Guide

Note

Workflows is in public beta, and any developer with a free or paid Workers plan can start using Workflows immediately.

To learn more about Workflows and how it works, read the beta announcement blog ↗.

Workflows allow you to build durable, multi-step applications using the Workers platform. A Workflow can automatically retry, persist state, run for hours or days, and coordinate between third-party APIs.

You can build Workflows to post-process file uploads to R2 object storage, automate generation of Workers AI embeddings into a Vectorize vector database, or to trigger user lifecycle emails using your favorite email API.

This guide will instruct you through:

• Defining your first Workflow and publishing it
• Deploying the Workflow to your Cloudflare account
• Running (triggering) your Workflow and observing its output

At the end of this guide, you should be able to author, deploy and debug your own Workflows applications.

Prerequisites

1. Sign up for a Cloudflare account ↗.
2. Install Node.js ↗.

Node.js version manager

Use a Node version manager like Volta ↗ or nvm ↗ to avoid permission issues and change Node.js versions. Wrangler, discussed later in this guide, requires a Node version of 16.17.0 or later.

1. Define your Workflow

To create your first Workflow, use the create cloudflare (C3) CLI tool, specifying the Workflows starter template:

npm create cloudflare@latest workflows-starter -- --template "cloudflare/workflows-starter"

This will create a new folder called workflows-starter.

Open the src/index.ts file in your text editor. This file contains the following code, which is the most basic instance of a Workflow definition:

src/index.ts

import { WorkflowEntrypoint, WorkflowStep, WorkflowEvent } from 'cloudflare:workers';

type Env = {
  // Add your bindings here, e.g. Workers KV, D1, Workers AI, etc.
  MY_WORKFLOW: Workflow;
};

// User-defined params passed to your workflow
type Params = {
  email: string;
  metadata: Record<string, string>;
};

export class MyWorkflow extends WorkflowEntrypoint<Env, Params> {
  async run(event: WorkflowEvent<Params>, step: WorkflowStep) {
    // Can access bindings on `this.env`
    // Can access params on `event.payload`

    const files = await step.do('my first step', async () => {
      // Fetch a list of files from $SOME_SERVICE
      return {
        inputParams: event,
        files: [
          'doc_7392_rev3.pdf',
          'report_x29_final.pdf',
          'memo_2024_05_12.pdf',
          'file_089_update.pdf',
          'proj_alpha_v2.pdf',
          'data_analysis_q2.pdf',
          'notes_meeting_52.pdf',
          'summary_fy24_draft.pdf',
        ],
      };
    });

    const apiResponse = await step.do('some other step', async () => {
      let resp = await fetch('https://api.cloudflare.com/client/v4/ips');
      return await resp.json<any>();
    });

    await step.sleep('wait on something', '1 minute');

    await step.do(
      'make a call to write that could maybe, just might, fail',
      // Define a retry strategy
      {
        retries: {
          limit: 5,
          delay: '5 second',
          backoff: 'exponential',
        },
        timeout: '15 minutes',
      },
      async () => {
        // Do stuff here, with access to the state from our previous steps
        if (Math.random() > 0.5) {
          throw new Error('API call to $STORAGE_SYSTEM failed');
        }
      },
    );
  }
}

A Workflow definition:

1. Defines a run method that contains the primary logic for your workflow.
2. Has at least one or more calls to step.do that encapsulates the logic of your Workflow.
3. Allows steps to return (optional) state, allowing a Workflow to continue execution even if subsequent steps fail, without having to re-run all previous steps.

A single Worker application can contain multiple Workflow definitions, as long as each Workflow has a unique class name. This can be useful for code re-use or to define Workflows which are related to each other conceptually.

Each Workflow is otherwise entirely independent: a Worker that defines multiple Workflows is no different from a set of Workers that define one Workflow each.

2. Create your Workflows steps

Each step in a Workflow is an independently retriable function.

A step is what makes a Workflow powerful, as you can encapsulate errors and persist state as your Workflow progresses from step to step, avoiding your application from having to start from scratch on failure and ultimately build more reliable applications.

