Stay organized with collections Save and categorize content based on your preferences.
Edit on GitHub
Report issue
Page history

Asynchronous patterns for Cloud Functions

Author(s): @ptone ,   Published: 2019-03-08

Preston Holmes | Solution Architect | Google

Contributed by Google employees.

This tutorial demonstrates how to use Cloud Functions to extend the synchronous web-hook-style request/response to longer-running jobs, with a focus on trackable and stateful long-running operations.

When a caller makes a request of a service, the caller is asking the service to perform some work. There are three integration patterns that can be applied, depending on the use-case:

Synchronous request/response


The caller will wait until the work is done, expecting a result. This pattern can be directly and simply handled by Cloud Functions with an HTTP trigger.

Asynchronous work queue


The caller does not need to wait for the work to be done, and does not need to follow up on the completion status.

There are a couple ways this can be solved on Google Cloud in a serverless way. You can use Pub/Sub patterns for long-running tasks or you can use a dedicated service with Cloud Tasks.

Asynchronous stateful jobs

The caller does not need to wait for the work to be done, but it does need the ability to inquire about the completion of the request.

This pattern is common enough that Google has defined a standard of a long-running-operation API contract used in multiple APIs.

This is a high-level pattern. What about a job state is tracked and how work is performed will vary by use-case. This tutorial goes deeper into this pattern.

Serverless stateful jobs

This tutorial uses several managed services to implement the asynchronous stateful jobs pattern, including Pub/Sub and Firestore.

You can implement this pattern with alternative components. You can use Cloud SQL for job state storage or Cloud Tasks for the work queue. This tutorial uses the support of Go in Cloud Functions, but any supported language can be used to fulfill the pattern.

Job resource definition

A simple custom job definition is defined as follows:

type JobState int

const (
    Created JobState = iota

type Job struct {
    ID         string    `json:"id"`
    CreateTime time.Time `json:"created-time"`
    DoneTime   time.Time `json:"done-time" firestore:"DoneTime,omitempty"`
    Done       bool
    Result     string
    State      JobState
    Task       map[string]interface{} `json:"-"`
    // can add a requester, source IP etc if needed

When a job request is received, it is given an ID, and the details of the work are included in a task field. This task is sent into the work queue as a Pub/Sub payload, with the job ID as a Cloud Pub/Sub attribute. When a worker picks up the task, it moves the job from Created to Running. When the task is complete, it moves from Running to Completed (success) or Failed and writes any result back into the Cloud Firestore document.


  1. Create a project in the Cloud Console.
  2. Enable billing for your project.
  3. Use Cloud Shell or install the Cloud SDK.
  4. Enable Cloud Functions, Firestore, and Pub/Sub APIs:

    gcloud services enable
  5. Install jq, curl, and watch. These tools are already in Cloud Shell, but install them if you are running this tutorial from your local machine.

Commands in this tutorial assume that you are running from the tutorial folder:

git clone
cd community/tutorials/cloud-functions-async

Create Cloud Pub/Sub resources

gcloud pubsub topics create jobs
gcloud pubsub subscriptions create worker-tasks --topic jobs

Create the Cloud Firestore database

Cloud Firestore is used to store the state of jobs. Follow the server quickstart documentation to create a Cloud Firestore database.

Deploy Cloud Functions

gcloud functions deploy Jobs --runtime go111 --trigger-http

Create jobs and check their state

After the function is deployed, we can use POST and GET methods to interact with long-running jobs.

Copy the URL

Put the URL of the deployed Cloud Function into an environment variable:

export URL=$(gcloud functions describe Jobs --format='value(httpsTrigger.url)')

Create a job

Call the function-based API to create a job with a simple task. Our task has one field, worktime, which determines how many seconds the task takes to complete:

JOBID=$(curl -s --header "Content-Type: application/json" --request POST --data '{"worktime":40}' $URL | jq -r '.id')

When the job is created, the function returns the ID of the job, which is put into an environment variable for reference later.

Check on job status

This command polls the state of the job every 2 seconds with the watch command.

watch -t "curl -s $URL/$JOBID | jq"

Perform the work

Start a worker in another tab in either your local terminal or in Cloud Shell, and then run a basic worker that will complete after waiting the worktime value in the original.

cd worker
go run main.go

You'll see output that looks like this:

2019/02/22 08:19:20 Starting on task 4d43b618-9696-4398-aad4-f0e4f7bfc0d7

If you switch back to the shell that is watching the state of the job, in about 40 seconds you will see it transition from this:

  "Done": false,
  "Result": "",
  "State": 1,
  "created-time": "2019-02-22T16:21:43.055Z",
  "done-time": "0001-01-01T00:00:00Z",
  "id": "ec82ea86-66d6-46c0-8e9f-ed0b073356ab"

to this:

  "Done": true,
  "Result": "OK completed",
  "State": 2,
  "created-time": "2019-02-22T16:21:43.055Z",
  "done-time": "2019-02-22T16:22:23.794021Z",
  "id": "ec82ea86-66d6-46c0-8e9f-ed0b073356ab"

At that point you can press CTRL-C to exit the watch command. Switch to the shell with the worker process and press CTRL-C to exit the worker.

Cleanup and next steps

Optionally, delete the deployed function and Cloud Pub/Sub resources:

gcloud functions delete Jobs
gcloud pubsub subscriptions delete worker-tasks
gcloud pubsub topics delete jobs

In production, you would want to add authorization to the API function, for example with Firebase Auth tokens.

Submit a tutorial

Share step-by-step guides

Submit a tutorial

Request a tutorial

Ask for community help

Submit a request

View tutorials

Search Google Cloud tutorials

View tutorials

Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see our Site Policies. Java is a registered trademark of Oracle and/or its affiliates.