Construct multi-step purposes and AI workflows with AWS Lambda sturdy features

Fashionable purposes more and more require complicated and long-running coordination between companies, equivalent to multi-step cost processing, AI agent orchestration, or approval processes awaiting human selections. Constructing these historically required vital effort to implement state administration, deal with failures, and combine a number of infrastructure companies.

Beginning in the present day, you should use AWS Lambda sturdy features to construct dependable multi-step purposes straight inside the acquainted AWS Lambda expertise. Sturdy features are common Lambda features with the identical occasion handler and integrations you already know. You write sequential code in your most well-liked programming language, and sturdy features observe progress, mechanically retry on failures, and droop execution for as much as one yr at outlined factors, with out paying for idle compute throughout waits.

AWS Lambda sturdy features use a checkpoint and replay mechanism, generally known as sturdy execution, to ship these capabilities. After enabling a operate for sturdy execution, you add the brand new open supply sturdy execution SDK to your operate code. You then use SDK primitives like “steps” so as to add automated checkpointing and retries to your small business logic and “waits” to effectively droop execution with out compute prices. When execution terminates unexpectedly, Lambda resumes from the final checkpoint, replaying your occasion handler from the start whereas skipping accomplished operations.

Getting began with AWS Lambda sturdy features

Let me stroll you thru tips on how to use sturdy features.

First, I create a brand new Lambda operate within the console and choose Writer from scratch. Within the Sturdy execution part, I choose Allow. Notice that, sturdy operate setting can solely be set throughout operate creation and at present can’t be modified for current Lambda features.

After I create my Lambda sturdy operate, I can get began with the supplied code.

Lambda sturdy features introduces two core primitives that deal with state administration and restoration:

• Steps—The context.step() methodology provides automated retries and checkpointing to your small business logic. After a step is accomplished, will probably be skipped throughout replay.
• Wait—The context.wait() methodology pauses execution for a specified length, terminating the operate, suspending and resuming execution with out compute prices.

Moreover, Lambda sturdy features offers different operations for extra complicated patterns: create_callback() creates a callback that you should use to await outcomes for exterior occasions like API responses or human approvals, wait_for_condition() pauses till a particular situation is met like polling a REST API for course of completion, and parallel() or map() operations for superior concurrency use circumstances.

Constructing a production-ready order processing workflow

Now let’s increase the default instance to construct a production-ready order processing workflow. This demonstrates tips on how to use callbacks for exterior approvals, deal with errors correctly, and configure retry methods. I maintain the code deliberately concise to give attention to these core ideas. In a full implementation, you possibly can improve the validation step with Amazon Bedrock so as to add AI-powered order evaluation.

Right here’s how the order processing workflow works:

• First, validate_order() checks order knowledge to make sure all required fields are current.
• Subsequent, send_for_approval() sends the order for exterior human approval and waits for a callback response, suspending execution with out compute prices.
• Then, process_order() completes order processing.
• All through the workflow, try-catch error dealing with distinguishes between terminal errors that cease execution instantly and recoverable errors inside steps that set off automated retries.

Right here’s the whole order processing workflow with step definitions and the principle handler:

import random
from aws_durable_execution_sdk_python import (
    DurableContext,
    StepContext,
    durable_execution,
    durable_step,
)
from aws_durable_execution_sdk_python.config import (
    Length,
    StepConfig,
    CallbackConfig,
)
from aws_durable_execution_sdk_python.retries import (
    RetryStrategyConfig,
    create_retry_strategy,
)


@durable_step
def validate_order(step_context: StepContext, order_id: str) -> dict:
    """Validates order knowledge utilizing AI."""
    step_context.logger.data(f"Validating order: {order_id}")
    # In manufacturing: calls Amazon Bedrock to validate order completeness and accuracy
    return {"order_id": order_id, "standing": "validated"}


@durable_step
def send_for_approval(step_context: StepContext, callback_id: str, order_id: str) -> dict:
    """Sends order for approval utilizing the supplied callback token."""
    step_context.logger.data(f"Sending order {order_id} for approval with callback_id: {callback_id}")
    
    # In manufacturing: ship callback_id to exterior approval system
    # The exterior system will name Lambda SendDurableExecutionCallbackSuccess or
    # SendDurableExecutionCallbackFailure APIs with this callback_id when approval is full
    
    return {
        "order_id": order_id,
        "callback_id": callback_id,
        "standing": "sent_for_approval"
    }


@durable_step
def process_order(step_context: StepContext, order_id: str) -> dict:
    """Processes the order with retry logic for transient failures."""
    step_context.logger.data(f"Processing order: {order_id}")
    # Simulate flaky API that generally fails
    if random.random() > 0.4:
        step_context.logger.data("Processing failed, will retry")
        elevate Exception("Processing failed")
    return {
        "order_id": order_id,
        "standing": "processed",
        "timestamp": "2025-11-27T10:00:00Z",
    }


@durable_execution
def lambda_handler(occasion: dict, context: DurableContext) -> dict:
    strive:
        order_id = occasion.get("order_id")
        