• A step can execute code (step.do) or sleep a Workflow (step.sleep).
• If a step fails (throws an exception), it will be automatically be retried based on your retry logic.
• If a step succeeds, any state it returns will be persisted within the Workflow.

At its most basic, a step looks like this:

src/index.ts

// Import the Workflow definition
import { WorkflowEntrypoint, WorkflowEvent, WorkflowStep } from "cloudflare:workers"

type Params = {}

// Create your own class that implements a Workflow
export class MyWorkflow extends WorkflowEntrypoint<Env, Params> {
    // Define a run() method
    async run(event: WorkflowEvent<Params>, step: WorkflowStep) {
        // Define one or more steps that optionally return state.
        let state = step.do("my first step", async () => {

        })

        step.do("my second step", async () => {

        })
    }
}

Each call to step.do accepts three arguments:

1. (Required) A step name, which identifies the step in logs and telemetry
2. (Required) A callback function that contains the code to run for your step, and any state you want the Workflow to persist
3. (Optional) A StepConfig that defines the retry configuration (max retries, delay, and backoff algorithm) for the step

When trying to decide whether to break code up into more than one step, a good rule of thumb is to ask "do I want all of this code to run again if just one part of it fails?". In many cases, you do not want to repeatedly call an API if the following data processing stage fails, or if you get an error when attempting to send a completion or welcome email.

For example, each of the below tasks is ideally encapsulated in its own step, so that any failure — such as a file not existing, a third-party API being down or rate limited — does not cause your entire program to fail.

• Reading or writing files from R2
• Running an AI task using Workers AI
• Querying a D1 database or a database via Hyperdrive
• Calling a third-party API

If a subsequent step fails, your Workflow can retry from that step, using any state returned from a previous step. This can also help you avoid unnecessarily querying a database or calling an paid API repeatedly for data you have already fetched.

Note

The term "Durable Execution" is widely used to describe this programming model.

"Durable" describes the ability of the program (application) to implicitly persist state without you having to manually write to an external store or serialize program state.

3. Configure your Workflow

Before you can deploy a Workflow, you need to configure it.

Open the wrangler.toml file at the root of your workflows-starter folder, which contains the following [[workflows]] configuration:

wrangler.toml

#:schema node_modules/wrangler/config-schema.json
name = "workflows-starter"
main = "src/index.ts"
compatibility_date = "2024-10-22"

[[workflows]]
# name of your workflow
name = "workflows-starter"
# binding name env.MY_WORKFLOW
binding = "MY_WORKFLOW"
# this is class that extends the Workflow class in src/index.ts
class_name = "MyWorkflow"

wrangler.json

{
  "name": "workflows-starter",
  "main": "src/index.ts",
  "compatibility_date": "2024-10-22",
  "workflows": [
    {
      "name": "workflows-starter",
      "binding": "MY_WORKFLOW",
      "class_name": "MyWorkflow"
    }
  ]
}

Note

If you have changed the name of the Workflow in your Wrangler commands, the JavaScript class name, or the name of the project you created, ensure that you update the values above to match the changes.

This configuration tells the Workers platform which JavaScript class represents your Workflow, and sets a binding name that allows you to run the Workflow from other handlers or to call into Workflows from other Workers scripts.

4. Bind to your Workflow

We have a very basic Workflow definition, but now need to provide a way to call it from within our code. A Workflow can be triggered by:

1. External HTTP requests via a fetch() handler
2. Messages from a Queue
3. A schedule via Cron Trigger
4. Via the Workflows REST API or wrangler CLI

Return to the src/index.ts file we created in the previous step and add a fetch handler that binds to our Workflow. This binding allows us to create new Workflow instances, fetch the status of an existing Workflow, pause and/or terminate a Workflow.

src/index.ts

// This is in the same file as your Workflow definition

export default {
  async fetch(req: Request, env: Env): Promise<Response> {
    let url = new URL(req.url);

    if (url.pathname.startsWith('/favicon')) {
      return Response.json({}, { status: 404 });
    }

