Triggering DAGs (workflows)

This page describes how to use Cloud Functions for event-based DAG triggers.

While Airflow is designed to run DAGs on a regular schedule, you can trigger DAGs in response to events, such as a change in a Cloud Storage bucket or a message pushed to Cloud Pub/Sub. To accomplish this, Cloud Composer DAGs can be triggered by Cloud Functions.

The example in this guide runs a DAG every time a change occurs in a Cloud Storage bucket. Object-change metadata are passed into the DAG's config.

Enabling APIs for your project

Enable the Cloud Composer, Cloud Functions, and Cloud Identity and Access Management (Cloud IAM) APIs.

Enable the APIs

Granting blob signing permissions to the Cloud Functions Service Account

To authenticate to Cloud IAP, grant the Appspot Service Account (used by Cloud Functions) the Service Account Token Creator role on itself. To do this, execute the following command in the gcloud command-line tool or Cloud Shell:

gcloud iam service-accounts add-iam-policy-binding \ \ \

Getting the client ID

To construct a token to authenticate to Cloud IAP, the function requires the client ID of the proxy that protects the Airflow webserver. The Cloud Composer API does not provide this information directly. Instead, make an unauthenticated request to the Airflow webserver and capture the client ID from the redirect URL. The following Python code sample demonstrates how to get the client ID:

import google.auth
import google.auth.transport.requests
import requests
import six.moves.urllib.parse

# Authenticate with Google Cloud.
# See:
credentials, _ = google.auth.default(
authed_session = google.auth.transport.requests.AuthorizedSession(

# project_id = 'YOUR_PROJECT_ID'
# location = 'us-central1'
# composer_environment = 'YOUR_COMPOSER_ENVIRONMENT_NAME'

environment_url = (
    '/environments/{}').format(project_id, location, composer_environment)
composer_response = authed_session.request('GET', environment_url)
environment_data = composer_response.json()
airflow_uri = environment_data['config']['airflowUri']

# The Composer environment response does not include the IAP client ID.
# Make a second, unauthenticated HTTP request to the web server to get the
# redirect URI.
redirect_response = requests.get(airflow_uri, allow_redirects=False)
redirect_location = redirect_response.headers['location']

# Extract the client_id query parameter from the redirect.
parsed = six.moves.urllib.parse.urlparse(redirect_location)
query_string = six.moves.urllib.parse.parse_qs(parsed.query)

Creating your function

Node.js 8/10

Create a function with the below index.js and package.json, filling in the first four constants (see Create a Function). Enable Retry on failure.

When you have finished, your function should resemble the following graphic:


'use strict';

const fetch = require('node-fetch');
const FormData = require('form-data');

 * Triggered from a message on a Cloud Storage bucket.
 * IAP authorization based on:
 * and
 * @param {!Object} data The Cloud Functions event data.
 * @returns {Promise}
exports.triggerDag = async data => {
  // Fill in your Composer environment information here.

  // The project that holds your function
  const PROJECT_ID = 'your-project-id';
  // Navigate to your webserver's login page and get this from the URL
  const CLIENT_ID = 'your-iap-client-id';
  // This should be part of your webserver's URL:
  // {tenant-project-id}
  const WEBSERVER_ID = 'your-tenant-project-id';
  // The name of the DAG you wish to trigger
  const DAG_NAME = 'composer_sample_trigger_response_dag';

  // Other constants
  const WEBSERVER_URL = `https://${WEBSERVER_ID}${DAG_NAME}/dag_runs`;
  const USER_AGENT = 'gcf-event-trigger';
  const BODY = {conf: JSON.stringify(data)};

  // Make the request
  try {
    const iap = await authorizeIap(CLIENT_ID, PROJECT_ID, USER_AGENT);

    return makeIapPostRequest(WEBSERVER_URL, BODY, iap.idToken, USER_AGENT);
  } catch (err) {
    console.error('Error authorizing IAP:', err.message);
    throw new Error(err);

 * @param {string} clientId The client id associated with the Composer webserver application.
 * @param {string} projectId The id for the project containing the Cloud Function.
 * @param {string} userAgent The user agent string which will be provided with the webserver request.
const authorizeIap = async (clientId, projectId, userAgent) => {
  const SERVICE_ACCOUNT = `${projectId}`;
  const JWT_HEADER = Buffer.from(
    JSON.stringify({alg: 'RS256', typ: 'JWT'})

  let jwt = '';
  let jwtClaimset = '';

