In modern data architectures, it is common to manage multiple ETL pipelines that must run in sequence or in parallel. AWS Glue provides a robust framework for building workflows, but when we need to orchestrate two or more Glue Workflows together, AWS Step Functions becomes the natural choice.

In this post, we will explain how to connect multiple Glue Workflows using Step Functions, so that one workflow can trigger another, and the execution can be monitored in a controlled manner.

Why Use Step Functions with Glue Workflows?

Although Glue Workflows allow you to chain jobs and crawlers, Step Functions provide several additional advantages:

• Cross-service orchestration: Not limited to Glue; you can also trigger Lambda functions, SNS notifications, or even invoke APIs.
• Error handling and retries: Fine-grained control when a step fails.
• Visualization: A clear state machine diagram to monitor execution.
• Conditional execution: Run Workflow B only if Workflow A succeeds, or take an alternate path if it fails.

Step Functions and Glue Integration

Step Functions provide a direct integration with Glue through the following actions:

• arn:aws:states:::glue:startWorkflowRun → Starts a Glue Workflow.
• arn:aws:states:::glue:getWorkflowRun → Retrieves the status of a Glue Workflow run.

By combining these two, we can implement a simple pattern:

1. Start Workflow A.
2. Wait for completion.
3. If successful, trigger Workflow B.
4. If failure, stop or take an alternative action.

Example: Sequential Execution of Two Workflows

Below is a simplified Step Functions state machine definition in Amazon States Language (ASL):

{
  "Comment": "Execute two Glue Workflows sequentially",
  "StartAt": "Start Workflow A",
  "States": {
    "Start Workflow A": {
      "Type": "Task",
      "Resource": "arn:aws:states:::glue:startWorkflowRun",
      "Parameters": {
        "Name": "WorkflowA"
      },
      "ResultPath": "$.WorkflowA",
      "Next": "Wait A"
    },
    "Wait A": {
      "Type": "Wait",
      "Seconds": 60,
      "Next": "Check A"
    },
    "Check A": {
      "Type": "Task",
      "Resource": "arn:aws:states:::glue:getWorkflowRun",
      "Parameters": {
        "Name": "WorkflowA",
        "RunId.$": "$.WorkflowA.RunId",
        "IncludeGraph": false
      },
      "ResultPath": "$.StatusA",
      "Next": "Decide A"
    },
    "Decide A": {
      "Type": "Choice",
      "Choices": [
        {
          "Variable": "$.StatusA.Run.Status",
          "StringEquals": "SUCCEEDED",
          "Next": "Start Workflow B"
        }
      ],
      "Default": "Fail A"
    },
    "Start Workflow B": {
      "Type": "Task",
      "Resource": "arn:aws:states:::glue:startWorkflowRun",
      "Parameters": {
        "Name": "WorkflowB"
      },
      "End": true
    },
    "Fail A": {
      "Type": "Fail",
      "Error": "WorkflowAFailed",
      "Cause": "Workflow A did not finish successfully"
    }
  }
}

This definition ensures that Workflow B only runs if Workflow A finishes with a SUCCEEDED status. Otherwise, the Step Function execution will fail, and you can configure alerts accordingly.

---

Best Practices

• Use retries and exponential backoff: Step Functions allow retries with backoff for transient Glue issues.
• Parameterize workflows: Pass dynamic parameters to workflows if you need flexible execution.
• Monitoring and alerts: Integrate with CloudWatch or SNS for notifications.
• Optimize wait logic: Instead of a fixed Wait, you can implement a loop with shorter intervals or exponential delays.

---

Conclusion

By combining AWS Glue Workflows with Step Functions, you gain powerful orchestration capabilities, ensuring your ETL pipelines are reliable, resilient, and easy to monitor. This approach lets you build complex data pipelines that span multiple workflows, while maintaining control over execution flow and error handling.

---

References

• AWS Step Functions + Glue: https://docs.aws.amazon.com/step-functions/latest/dg/connect-glue.html
• Glue Workflow API: https://docs.aws.amazon.com/glue/latest/dg/aws-glue-api-workflow-pysdk.html