AWS Lambda Durable Functions
AWSOfficialPublished by AWS
- Skill
- Tool integration
- Cloud and DevOps
- Apache-2.0
- Low risk
AutoGineer summary
Written by AutoGineer from the skill’s own files, not by the author
This skill provides guidance and patterns for implementing durable, multi-step applications on AWS Lambda. It focuses on the replay model, which allows functions to maintain state across long-running executions—up to one year—by automatically checkpointing progress. The skill covers essential architectural patterns such as the saga pattern for error handling, wait/callback mechanisms for external system integration, and testing strategies using the local test runner. It emphasizes strict adherence to deterministic coding practices required by the replay model to ensure reliable execution.
What it does
- Enables state persistence for long-running Lambda functions.
- Provides patterns for implementing the saga pattern and compensating transactions.
- Supports wait/callback patterns for integrating with external systems.
- Offers guidance on testing durable functions locally and in the cloud.
- Defines rules for deterministic code to prevent state corruption during replays.
- Outlines requirements for qualified ARN invocation and IAM permissions.
When to use it
- When building workflows that require execution times longer than standard Lambda timeouts.
- When orchestrating multi-step processes that need to maintain state across interruptions.
- When implementing reliable, fault-tolerant serverless integrations.
- When managing complex state machines or human-in-the-loop approval processes.
- When you need automatic retry logic and checkpointing for critical business processes.
Advantages
- Resilient execution that survives infrastructure interruptions.
- Simplified state management without external databases for simple workflows.
- Native support for complex orchestration patterns like sagas and parallel execution.
- Built-in replay-aware logging and error handling.
- Local testing capabilities to validate complex workflow logic.
Usage examples
- “How do I implement a multi-step workflow with a wait period in AWS Lambda?”
- The agent provides a code example using context.step for logic and context.wait for the delay, ensuring the execution state is persisted correctly.
- “Why is my durable function failing during replay?”
- The agent helps diagnose non-deterministic code (like Date.now() or Math.random()) placed outside of steps, which causes state corruption.
- “How do I handle errors in a long-running durable function?”
- The agent explains the saga pattern and how to implement compensating transactions within steps to roll back state if a step fails.
- “What are the requirements for invoking a durable Lambda function?”
- The agent clarifies that durable functions must be invoked using a qualified ARN (version, alias, or $LATEST) and provides the corresponding AWS CLI command.
Install
Download the skill as a ZIP:
https://mp-staging.autogineer.com/agent-skills/aws/aws-lambda-durable-functions/downloadUnzip it into the cross-client skills folder of your project:
mkdir -p .agents/skills && unzip aws-lambda-durable-functions.zip -d .agents/skills
Documentation by AWS
AWS Lambda durable functions
Build resilient multi-step applications and AI workflows that can execute for up to 1 year while maintaining reliable progress despite interruptions.
Works best with the AWS MCP server but is not required. All AWS interactions in this skill use standard AWS CLI commands that work in any environment with configured AWS credentials.
Critical Rules
Read these before writing any code. Each one is a constraint that will silently break a function if violated.
Durable execution must be enabled at function creation time — it cannot be retrofitted. A new Lambda function must be created with durable execution turned on. Migrate the logic into the new function; do not attempt to install the SDK and wrap the handler of the existing function and expect it to work.
Durable functions must be invoked with a qualified ARN — a specific version, an alias, or the literal
$LATESTsuffix. An unqualified function name will fail. See the Invocation Requirements section below for examples.Durable operations cannot be nested. You cannot call
context.step(),context.wait(), orcontext.invoke()from inside another step's callback. Usecontext.runInChildContext()to group operations instead.All non-deterministic code must run inside steps.
