task state.

tasks.GlueStartJobRun(self, "Task",
    glue_job_name="my-glue-job",
    arguments=sfn.TaskInput.from_object({
        "key": "value"
    }),
    task_timeout=sfn.Timeout.duration(Duration.minutes(30)),
    notify_delay_after=Duration.minutes(5)
)

StartCrawlerRun

You can call the StartCrawler API from a Task state through AWS SDK service integrations.

import aws_cdk.aws_glue as glue

# my_crawler: glue.CfnCrawler


# You can get the crawler name from `crawler.ref`
tasks.GlueStartCrawlerRun(self, "Task1",
    crawler_name=my_crawler.ref
)

# Of course, you can also specify the crawler name directly.
tasks.GlueStartCrawlerRun(self, "Task2",
    crawler_name="my-crawler-job"
)

Glue DataBrew

Step Functions supports AWS Glue DataBrew through the service integration pattern.

Start Job Run

You can call the StartJobRun API from a Task state.

tasks.GlueDataBrewStartJobRun(self, "Task",
    name="databrew-job"
)

Invoke HTTP API

Step Functions supports calling third-party APIs with credentials managed by Amazon EventBridge Connections.

The following snippet creates a new API destination connection, and uses it to make a POST request to the specified URL. The endpoint response is available at the $.ResponseBody path.

import aws_cdk.aws_events as events


connection = events.Connection(self, "Connection",
    authorization=events.Authorization.basic("username", SecretValue.unsafe_plain_text("password"))
)

tasks.HttpInvoke(self, "Invoke HTTP API",
    api_root="https://api.example.com",
    api_endpoint=sfn.TaskInput.from_text("https://api.example.com/path/to/resource"),
    body=sfn.TaskInput.from_object({"foo": "bar"}),
    connection=connection,
    headers=sfn.TaskInput.from_object({"Content-Type": "application/json"}),
    method=sfn.TaskInput.from_text("POST"),
    query_string_parameters=sfn.TaskInput.from_object({"id": "123"}),
    url_encoding_format=tasks.URLEncodingFormat.BRACKETS
)

Lambda

Step Functions supports AWS Lambda through the service integration pattern.

Invoke

Invoke a Lambda function.

You can specify the input to your Lambda function through the payload attribute. By default, Step Functions invokes Lambda function with the state input (JSON path ‘$’) as the input.

The following snippet invokes a Lambda Function with the state input as the payload by referencing the $ path.

# fn: lambda.Function

tasks.LambdaInvoke(self, "Invoke with state input",
    lambda_function=fn
)

When a function is invoked, the Lambda service sends these response elements back.

⚠️ The response from the Lambda function is in an attribute called Payload

The following snippet invokes a Lambda Function by referencing the $.Payload path to reference the output of a Lambda executed before it.

# fn: lambda.Function

tasks.LambdaInvoke(self, "Invoke with empty object as payload",
    lambda_function=fn,
    payload=sfn.TaskInput.from_object({})
)

# use the output of fn as input
tasks.LambdaInvoke(self, "Invoke with payload field in the state input",
    lambda_function=fn,
    payload=sfn.TaskInput.from_json_path_at("$.Payload")
)

The following snippet invokes a Lambda and sets the task output to only include the Lambda function response.

# fn: lambda.Function

tasks.LambdaInvoke(self, "Invoke and set function response as task output",
    lambda_function=fn,
    output_path="$.Payload"
)

If you want to combine the input and the Lambda function response you can use the payloadResponseOnly property and specify the resultPath. This will put the Lambda function ARN directly in the “Resource” string, but it conflicts with the integrationPattern, invocationType, clientContext, and qualifier properties.

# fn: lambda.Function

tasks.LambdaInvoke(self, "Invoke and combine function response with task input",
    lambda_function=fn,
    payload_response_only=True,
    result_path="$.fn"
)

You can have Step Functions pause a task, and wait for an external process to return a task token. Read more about the callback pattern

To use the callback pattern, set the token property on the task. Call the Step Functions SendTaskSuccess or SendTaskFailure APIs with the token to indicate that the task has completed and the state machine should resume execution.

The following snippet invokes a Lambda with the task token as part of the input to the Lambda.

# fn: lambda.Function

tasks.LambdaInvoke(self, "Invoke with callback",
    lambda_function=fn,
    integration_pattern=sfn.IntegrationPattern.WAIT_FOR_TASK_TOKEN,
    payload=sfn.TaskInput.from_object({
        "token": sfn.JsonPath.task_token,
        "input": sfn.JsonPath.string_at("$.someField")
    })
)

⚠️ The task will pause until it receives that task token back with a SendTaskSuccess or SendTaskFailure call. Learn more about Callback with the Task Token.

AWS Lambda can occasionally experience transient service errors. In this case, invoking Lambda results in a 500 error, such as ClientExecutionTimeoutException, ServiceException, AWSLambdaException, or SdkClientException. As a best practice, the LambdaInvoke task will retry on those errors with an interval of 2 seconds, a back-off rate of 2 and 6 maximum attempts. Set the retryOnServiceExceptions prop to false to disable this behavior.

