使用Python安排Google Cloud Dataflow作业

Question

Currently these are the options to schedule the execution of a Dataflow's job that I know:

• Using App Engine Cron Service or Cloud Functions.

  • This example is with Java, There are any official example with Python as simple?
  • This example is with Python but I'm not sure if currently is still a good option or is "deprecated"

• From a cron job in a Compute Engine

  • Any tutorial of this?

• Using windowing in a streaming pipeline

  • I think this is the easiest but, is the best thinking in total cost?

• Cloud Scheduler

  • Is this a valid method?

Answer 1

I use App Engine Flex as my Dataflow launcher. This microservice has endpoints to launch dataflow jobs on demand, which cron can hit too.

This is my project structure:

df_tasks/
- __init__.py
- datastore_to_csv.py
- ...other_piplines
__init__.py
dflaunch.yaml
main.py
setup.py <-- used by pipelines

The trick with this for me was getting my pipeline dependencies set up correctly. Namely, using a setup.py for pipeline dependencies. Setting it up like this example helped out the most: https://github.com/apache/beam/tree/master/sdks/python/apache_beam/examples/complete/juliaset

setup.py:

import setuptools

setuptools.setup(
    name='dataflow_python_pipeline',
    version='1.0.0',
    description='DataFlow Python Pipeline',
    packages=setuptools.find_packages(),
)

My pipline configs in df_tasks then looks like this:

pipeline_options = PipelineOptions.from_dictionary({
        'project': project,
        'runner': 'DataflowRunner',
        'staging_location': bucket_path+'/staging',
        'temp_location': bucket_path+'/temp',
        'setup_file': './setup.py'
    })

Then in main.py:

from df_tasks import datastore_to_csv

project_id = os.environ['GCLOUD_PROJECT']

@app.route('/datastore-to-csv', methods=['POST'])
def df_day_summary():
    # Extract Payload
        payload = request.get_json()
        model = payload['model']
        for_date = datetime.datetime.strptime(payload['for_date'], '%Y/%m/%d')
    except Exception as e:
        print traceback.format_exc()
        return traceback.format_exc()
    # launch the job
    try:
        job_id, job_name = datastore_to_csv.run(
            project=project_id,
            model=model,
            for_date=for_date,
        )
        # return the job id
        return jsonify({'jobId': job_id, 'jobName': job_name})
    except Exception as e:
        print traceback.format_exc()
        return traceback.format_exc()

Answer 2

There are multiple ways, but one that I think would be very convenient for you would be using the DataflowPythonOperator of Apache Airflow.

GCP offers a managed service for Apache Airflow in the form of Cloud Composer , which you can use to schedule Dataflow pipelines, or other GCP operations.