Date.now(),Math.random(), UUID generation, API calls, and database queries outside a step will produce different values on replay and corrupt execution state.Closure mutations are lost on replay - return values from steps
Side effects outside steps repeat - use
context.logger(replay-aware)
When to Load Reference Files
Load the appropriate reference file based on what the user is working on:
Getting started, basic setup, example, ESLint, or Jest setup -> see getting-started.md
Understanding replay model, determinism, or non-deterministic errors -> see replay-model-rules.md
Creating steps, atomic operations, or retry logic -> see step-operations.md
Waiting, delays, callbacks, external systems, or polling -> see wait-operations.md
Parallel execution, map operations, batch processing, or concurrency -> see concurrent-operations.md
Error handling, retry strategies, saga pattern, or compensating transactions -> see error-handling.md
Advanced error handling, timeout handling, circuit breakers, or conditional retries -> see advanced-error-handling.md
Testing, local testing, cloud testing, test runner, or flaky tests -> see testing-patterns.md
Deployment, CloudFormation, CDK, SAM, log groups, deploy, or infrastructure -> see deployment-iac.md
Advanced patterns, GenAI agents, completion policies, step semantics, or custom serialization -> see advanced-patterns.md
troubleshooting, stuck execution, failed execution, debug execution ID, execution history, execution error, why did my execution fail, execution timed out, callback not received, diagnose execution, or root cause execution -> see troubleshooting-executions.md
Quick Reference
Basic Handler Pattern
TypeScript:
import { withDurableExecution, DurableContext } from '@aws/durable-execution-sdk-js';
export const handler = withDurableExecution(async (event, context: DurableContext) => {
const result = await context.step('process', async () => processData(event));
return result;
});Python:
from aws_durable_execution_sdk_python import durable_execution, DurableContext
@durable_execution
def handler(event: dict, context: DurableContext) -> dict:
result = context.step(lambda _: process_data(event), name='process')
return resultPython API Differences
The Python SDK differs from TypeScript in several key areas:
Steps: Use
@durable_stepdecorator +context.step(my_step(args)), or inlinecontext.step(lambda _: ..., name='...'). Prefer the decorator for automatic step naming.Wait:
context.wait(duration=Duration.from_seconds(n), name='...')Exceptions:
ExecutionError(permanent),InvocationError(transient),CallbackError(callback failures)Testing: Use
DurableFunctionTestRunnerclass directly - instantiate with handler, use context manager, callrun(input=...)
Invocation Requirements
Durable functions require qualified ARNs (version, alias, or $LATEST):
# Valid
aws lambda invoke --function-name my-function:1 output.json
aws lambda invoke --function-name my-function:live output.json
# Invalid - will fail
aws lambda invoke --function-name my-function output.jsonIAM Permissions
Your Lambda execution role MUST have the AWSLambdaBasicDurableExecutionRolePolicy managed policy attached. This includes:
lambda:CheckpointDurableExecution- Persist execution statelambda:GetDurableExecutionState- Retrieve execution stateCloudWatch Logs permissions
Additional permissions needed for:
Durable invokes:
lambda:InvokeFunctionon target function ARNsExternal callbacks: Systems need
lambda:SendDurableExecutionCallbackSuccessandlambda:SendDurableExecutionCallbackFailure
Validation Guidelines
When writing or reviewing durable function code, ALWAYS check for these replay model violations:
Non-deterministic code outside steps:
Date.now(),Math.random(), UUID generation, API calls, database queries must all be inside stepsNested durable operations in step functions: Cannot call
context.step(),context.wait(), orcontext.invoke()inside a step function — usecontext.runInChildContext()insteadClosure mutations that won't persist: Variables mutated inside steps are NOT preserved across replays — return values from steps instead
Side effects outside steps that repeat on replay: Use
context.loggerfor logging (it is replay-aware and deduplicates automatically)
When implementing or modifying tests for durable functions, ALWAYS verify:
All operations have descriptive names
Tests get operations by NAME, never by index
Replay behavior is tested with multiple invocations
Use
LocalDurableTestRunnerfor local testing
Security Considerations
Checkpoint data encryption: Execution state is persisted automatically. Enable KMS encryption on associated CloudWatch Log Groups to protect checkpointed data at rest.
Sensitive data in step results: Step return values are checkpointed and persisted. Do not return secrets, raw credentials, or PII from steps — store sensitive data in Secrets Manager or SSM Parameter Store and return references instead.
Input validation: Validate and sanitize event payloads at the handler entry point before passing data to steps.
Credential management: Retrieve secrets from AWS Secrets Manager or SSM Parameter Store within steps.
Callback payload validation: Data received via
waitForCallbackoriginates from external systems — validate and sanitize before processing.Logging: Avoid
DEBUGlog level in non-development environments as it may expose step results and execution state. Enable CloudWatch Logs encryption with KMS.
Resources
Files
- ATTRIBUTION.mdAdded by AutoGineer946 B
- LICENSELicence file added by AutoGineer9.9 KiB
- NOTICELicence file added by AutoGineer67 B
- SKILL.md8.9 KiB
- references/advanced-error-handling.md4.9 KiB
- references/advanced-patterns.md10.4 KiB
- references/concurrent-operations.md8.4 KiB
- references/deployment-iac.md14.0 KiB
- references/error-handling.md10.5 KiB
- references/getting-started.md12.6 KiB
- references/replay-model-rules.md8.3 KiB
- references/step-operations.md8.3 KiB
- references/testing-patterns.md17.4 KiB
- references/troubleshooting-executions.md8.5 KiB
- references/wait-operations.md9.5 KiB
references/error-handling.md
Error Handling and Retry Strategies
Comprehensive error handling patterns for durable functions.
