The manager — Step Functions and sagas

Some work is too important to leave to a crowd: taking a payment, cooking, and refunding if the kitchen runs out of ingredients. Here a manager helps — someone who knows the whole plan, tells each human what to do, waits for results and decides what happens when something goes wrong. AWS Step Functions is that manager.

What a manager does

• Knows the steps and their order.
• Waits — seconds, hours or days (Standard workflows can run for up to a year).
• Retries failed steps with backoff.
• Branches on results ("if payment fails, notify the customer").
• Runs steps in parallel.
• Records every step, so you can see exactly where a process is — or where it failed.

The order lifecycle as a workflow

Start
  → Cashier: authorise payment ──(fails)──► Messenger: "payment failed" → End
  → Chef: cook order ──(out of ingredients)──► Cashier: refund (apology) → Messenger: "sorry" → End
  → Wait for "order ready"
  → Messenger: "your food is ready!"
End

Sagas: apologising instead of rolling back

In one database, a transaction can roll back everything. Across many humans, each with their own memory, you can't. Instead you use a saga: if a later step fails, run compensating actions for the earlier ones — like a refund after a payment. It's how people handle mistakes: you can't undo, but you can apologise and make it right.

The manager in CDK

import * as sfn from "aws-cdk-lib/aws-stepfunctions";
import * as tasks from "aws-cdk-lib/aws-stepfunctions-tasks";

const authorise = new tasks.LambdaInvoke(this, "Cashier: authorise payment", {
  lambdaFunction: cashierAuthoriseFn,
  outputPath: "$.Payload",
});
authorise.addRetry({ errors: ["States.TaskFailed"], maxAttempts: 2, backoffRate: 2 });

const cook = new tasks.LambdaInvoke(this, "Chef: cook order", {
  lambdaFunction: chefCookFn,
  outputPath: "$.Payload",
});

const refund = new tasks.LambdaInvoke(this, "Cashier: refund (apology)", {
  lambdaFunction: cashierRefundFn,
  outputPath: "$.Payload",
});

const notifyReady = new tasks.LambdaInvoke(this, "Messenger: food is ready", { lambdaFunction: messengerFn });
const notifySorry = new tasks.LambdaInvoke(this, "Messenger: sorry", { lambdaFunction: messengerFn });

cook.addCatch(refund.next(notifySorry), { errors: ["OutOfIngredients"] });

const definition = authorise.next(cook).next(notifyReady);

new sfn.StateMachine(this, "OrderManager", {
  definitionBody: sfn.DefinitionBody.fromChainable(definition),
  tracingEnabled: true,
});

Step Functions can also call many AWS services directly (DynamoDB, SQS, SNS, EventBridge and more) without a Lambda in between — fewer cells to maintain.

Standard vs Express

When to hire a manager

Written by Bipin Singh

Senior Full-Stack Engineer · AI & AWS. I design and build serverless systems on AWS — and love explaining them simply.

Work with me