Trigger a pipeline run with Cloud Pub/Sub

The following code samples show you how to write, deploy, and trigger a pipeline using an Event-Driven Cloud Function with a Cloud Pub/Sub trigger.

Build and compile a simple Pipeline

Using Kubeflow Pipelines SDK, build a scheduled pipeline and compile it into a YAML file.

Sample hello-world-scheduled-pipeline:

from kfp import compiler
from kfp import dsl

# A simple component that prints and returns a greeting string
def hello_world(message: str) -> str:
    greeting_str = f'Hello, {message}'
    return greeting_str

# A simple pipeline that contains a single hello_world task
def hello_world_scheduled_pipeline(greet_name: str):
    hello_world_task = hello_world(greet_name)

# Compile the pipeline and generate a YAML file

Upload compiled pipeline YAML to Cloud Storage bucket

  1. Open the Cloud Storage browser in the Google Cloud console.

    Cloud Storage Browser

  2. Click the Cloud Storage bucket you created when you configured your project.

  3. Using either an existing folder or a new folder, upload your compiled pipeline YAML (in this example hello_world_scheduled_pipeline.yaml) to the selected folder.

  4. Click the uploaded YAML file to access the details. Copy the gsutil URI for later use.

Create a Cloud Function with Pub/Sub Trigger

  1. Visit the Cloud Functions page in the console.

    Go to the Cloud Functions page

  2. Click the Create function button.

  3. In the Basics section, give your function a name (for example my-scheduled-pipeline-function).

  4. In the Trigger section, select Cloud Pub/Sub as the Trigger type.

    create function configuration choose pubsub as Trigger type image

  5. In the Select a Cloud Pub/Sub topic dropdown, click Create a topic.

  6. In the Create a topic box, give your new topic a name (for example my-scheduled-pipeline-topic), and select Create topic.

  7. Leave all other fields as default and click Save to save the Trigger section configuration.

  8. Leave all other fields as default and click Next to proceed to the Code section.

  9. Under Runtime, select Python 3.7.

  10. In Entry point, input "subscribe" (the example code entry point function name).

  11. Under Source code, select Inline Editor if it's not already selected.

  12. In the file, add in the following code:

      import base64
      import json
      from import aiplatform
      PROJECT_ID = 'your-project-id'                     # <---CHANGE THIS
      REGION = 'your-region'                             # <---CHANGE THIS
      PIPELINE_ROOT = 'your-cloud-storage-pipeline-root' # <---CHANGE THIS
      def subscribe(event, context):
        """Triggered from a message on a Cloud Pub/Sub topic.
              event (dict): Event payload.
              context ( Metadata for the event.
        # decode the event payload string
        payload_message = base64.b64decode(event['data']).decode('utf-8')
        # parse payload string into JSON object
        payload_json = json.loads(payload_message)
        # trigger pipeline run with payload
      def trigger_pipeline_run(payload_json):
        """Triggers a pipeline run
              payload_json: expected in the following format:
                  "pipeline_spec_uri": "<path-to-your-compiled-pipeline>",
                  "parameter_values": {
                    "greet_name": "<any-greet-string>"
        pipeline_spec_uri = payload_json['pipeline_spec_uri']
        parameter_values = payload_json['parameter_values']
        # Create a PipelineJob using the compiled pipeline from pipeline_spec_uri
        job = aiplatform.PipelineJob(
        # Submit the PipelineJob

    Replace the following:

    • PROJECT_ID: The Google Cloud project that this pipeline runs in.
    • REGION: The region that this pipeline runs in.
    • PIPELINE_ROOT: Specify a Cloud Storage URI that your pipelines service account can access. The artifacts of your pipeline runs are stored in the pipeline root.
  13. In the requirements.txt file, replace the contents with the following package requirements:

  14. Click deploy to deploy the Function.