TypeScript:
import { createRetryStrategy, JitterStrategy } from '@aws/durable-execution-sdk-js';
// Exponential backoff with jitter
const result = await context.step(
'api-call',
async () => callAPI(),
{
retryStrategy: createRetryStrategy({
maxAttempts: 5,
initialDelay: { seconds: 1 },
maxDelay: { seconds: 60 },
backoffRate: 2.0,
jitter: JitterStrategy.FULL
})
}
);
// Fixed delay
const result = await context.step(
'simple-retry',
async () => operation(),
{
retryStrategy: createRetryStrategy({
maxAttempts: 3,
delay: { seconds: 5 },
backoffRate: 1
})
}
);Python:
from aws_durable_execution_sdk_python.retries import RetryStrategyConfig, create_retry_strategy, JitterStrategy
retry_config = RetryStrategyConfig(
max_attempts=5,
initial_delay=Duration.from_seconds(1),
max_delay=Duration.from_seconds(60),
backoff_rate=2.0,
jitter_strategy=JitterStrategy.FULL
)
result = context.step(
func=api_call(),
config=StepConfig(retry_strategy=create_retry_strategy(retry_config))
)Custom Retry Logic
TypeScript:
const result = await context.step(
'custom-retry',
async () => riskyOperation(),
{
retryStrategy: (error, attemptCount) => {
// Don't retry client errors
if (error.statusCode >= 400 && error.statusCode < 500) {
return { shouldRetry: false };
}
// Retry server errors with exponential backoff
if (attemptCount < 5) {
return {
shouldRetry: true,
delay: { seconds: Math.pow(2, attemptCount) }
};
}
return { shouldRetry: false };
}
}
);Python:
def custom_retry(error: Exception, attempt: int) -> RetryDecision:
if hasattr(error, 'status_code') and 400 <= error.status_code < 500:
return RetryDecision.no_retry()
if attempt < 5:
return RetryDecision(
should_retry=True,
delay=Duration.from_seconds(2 ** attempt)
)
return RetryDecision.no_retry()Error Classification
Retryable vs Non-Retryable
TypeScript:
class ValidationError extends Error {
name = 'ValidationError';
}
class NetworkError extends Error {
name = 'NetworkError';
}
const result = await context.step(
'selective-retry',
async () => operation(),
{
retryStrategy: createRetryStrategy({
maxAttempts: 3,
retryableErrorTypes: [NetworkError],
// ValidationError won't be retried
})
}
);Python:
retry_config = RetryStrategyConfig(
max_attempts=3,
retryable_error_types=[NetworkError, TimeoutError]
)Saga Pattern
Implement compensating transactions for distributed workflows:
TypeScript:
export const handler = withDurableExecution(async (event, context: DurableContext) => {
const compensations: Array<{
name: string;
fn: () => Promise<void>;
}> = [];
try {
// Step 1: Reserve inventory
const reservation = await context.step('reserve-inventory', async () =>
inventoryService.reserve(event.items)
);
compensations.push({
name: 'cancel-reservation',
fn: () => inventoryService.cancelReservation(reservation.id)
});
// Step 2: Charge payment
const payment = await context.step('charge-payment', async () =>
paymentService.charge(event.paymentMethod, event.amount)
);
compensations.push({
name: 'refund-payment',
fn: () => paymentService.refund(payment.id)
});
// Step 3: Create shipment
const shipment = await context.step('create-shipment', async () =>
shippingService.createShipment(event.address, event.items)
);
compensations.push({
name: 'cancel-shipment',
fn: () => shippingService.cancelShipment(shipment.id)
});
return { success: true, orderId: shipment.orderId };
} catch (error) {
context.logger.error('Order failed, executing compensations', error);
// Execute compensations in reverse order
for (const comp of compensations.reverse()) {
try {
await context.step(comp.name, async () => comp.fn());
} catch (compError) {
context.logger.error(`Compensation ${comp.name} failed`, compError);
// Continue with other compensations
}
}
throw error;
}
});Python:
# Note: All service methods are decorated with @durable_step
@durable_execution
def handler(event: dict, context: DurableContext) -> dict:
compensations = []
try:
# Step 1: Reserve inventory
reservation = context.step(reserve_inventory(event['items']))
compensations.append(('cancel-reservation', cancel_reservation, reservation['id']))
# Step 2: Charge payment
payment = context.step(charge_payment(event['payment_method'], event['amount']))
compensations.append(('refund-payment', refund_payment, payment['id']))
# Step 3: Create shipment
shipment = context.step(create_shipment(event['address'], event['items']))
return {'success': True, 'order_id': shipment['order_id']}
except Exception as error:
context.logger.error(f'Order failed, executing compensations: {error}')
for name, comp_step, resource_id in reversed(compensations):
try:
context.step(comp_step(resource_id))
except Exception as comp_error:
context.logger.error(f'Compensation {name} failed: {comp_error}')
raise errorUnrecoverable Errors
Mark errors as unrecoverable to stop execution immediately:
TypeScript:
export const handler = withDurableExecution(async (event, context: DurableContext) => {
const user = await context.step(
'fetch-user',
async () => {
const user = await fetchUser(event.userId);
if (!user) {
throw new Error('User not found');
}
return user;
},
{ retryStrategy: () => ({ shouldRetry: false }) }
);
// Continue processing...