    // Get the status of an existing instance, if provided
    let id = url.searchParams.get('instanceId');
    if (id) {
      let instance = await env.MY_WORKFLOW.get(id);
      return Response.json({
        status: await instance.status(),
      });
    }

    // Spawn a new instance and return the ID and status
    let instance = await env.MY_WORKFLOW.create();
    return Response.json({
      id: instance.id,
      details: await instance.status(),
    });
  },
};

The code here exposes a HTTP endpoint that generates a random ID and runs the Workflow, returning the ID and the Workflow status. It also accepts an optional instanceId query parameter that retrieves the status of a Workflow instance by its ID.

Note

In a production application, you might choose to put authentication in front of your endpoint so that only authorized users can run a Workflow. Alternatively, you could pass messages to a Workflow from a Queue consumer in order to allow for long-running tasks.

Review your Workflow code

Note

This is the full contents of the src/index.ts file pulled down when you used the cloudflare/workflows-starter template at the beginning of this guide.

Before you deploy, you can review the full Workflows code and the fetch handler that will allow you to trigger your Workflow over HTTP:

src/index.ts

import { WorkflowEntrypoint, WorkflowStep, WorkflowEvent } from 'cloudflare:workers';

type Env = {
  // Add your bindings here, e.g. Workers KV, D1, Workers AI, etc.
  MY_WORKFLOW: Workflow;
};

// User-defined params passed to your workflow
type Params = {
  email: string;
  metadata: Record<string, string>;
};

export class MyWorkflow extends WorkflowEntrypoint<Env, Params> {
  async run(event: WorkflowEvent<Params>, step: WorkflowStep) {
    // Can access bindings on `this.env`
    // Can access params on `event.params`

    const files = await step.do('my first step', async () => {
      // Fetch a list of files from $SOME_SERVICE
      return {
        inputParams: event,
        files: [
          'doc_7392_rev3.pdf',
          'report_x29_final.pdf',
          'memo_2024_05_12.pdf',
          'file_089_update.pdf',
          'proj_alpha_v2.pdf',
          'data_analysis_q2.pdf',
          'notes_meeting_52.pdf',
          'summary_fy24_draft.pdf',
        ],
      };
    });

    const apiResponse = await step.do('some other step', async () => {
      let resp = await fetch('https://api.cloudflare.com/client/v4/ips');
      return await resp.json<any>();
    });

    await step.sleep('wait on something', '1 minute');

    await step.do(
      'make a call to write that could maybe, just might, fail',
      // Define a retry strategy
      {
        retries: {
          limit: 5,
          delay: '5 second',
          backoff: 'exponential',
        },
        timeout: '15 minutes',
      },
      async () => {
        // Do stuff here, with access to the state from our previous steps
        if (Math.random() > 0.5) {
          throw new Error('API call to $STORAGE_SYSTEM failed');
        }
      },
    );
  }
}

export default {
  async fetch(req: Request, env: Env): Promise<Response> {
    let url = new URL(req.url);

    if (url.pathname.startsWith('/favicon')) {
      return Response.json({}, { status: 404 });
    }

    // Get the status of an existing instance, if provided
    let id = url.searchParams.get('instanceId');
    if (id) {
      let instance = await env.MY_WORKFLOW.get(id);
      return Response.json({
        status: await instance.status(),
      });
    }

    // Spawn a new instance and return the ID and status
    let instance = await env.MY_WORKFLOW.create();
    return Response.json({
      id: instance.id,
      details: await instance.status(),
    });
  },
};

5. Deploy your Workflow

Deploying a Workflow is identical to deploying a Worker.

npx wrangler deploy

# Note the "Workflows" binding mentioned here, showing that
# wrangler has detected your Workflow
Your worker has access to the following bindings:
- Workflows:
  - MY_WORKFLOW: MyWorkflow (defined in workflows-starter)
Uploaded workflows-starter (2.53 sec)
Deployed workflows-starter triggers (1.12 sec)
  https://workflows-starter.YOUR_WORKERS_SUBDOMAIN.workers.dev
  workflow: workflows-starter