        # Step 1: Validate the order
        validated = context.step(validate_order(order_id))
        if validated["status"] != "validated":
            elevate Exception("Validation failed")  # Terminal error - stops execution
        context.logger.data(f"Order validated: {validated}")
        
        # Step 2: Create callback
        callback = context.create_callback(
            identify="awaiting-approval",
            config=CallbackConfig(timeout=Length.from_minutes(3))
        )
        context.logger.data(f"Created callback with id: {callback.callback_id}")
        
        # Step 3: Ship for approval with the callback_id
        approval_request = context.step(send_for_approval(callback.callback_id, order_id))
        context.logger.data(f"Approval request despatched: {approval_request}")
        
        # Step 4: Look ahead to the callback outcome
        # This blocks till exterior system calls SendDurableExecutionCallbackSuccess or SendDurableExecutionCallbackFailure
        approval_result = callback.outcome()
        context.logger.data(f"Approval obtained: {approval_result}")
        
        # Step 5: Course of the order with customized retry technique
        retry_config = RetryStrategyConfig(max_attempts=3, backoff_rate=2.0)
        processed = context.step(
            process_order(order_id),
            config=StepConfig(retry_strategy=create_retry_strategy(retry_config)),
        )
        if processed["status"] != "processed":
            elevate Exception("Processing failed")  # Terminal error
        
        context.logger.data(f"Order efficiently processed: {processed}")
        return processed
        
    besides Exception as error:
        context.logger.error(f"Error processing order: {error}")
        elevate error  # Re-raise to fail the execution

This code demonstrates a number of necessary ideas:

• Error dealing with—The try-catch block handles terminal errors. When an unhandled exception is thrown exterior of a step (just like the validation test), it terminates the execution instantly. That is helpful when there’s no level in retrying, equivalent to invalid order knowledge.
• Step retries—Contained in the process_order step, exceptions set off automated retries primarily based on the default (step 1) or configured RetryStrategy (step 5). This handles transient failures like short-term API unavailability.
• Logging—I exploit context.logger for the principle handler and step_context.logger inside steps. The context logger suppresses duplicate logs throughout replay.

Now I create a check occasion with order_id and invoke the operate asynchronously to begin the order workflow. I navigate to the Check tab and fill within the non-compulsory Sturdy execution identify to establish this execution. Notice that, sturdy features offers built-in idempotency. If I invoke the operate twice with the identical execution identify, the second invocation returns the prevailing execution outcome as an alternative of making a replica.

I can monitor the execution by navigating to the Sturdy executions tab within the Lambda console:

Right here I can see every step’s standing and timing. The execution exhibits CallbackStarted adopted by InvocationCompleted, which signifies the operate has terminated and execution is suspended to keep away from idle prices whereas ready for the approval callback.

I can now full the callback straight from the console by selecting Ship success or Ship failure, or programmatically utilizing the Lambda API.

I select Ship success.

After the callback completes, the execution resumes and processes the order. If the process_order step fails as a result of simulated flaky API, it mechanically retries primarily based on the configured technique. As soon as all retries succeed, the execution completes efficiently.

Monitoring executions with Amazon EventBridge

You may as well monitor sturdy operate executions utilizing Amazon EventBridge. Lambda mechanically sends execution standing change occasions to the default occasion bus, permitting you to construct downstream workflows, ship notifications, or combine with different AWS companies.

To obtain these occasions, create an EventBridge rule on the default occasion bus with this sample:

{
  "supply": ["aws.lambda"],
  "detail-type": ["Durable Execution Status Change"]
}

Issues to know

Listed below are key factors to notice:

• Availability—Lambda sturdy features at the moment are accessible in US East (Ohio) AWS Area. For the newest Area availability, go to the AWS Capabilities by Area web page.
• Programming language assist—At launch, AWS Lambda sturdy features helps JavaScript/TypeScript (Node.js 22/24) and Python (3.13/3.14). We advocate bundling the sturdy execution SDK together with your operate code utilizing your most well-liked bundle supervisor. The SDKs are fast-moving, so you’ll be able to simply replace dependencies as new options turn into accessible.
• Utilizing Lambda variations—When deploying sturdy features to manufacturing, use Lambda variations to make sure replay all the time occurs on the identical code model. Should you replace your operate code whereas an execution is suspended, replay will use the model that began the execution, stopping inconsistencies from code adjustments throughout long-running workflows.
• Testing your sturdy features—You may check sturdy features domestically with out AWS credentials utilizing the separate testing SDK with pytest integration and the AWS Serverless Utility Mannequin (AWS SAM) command line interface (CLI) for extra complicated integration testing.
• Open supply SDKs—The sturdy execution SDKs are open supply for JavaScript/TypeScript and Python. You may evaluation the supply code, contribute enhancements, and keep up to date with the newest options.
• Pricing—To be taught extra on AWS Lambda sturdy features pricing, confer with the AWS Lambda pricing web page.

Get began with AWS Lambda sturdy features by visiting the AWS Lambda console. To be taught extra, confer with AWS Lambda sturdy features documentation web page.

Glad constructing!

— Donnie