  // Obtain an Oauth2 access token for the appspot service account
  const res = await fetch(
      headers: {'User-Agent': userAgent, 'Metadata-Flavor': 'Google'},
  const tokenResponse = await res.json();
  if (tokenResponse.error) {
    console.error('Error in token reponse:', tokenResponse.error.message);
    return Promise.reject(tokenResponse.error);

  const accessToken = tokenResponse.access_token;
  const iat = Math.floor(new Date().getTime() / 1000);
  const claims = {
    aud: '',
    iat: iat,
    exp: iat + 60,
    target_audience: clientId,
  jwtClaimset = Buffer.from(JSON.stringify(claims)).toString('base64');
  const toSign = [JWT_HEADER, jwtClaimset].join('.');

  const blob = await fetch(
      method: 'POST',
      body: JSON.stringify({
        bytesToSign: Buffer.from(toSign).toString('base64'),
      headers: {
        'User-Agent': userAgent,
        Authorization: `Bearer ${accessToken}`,
  const blobJson = await blob.json();
  if (blobJson.error) {
    console.error('Error in blob signing:', blobJson.error.message);
    return Promise.reject(blobJson.error);

  // Request service account signature on header and claimset
  const jwtSignature = blobJson.signature;
  jwt = [JWT_HEADER, jwtClaimset, jwtSignature].join('.');
  const form = new FormData();
  form.append('grant_type', 'urn:ietf:params:oauth:grant-type:jwt-bearer');
  form.append('assertion', jwt);

  const token = await fetch('', {
    method: 'POST',
    body: form,
  const tokenJson = await token.json();
  if (tokenJson.error) {
    console.error('Error fetching token:', tokenJson.error.message);
    return Promise.reject(tokenJson.error);

  return {
    idToken: tokenJson.id_token,

 * @param {string} url The url that the post request targets.
 * @param {string} body The body of the post request.
 * @param {string} idToken Bearer token used to authorize the iap request.
 * @param {string} userAgent The user agent to identify the requester.
const makeIapPostRequest = async (url, body, idToken, userAgent) => {
  const res = await fetch(url, {
    method: 'POST',
    headers: {
      'User-Agent': userAgent,
      Authorization: `Bearer ${idToken}`,
    body: JSON.stringify(body),

  if (!res.ok) {
    const err = await res.text();
    console.error('Error making IAP post request:', err.message);
    throw new Error(err);

Update your package.json with the following dependencies.


  "name": "nodejs-docs-samples-functions-composer-storage-trigger",
  "version": "0.0.1",
  "dependencies": {
    "form-data": "^2.3.2",
    "node-fetch": "^2.2.0"
  "engines": {
    "node": ">=8.0.0"
  "private": true,
  "license": "Apache-2.0",
  "author": "Google Inc.",
  "repository": {
    "type": "git",
    "url": ""
  "devDependencies": {
    "@google-cloud/nodejs-repo-tools": "^3.3.0",
    "mocha": "^6.0.0",
    "proxyquire": "^2.1.0",
    "sinon": "^7.2.7"
  "scripts": {
    "test": "mocha test/*.test.js --timeout=20000"


Create a function with the below and requirements.txt, filling in the first four constants (see Create a Function). Enable Retry on failure.

When you have finished, your function should resemble the following graphic:

import google.auth
import google.auth.compute_engine.credentials
import google.auth.iam
from google.auth.transport.requests import Request
import google.oauth2.credentials
import google.oauth2.service_account
import requests


def trigger_dag(data, context=None):
    """Makes a POST request to the Composer DAG Trigger API

    When called via Google Cloud Functions (GCF),
    data and context are Background function parameters.

    For more info, refer to

    To call this function from a Python script, omit the ``context`` argument
    and pass in a non-null value for the ``data`` argument.

    # Fill in with your Composer info here
    # Navigate to your webserver's login page and get this from the URL
    # Or use the script found at
    client_id = 'YOUR-CLIENT-ID'
    # This should be part of your webserver's URL:
    # {tenant-project-id}
    webserver_id = 'YOUR-TENANT-PROJECT'
    # The name of the DAG you wish to trigger
    dag_name = 'composer_sample_trigger_response_dag'
    webserver_url = (
        + webserver_id
        + ''
        + dag_name
        + '/dag_runs'
    # Make a POST request to IAP which then Triggers the DAG
    make_iap_request(webserver_url, client_id, method='POST', json=data)

# This code is copied from
def make_iap_request(url, client_id, method='GET', **kwargs):
    """Makes a request to an application protected by Identity-Aware Proxy.

      url: The Identity-Aware Proxy-protected URL to fetch.
      client_id: The client ID used by Identity-Aware Proxy.
      method: The request method to use
              ('GET', 'OPTIONS', 'HEAD', 'POST', 'PUT', 'PATCH', 'DELETE')
      **kwargs: Any of the parameters defined for the request function:
                If no timeout is provided, it is set to 90 by default.