A Worker with a valid Workflow definition will be automatically registered by Workflows. You can list your current Workflows using Wrangler:

npx wrangler workflows list

Showing last 1 workflow:
┌───────────────────┬───────────────────┬────────────┬─────────────────────────┬─────────────────────────┐
│ Name              │ Script name       │ Class name │ Created                 │ Modified                │
├───────────────────┼───────────────────┼────────────┼─────────────────────────┼─────────────────────────┤
│ workflows-starter │ workflows-starter │ MyWorkflow │ 10/23/2024, 11:33:58 AM │ 10/23/2024, 11:33:58 AM │
└───────────────────┴───────────────────┴────────────┴─────────────────────────┴─────────────────────────┘

6. Run and observe your Workflow

With your Workflow deployed, you can now run it.

1. A Workflow can run in parallel: each unique invocation of a Workflow is an instance of that Workflow.
2. An instance will run to completion (success or failure).
3. Deploying newer versions of a Workflow will cause all instances after that point to run the newest Workflow code.

Note

Because Workflows can be long running, it is possible to have running instances that represent different versions of your Workflow code over time.

To trigger our Workflow, we will use the wrangler CLI and pass in an optional --payload. The payload will be passed to your Workflow's run method handler as an Event.

npx wrangler workflows trigger workflows-starter '{"hello":"world"}'

# Workflow instance "12dc179f-9f77-4a37-b973-709dca4189ba" has been queued successfully

To inspect the current status of the Workflow instance we just triggered, we can either reference it by ID or by using the keyword latest:

npx wrangler@latest workflows instances describe workflows-starter latest
# Or by ID:
# npx wrangler@latest workflows instances describe workflows-starter 12dc179f-9f77-4a37-b973-709dca4189ba

Workflow Name:         workflows-starter
Instance Id:           f72c1648-dfa3-45ea-be66-b43d11d216f8
Version Id:            cedc33a0-11fa-4c26-8a8e-7d28d381a291
Status:                ✅ Completed
Trigger:               🌎 API
Queued:                10/15/2024, 1:55:31 PM
Success:               ✅ Yes
Start:                 10/15/2024, 1:55:31 PM
End:                   10/15/2024, 1:56:32 PM
Duration:              1 minute
Last Successful Step:  make a call to write that could maybe, just might, fail-1
Steps:

  Name:      my first step-1
  Type:      🎯 Step
  Start:     10/15/2024, 1:55:31 PM
  End:       10/15/2024, 1:55:31 PM
  Duration:  0 seconds
  Success:   ✅ Yes
  Output:    "{\"inputParams\":[{\"timestamp\":\"2024-10-15T13:55:29.363Z\",\"payload\":{\"hello\":\"world\"}}],\"files\":[\"doc_7392_rev3.pdf\",\"report_x29_final.pdf\",\"memo_2024_05_12.pdf\",\"file_089_update.pdf\",\"proj_alpha_v2.pdf\",\"data_analysis_q2.pdf\",\"notes_meeting_52.pdf\",\"summary_fy24_draft.pdf\",\"plan_2025_outline.pdf\"]}"
┌────────────────────────┬────────────────────────┬───────────┬────────────┐
│ Start                  │ End                    │ Duration  │ State      │
├────────────────────────┼────────────────────────┼───────────┼────────────┤
│ 10/15/2024, 1:55:31 PM │ 10/15/2024, 1:55:31 PM │ 0 seconds │ ✅ Success │
└────────────────────────┴────────────────────────┴───────────┴────────────┘

