Skip to main content

Workflow Control Flow

Workflow Control Flow defines the execution path and behavior of a workflow, particularly how it manages dependencies, sequences operations, and responds to failures within its constituent nodes. Its primary purpose is to ensure predictable, robust, and efficient execution of multi-step processes, allowing developers to orchestrate complex operations with defined error handling strategies.

Core Features​

The core features of Workflow Control Flow revolve around managing the execution lifecycle and defining responses to various events, most notably node failures.

Failure Handling with WorkflowFailurePolicy​

A critical aspect of workflow control flow is how the system reacts when an individual node within a workflow fails. The WorkflowFailurePolicy mechanism provides explicit control over this behavior, allowing developers to define the workflow's state transition upon observing a node execution failure.

The WorkflowFailurePolicy enum offers two distinct strategies:

  • FAIL_IMMEDIATELY: This policy dictates that the entire workflow execution will transition to a failed state as soon as any component node fails. No further runnable nodes will be initiated, and any currently running nodes may be terminated, depending on the underlying execution engine.

    • Use Case: Employ FAIL_IMMEDIATELY for workflows where the failure of any single step renders subsequent steps meaningless or potentially harmful. Examples include critical data transformations where corrupted intermediate data would invalidate downstream processing, or deployment pipelines where a failed build step should halt the entire release process.
    from your_workflow_package import WorkflowFailurePolicy

    # Example: Configuring a workflow to fail immediately on any node failure
    my_workflow_definition.on_failure_policy = WorkflowFailurePolicy.FAIL_IMMEDIATELY

  • FAIL_AFTER_EXECUTABLE_NODES_COMPLETE: With this policy, if a component node fails, the workflow execution will continue to run any other nodes that are still runnable and not dependent on the failed node. The workflow will only transition to a failed state once all other executable nodes have either completed successfully or failed themselves.

    • Use Case: Choose FAIL_AFTER_EXECUTABLE_NODES_COMPLETE when a partial failure does not necessitate an immediate halt of the entire workflow. This is useful for scenarios where independent cleanup tasks or non-critical parallel operations can proceed even if one part of the workflow encounters an issue. For instance, in a data ingestion pipeline, if one data source fails, other independent sources might still be processed, or logging/notification steps can complete.
    from your_workflow_package import WorkflowFailurePolicy

    # Example: Configuring a workflow to complete other tasks before failing
    my_workflow_definition.on_failure_policy = WorkflowFailurePolicy.FAIL_AFTER_EXECUTABLE_NODES_COMPLETE

Common Use Cases​

Workflow Control Flow is fundamental to building robust and scalable automated systems.

  • Data Processing Pipelines: Orchestrating complex ETL (Extract, Transform, Load) jobs where data validation, transformation, and loading steps must execute in a specific order, with defined error handling for data quality issues.
  • Automated Deployment and Release Management: Managing multi-stage deployments, including building artifacts, running tests, deploying to various environments, and performing post-deployment validations. Control flow ensures that a failure in an early stage (e.g., build) prevents progression to later, more critical stages (e.g., production deployment).
  • Business Process Automation: Automating multi-step business processes, such as order fulfillment, customer onboarding, or financial transaction processing, where each step might involve different systems and require specific conditional logic or failure recovery.
  • Machine Learning Model Training and Deployment: Sequencing data preparation, model training, evaluation, and deployment steps. Control flow ensures that models are only deployed if they meet performance criteria and that training failures are handled gracefully.

Best Practices and Considerations​

  • Policy Selection: Carefully consider the criticality of each workflow and its constituent nodes when selecting a WorkflowFailurePolicy. For mission-critical workflows where any failure is unacceptable, FAIL_IMMEDIATELY is often the safer choice. For workflows with independent branches or where partial results are acceptable, FAIL_AFTER_EXECUTABLE_NODES_COMPLETE can improve resilience and resource utilization.
  • Granularity of Failure Handling: While WorkflowFailurePolicy defines the workflow-level behavior, consider implementing node-level retry mechanisms or specific error handling within individual nodes for more fine-grained control.
  • Observability: Ensure that workflows are instrumented with logging and monitoring to provide clear visibility into execution status, node failures, and the application of the chosen failure policy. This is crucial for debugging and understanding workflow behavior in production.
  • Resource Management: The FAIL_AFTER_EXECUTABLE_NODES_COMPLETE policy might lead to continued resource consumption even after a critical node has failed. Design workflows to manage resources effectively under both success and failure scenarios.