      The page body, or raises an exception if the page couldn't be retrieved.
    # Set the default timeout, if missing
    if 'timeout' not in kwargs:
        kwargs['timeout'] = 90

    # Figure out what environment we're running in and get some preliminary
    # information about the service account.
    bootstrap_credentials, _ = google.auth.default(

    # For service account's using the Compute Engine metadata service,
    # service_account_email isn't available until refresh is called.

    signer_email = bootstrap_credentials.service_account_email
    if isinstance(bootstrap_credentials,
        # Since the Compute Engine metadata service doesn't expose the service
        # account key, we use the IAM signBlob API to sign instead.
        # In order for this to work:
        # 1. Your VM needs the scope.
        #    You can specify this specific scope when creating a VM
        #    through the API or gcloud. When using Cloud Console,
        #    you'll need to specify the "full access to all Cloud APIs"
        #    scope. A VM's scopes can only be specified at creation time.
        # 2. The VM's default service account needs the "Service Account Actor"
        #    role. This can be found under the "Project" category in Cloud
        #    Console, or roles/iam.serviceAccountActor in gcloud.
        signer = google.auth.iam.Signer(
            Request(), bootstrap_credentials, signer_email)
        # A Signer object can sign a JWT using the service account's key.
        signer = bootstrap_credentials.signer

    # Construct OAuth 2.0 service account credentials using the signer
    # and email acquired from the bootstrap credentials.
    service_account_credentials = google.oauth2.service_account.Credentials(
        signer, signer_email, token_uri=OAUTH_TOKEN_URI, additional_claims={
            'target_audience': client_id
    # service_account_credentials gives us a JWT signed by the service
    # account. Next, we use that to obtain an OpenID Connect token,
    # which is a JWT signed by Google.
    google_open_id_connect_token = get_google_open_id_connect_token(

    # Fetch the Identity-Aware Proxy-protected URL, including an
    # Authorization header containing "Bearer " followed by a
    # Google-issued OpenID Connect token for the service account.
    resp = requests.request(
        method, url,
        headers={'Authorization': 'Bearer {}'.format(
            google_open_id_connect_token)}, **kwargs)
    if resp.status_code == 403:
        raise Exception('Service account {} does not have permission to '
                        'access the IAP-protected application.'.format(
    elif resp.status_code != 200:
        raise Exception(
            'Bad response from application: {!r} / {!r} / {!r}'.format(
                resp.status_code, resp.headers, resp.text))
        return resp.text

def get_google_open_id_connect_token(service_account_credentials):
    """Get an OpenID Connect token issued by Google for the service account.

    This function:

      1. Generates a JWT signed with the service account's private key
         containing a special "target_audience" claim.

      2. Sends it to the OAUTH_TOKEN_URI endpoint. Because the JWT in #1
         has a target_audience claim, that endpoint will respond with
         an OpenID Connect token for the service account -- in other words,
         a JWT signed by *Google*. The aud claim in this JWT will be
         set to the value from the target_audience claim in #1.

    For more information, see .
    The HTTP/REST example on that page describes the JWT structure and
    demonstrates how to call the token endpoint. (The example on that page
    shows how to get an OAuth2 access token; this code is using a
    modified version of it to get an OpenID Connect token.)

    service_account_jwt = (
    request = google.auth.transport.requests.Request()
    body = {
        'assertion': service_account_jwt,
        'grant_type': google.oauth2._client._JWT_GRANT_TYPE,
    token_response = google.oauth2._client._token_endpoint_request(
        request, OAUTH_TOKEN_URI, body)
    return token_response['id_token']

Update your requirements.txt with the following dependencies.



Setting up your DAG

The following DAG contains a BashOperator which prints object-change metadata. In order for it to run in response to Cloud Functions events, copy it into your environment's DAGs folder (see Manage DAGs).

import datetime

import airflow
from airflow.operators import bash_operator

default_args = {
    'owner': 'Composer Example',
    'depends_on_past': False,
    'email': [''],
    'email_on_failure': False,
    'email_on_retry': False,
    'retries': 1,
    'retry_delay': datetime.timedelta(minutes=5),
    'start_date': datetime.datetime(2017, 1, 1),

with airflow.DAG(
        # Not scheduled, trigger only
        schedule_interval=None) as dag:

    # Print the dag_run's configuration, which includes information about the
    # Cloud Storage object change.
    print_gcs_info = bash_operator.BashOperator(
        task_id='print_gcs_info', bash_command='echo {{ dag_run.conf }}')

Testing your function

Upload a file to your Cloud Storage bucket. You should see that the DAG has been triggered and information about the Cloud Storage change has been logged by the BashOperator.

What's next

Σας βοήθησε αυτή η σελίδα; Πείτε μας τη γνώμη σας:

Αποστολή σχολίων σχετικά με…

Αυτή η σελίδα