  Name:      some other step-1
  Type:      🎯 Step
  Start:     10/15/2024, 1:55:31 PM
  End:       10/15/2024, 1:55:31 PM
  Duration:  0 seconds
  Success:   ✅ Yes
  Output:    "{\"result\":{\"ipv4_cidrs\":[\"173.245.48.0/20\",\"103.21.244.0/22\",\"103.22.200.0/22\",\"103.31.4.0/22\",\"141.101.64.0/18\",\"108.162.192.0/18\",\"190.93.240.0/20\",\"188.114.96.0/20\",\"197.234.240.0/22\",\"198.41.128.0/17\",\"162.158.0.0/15\",\"104.16.0.0/13\",\"104.24.0.0/14\",\"172.64.0.0/13\",\"131.0.72.0/22\"],\"ipv6_cidrs\":[\"2400:cb00::/32\",\"2606:4700::/32\",\"2803:f800::/32\",\"2405:b500::/32\",\"2405:8100::/32\",\"2a06:98c0::/29\",\"2c0f:f248::/32\"],\"etag\":\"38f79d050aa027e3be3865e495dcc9bc\"},\"success\":true,\"errors\":[],\"messages\":[]}"
┌────────────────────────┬────────────────────────┬───────────┬────────────┐
│ Start                  │ End                    │ Duration  │ State      │
├────────────────────────┼────────────────────────┼───────────┼────────────┤
│ 10/15/2024, 1:55:31 PM │ 10/15/2024, 1:55:31 PM │ 0 seconds │ ✅ Success │
└────────────────────────┴────────────────────────┴───────────┴────────────┘

  Name:      wait on something-1
  Type:      💤 Sleeping
  Start:     10/15/2024, 1:55:31 PM
  End:       10/15/2024, 1:56:31 PM
  Duration:  1 minute

  Name:      make a call to write that could maybe, just might, fail-1
  Type:      🎯 Step
  Start:     10/15/2024, 1:56:31 PM
  End:       10/15/2024, 1:56:32 PM
  Duration:  1 second
  Success:   ✅ Yes
  Output:    null
┌────────────────────────┬────────────────────────┬───────────┬────────────┬───────────────────────────────────────────┐
│ Start                  │ End                    │ Duration  │ State      │ Error                                     │
├────────────────────────┼────────────────────────┼───────────┼────────────┼───────────────────────────────────────────┤
│ 10/15/2024, 1:56:31 PM │ 10/15/2024, 1:56:31 PM │ 0 seconds │ ❌ Error   │ Error: API call to $STORAGE_SYSTEM failed │
├────────────────────────┼────────────────────────┼───────────┼────────────┼───────────────────────────────────────────┤
│ 10/15/2024, 1:56:32 PM │ 10/15/2024, 1:56:32 PM │ 0 seconds │ ✅ Success │                                           │
└────────────────────────┴────────────────────────┴───────────┴────────────┴───────────────────────────────────────────┘

From the output above, we can inspect:

• The status (success, failure, running) of each step
• Any state emitted by the step
• Any sleep state, including when the Workflow will wake up
• Retries associated with each step
• Errors, including exception messages

Note

You do not have to wait for a Workflow instance to finish executing to inspect its current status. The wrangler workflows instances describe sub-command will show the status of an in-progress instance, including any persisted state, if it is sleeping, and any errors or retries. This can be especially useful when debugging a Workflow during development.

In the previous step, we also bound a Workers script to our Workflow. You can trigger a Workflow by visiting the (deployed) Workers script in a browser or with any HTTP client.

# This must match the URL provided in step 6
curl -s https://workflows-starter.YOUR_WORKERS_SUBDOMAIN.workers.dev/

{"id":"16ac31e5-db9d-48ae-a58f-95b95422d0fa","details":{"status":"queued","error":null,"output":null}}

---

Next steps

• Learn more about how events are passed to a Workflow.
• Learn more about binding to and triggering Workflow instances using the Workers API.
• Learn more about the Rules of Workflows and best practices for building applications using Workflows.

If you have any feature requests or notice any bugs, share your feedback directly with the Cloudflare team by joining the Cloudflare Developers community on Discord ↗.

Was this helpful?

What did you like?

Accurate

Easy to understand

Solved my problem

Helped me decide to use the product

Other

What went wrong?

Hard to understand

Incorrect information

Missing the information

Other

Thank you for helping improve Cloudflare's documentation!