SageMaker

Step Functions supports AWS SageMaker through the service integration pattern.

If your training job or model uses resources from AWS Marketplace, network isolation is required. To do so, set the enableNetworkIsolation property to true for SageMakerCreateModel or SageMakerCreateTrainingJob.

To set environment variables for the Docker container use the environment property.

Create Training Job

You can call the CreateTrainingJob API from a Task state.

tasks.SageMakerCreateTrainingJob(self, "TrainSagemaker",
    training_job_name=sfn.JsonPath.string_at("$.JobName"),
    algorithm_specification=tasks.AlgorithmSpecification(
        algorithm_name="BlazingText",
        training_input_mode=tasks.InputMode.FILE
    ),
    input_data_config=[tasks.Channel(
        channel_name="train",
        data_source=tasks.DataSource(
            s3_data_source=tasks.S3DataSource(
                s3_data_type=tasks.S3DataType.S3_PREFIX,
                s3_location=tasks.S3Location.from_json_expression("$.S3Bucket")
            )
        )
    )],
    output_data_config=tasks.OutputDataConfig(
        s3_output_location=tasks.S3Location.from_bucket(s3.Bucket.from_bucket_name(self, "Bucket", "mybucket"), "myoutputpath")
    ),
    resource_config=tasks.ResourceConfig(
        instance_count=1,
        instance_type=ec2.InstanceType(sfn.JsonPath.string_at("$.InstanceType")),
        volume_size=Size.gibibytes(50)
    ),  # optional: default is 1 instance of EC2 `M4.XLarge` with `10GB` volume
    stopping_condition=tasks.StoppingCondition(
        max_runtime=Duration.hours(2)
    )
)

You can specify TrainingInputMode via the trainingInputMode property.

To download the data from Amazon Simple Storage Service (Amazon S3) to the provisioned ML storage volume, and mount the directory to a Docker volume, choose InputMode.FILE if an algorithm supports it.
To stream data directly from Amazon S3 to the container, choose InputMode.PIPE if an algorithm supports it.
To stream data directly from Amazon S3 to the container with no code changes and to provide file system access to the data, choose InputMode.FAST_FILE if an algorithm supports it.

Create Transform Job

You can call the CreateTransformJob API from a Task state.

tasks.SageMakerCreateTransformJob(self, "Batch Inference",
    transform_job_name="MyTransformJob",
    model_name="MyModelName",
    model_client_options=tasks.ModelClientOptions(
        invocations_max_retries=3,  # default is 0
        invocations_timeout=Duration.minutes(5)
    ),
    transform_input=tasks.TransformInput(
        transform_data_source=tasks.TransformDataSource(
            s3_data_source=tasks.TransformS3DataSource(
                s3_uri="s3://inputbucket/train",
                s3_data_type=tasks.S3DataType.S3_PREFIX
            )
        )
    ),
    transform_output=tasks.TransformOutput(
        s3_output_path="s3://outputbucket/TransformJobOutputPath"
    ),
    transform_resources=tasks.TransformResources(
        instance_count=1,
        instance_type=ec2.InstanceType.of(ec2.InstanceClass.M4, ec2.InstanceSize.XLARGE)
    )
)

Create Endpoint

You can call the CreateEndpoint API from a Task state.

tasks.SageMakerCreateEndpoint(self, "SagemakerEndpoint",
    endpoint_name=sfn.JsonPath.string_at("$.EndpointName"),
    endpoint_config_name=sfn.JsonPath.string_at("$.EndpointConfigName")
)

Create Endpoint Config

You can call the CreateEndpointConfig API from a Task state.

tasks.SageMakerCreateEndpointConfig(self, "SagemakerEndpointConfig",
    endpoint_config_name="MyEndpointConfig",
    production_variants=[tasks.ProductionVariant(
        initial_instance_count=2,
        instance_type=ec2.InstanceType.of(ec2.InstanceClass.M5, ec2.InstanceSize.XLARGE),
        model_name="MyModel",
        variant_name="awesome-variant"
    )]
)

Create Model

You can call the CreateModel API from a Task state.

tasks.SageMakerCreateModel(self, "Sagemaker",
    model_name="MyModel",
    primary_container=tasks.ContainerDefinition(
        image=tasks.DockerImage.from_json_expression(sfn.JsonPath.string_at("$.Model.imageName")),
        mode=tasks.Mode.SINGLE_MODEL,
        model_s3_location=tasks.S3Location.from_json_expression("$.TrainingJob.ModelArtifacts.S3ModelArtifacts")
    )
)

Update Endpoint

You can call the UpdateEndpoint API from a Task state.

tasks.SageMakerUpdateEndpoint(self, "SagemakerEndpoint",
    endpoint_name=sfn.JsonPath.string_at("$.Endpoint.Name"),
    endpoint_config_name=sfn.JsonPath.string_at("$.Endpoint.EndpointConfig")
)

SNS

Step Functions supports Amazon SNS through the service integration pattern.