});Python:
from aws_durable_execution_sdk_python.exceptions import ExecutionError
@durable_execution
def handler(event: dict, context: DurableContext) -> dict:
@durable_step
def fetch_user_step(step_ctx: StepContext):
user = fetch_user(event['user_id'])
if not user:
# Stop execution immediately — permanent failure, no retry
raise ExecutionError('User not found')
return user
user = context.step(fetch_user_step())
# Continue processing...The SDK provides these exception types for different failure scenarios:
| Exception | Retryable | Use case |
|---|---|---|
ExecutionError | No | Permanent business logic failures (returns FAILED status) |
InvocationError | Yes (by Lambda) | Transient infrastructure issues (Lambda retries invocation) |
CallbackError | No | Callback handling failures |
DurableExecutionsError | — | Base class for all SDK exceptions |
Error Determinism
Ensure errors are deterministic across replays:
TypeScript:
class CustomBusinessError extends Error {
constructor(
message: string,
public readonly code: string,
public readonly details: any
) {
super(message);
this.name = 'CustomBusinessError';
}
}
const result = await context.step('validate', async () => {
if (!isValid(data)) {
// ✅ Deterministic error
throw new CustomBusinessError(
'Validation failed',
'INVALID_DATA',
{ field: 'email', reason: 'invalid format' }
);
}
return processData(data);
});Circuit Breaker Pattern
TypeScript:
class CircuitBreaker {
private failures = 0;
private lastFailureTime = 0;
private readonly threshold = 5;
private readonly timeout = 60000; // 1 minute
async execute<T>(fn: () => Promise<T>): Promise<T> {
if (this.isOpen()) {
throw new Error('Circuit breaker is open');
}
try {
const result = await fn();
this.onSuccess();
return result;
} catch (error) {
this.onFailure();
throw error;
}
}
private isOpen(): boolean {
if (this.failures >= this.threshold) {
const elapsed = Date.now() - this.lastFailureTime;
return elapsed < this.timeout;
}
return false;
}
private onSuccess() {
this.failures = 0;
}
private onFailure() {
this.failures++;
this.lastFailureTime = Date.now();
}
}
// Use in handler
const breaker = new CircuitBreaker();
export const handler = withDurableExecution(async (event, context: DurableContext) => {
const result = await context.step('api-call', async () => {
return await breaker.execute(() => callExternalAPI());
});
return result;
});Partial Failure Handling
TypeScript:
export const handler = withDurableExecution(async (event, context: DurableContext) => {
const results = await context.map(
'process-items',
event.items,
async (ctx, item, index) => {
return await ctx.step(async () => processItem(item));
},
{
completionConfig: {
toleratedFailurePercentage: 10 // Allow 10% failures
}
}
);
if (results.hasFailure) {
// Log failures but continue
context.logger.warn('Some items failed', {
failureCount: results.failureCount,
failures: results.failed().map(f => ({
index: f.index,
error: f.error?.message
}))
});
// Store failed items for later retry
await context.step('store-failures', async () => {
const failedItems = results.failed().map(f => event.items[f.index]);
return await storeFailedItems(failedItems);
});
}
return {
totalProcessed: results.successCount,
failed: results.failureCount
};
});Best Practices
Use appropriate retry strategies - exponential backoff for most cases
Classify errors correctly - distinguish retryable from non-retryable
Implement compensating transactions for distributed workflows
Make errors deterministic - same input produces same error
Use unrecoverable errors to stop execution early when appropriate
Log errors with context using
context.loggerHandle partial failures gracefully in batch operations
Implement circuit breakers for external service calls
Test error scenarios thoroughly with test runners
Monitor error rates and adjust retry strategies accordingly
Requirements and compatibility
Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta
“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.
Needs: Needs an MCP server, Needs a command-line tool, Needs Python
Security
What our scanner noticed· Low risk
- Instructions that use the network
A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.
Attribution and licence
- Published by
- AWS
- Publisher
- AWS (Official)
- Source
- aws/agent-toolkit-for-aws · skills/specialized-skills/serverless-skills/aws-lambda-durable-functions · commit 0a218f8
- Licence
- Apache License 2.0 (redistribution allowed)
- LICENSE (added by AutoGineer from LICENSE)
- NOTICE (added by AutoGineer from NOTICE)
- ATTRIBUTION.md (written by AutoGineer)
- Synced
AutoGineer is not affiliated with or endorsed by AWS.
Download ZIP
The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download aws-lambda-durable-functions.zip