Publish

You can call the Publish API from a Task state to publish to an SNS topic.

topic = sns.Topic(self, "Topic")

# Use a field from the execution data as message.
task1 = tasks.SnsPublish(self, "Publish1",
    topic=topic,
    integration_pattern=sfn.IntegrationPattern.REQUEST_RESPONSE,
    message=sfn.TaskInput.from_data_at("$.state.message"),
    message_attributes={
        "place": tasks.MessageAttribute(
            value=sfn.JsonPath.string_at("$.place")
        ),
        "pic": tasks.MessageAttribute(
            # BINARY must be explicitly set
            data_type=tasks.MessageAttributeDataType.BINARY,
            value=sfn.JsonPath.string_at("$.pic")
        ),
        "people": tasks.MessageAttribute(
            value=4
        ),
        "handles": tasks.MessageAttribute(
            value=["@kslater", "@jjf", null, "@mfanning"]
        )
    }
)

# Combine a field from the execution data with
# a literal object.
task2 = tasks.SnsPublish(self, "Publish2",
    topic=topic,
    message=sfn.TaskInput.from_object({
        "field1": "somedata",
        "field2": sfn.JsonPath.string_at("$.field2")
    })
)

Step Functions

Step Functions supports AWS Step Functions through the service integration pattern.

Start Execution

You can manage AWS Step Functions executions.

AWS Step Functions supports it’s own StartExecution API as a service integration.

# Define a state machine with one Pass state
child = sfn.StateMachine(self, "ChildStateMachine",
    definition=sfn.Chain.start(sfn.Pass(self, "PassState"))
)

# Include the state machine in a Task state with callback pattern
task = tasks.StepFunctionsStartExecution(self, "ChildTask",
    state_machine=child,
    integration_pattern=sfn.IntegrationPattern.WAIT_FOR_TASK_TOKEN,
    input=sfn.TaskInput.from_object({
        "token": sfn.JsonPath.task_token,
        "foo": "bar"
    }),
    name="MyExecutionName"
)

# Define a second state machine with the Task state above
sfn.StateMachine(self, "ParentStateMachine",
    definition=task
)

You can utilize Associate Workflow Executions via the associateWithParent property. This allows the Step Functions UI to link child executions from parent executions, making it easier to trace execution flow across state machines.

# child: sfn.StateMachine

task = tasks.StepFunctionsStartExecution(self, "ChildTask",
    state_machine=child,
    associate_with_parent=True
)

This will add the payload AWS_STEP_FUNCTIONS_STARTED_BY_EXECUTION_ID.$: $$.Execution.Id to the inputproperty for you, which will pass the execution ID from the context object to the execution input. It requires input to be an object or not be set at all.

Invoke Activity

You can invoke a Step Functions Activity which enables you to have a task in your state machine where the work is performed by a worker that can be hosted on Amazon EC2, Amazon ECS, AWS Lambda, basically anywhere. Activities are a way to associate code running somewhere (known as an activity worker) with a specific task in a state machine.

When Step Functions reaches an activity task state, the workflow waits for an activity worker to poll for a task. An activity worker polls Step Functions by using GetActivityTask, and sending the ARN for the related activity.

After the activity worker completes its work, it can provide a report of its success or failure by using SendTaskSuccess or SendTaskFailure. These two calls use the taskToken provided by GetActivityTask to associate the result with that task.

The following example creates an activity and creates a task that invokes the activity.

submit_job_activity = sfn.Activity(self, "SubmitJob")

tasks.StepFunctionsInvokeActivity(self, "Submit Job",
    activity=submit_job_activity
)

Use the Parameters field to create a collection of key-value pairs that are passed as input. The values of each can either be static values that you include in your state machine definition, or selected from either the input or the context object with a path.

submit_job_activity = sfn.Activity(self, "SubmitJob")

tasks.StepFunctionsInvokeActivity(self, "Submit Job",
    activity=submit_job_activity,
    parameters={
        "comment": "Selecting what I care about.",
        "MyDetails": {
            "size": sfn.JsonPath.string_at("$.product.details.size"),
            "exists": sfn.JsonPath.string_at("$.product.availability"),
            "StaticValue": "foo"
        }
    }
)

SQS

Step Functions supports Amazon SQS

Send Message

You can call the SendMessage API from a Task state to send a message to an SQS queue.

queue = sqs.Queue(self, "Queue")

# Use a field from the execution data as message.
task1 = tasks.SqsSendMessage(self, "Send1",
    queue=queue,
    message_body=sfn.TaskInput.from_json_path_at("$.message")
)

# Combine a field from the execution data with
# a literal object.
task2 = tasks.SqsSendMessage(self, "Send2",
    queue=queue,
    message_body=sfn.TaskInput.from_object({
        "field1": "somedata",
        "field2": sfn.JsonPath.string_at("$.field2")
    })
)