Skip to content

It's like Celery, but with Serverless Google Cloud products.

License

Notifications You must be signed in to change notification settings

flamingo-run/django-cloud-tasks

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Github CI Maintainability Test Coverage python python

Django Cloud Tasks

Powered by GCP Pilot.

Installation

pip install django-google-cloud-tasks

APIs

The following APIs must be enabled in your project(s):

Configuration

Add DJANGO_CLOUD_TASKS_ENDPOINT to your Django settings or as an environment variable

image

Additionally, you can configure with more settings or environment variables:

  • DJANGO_CLOUD_TASKS_APP_NAME: uses this name as the queue name, prefix to topics and subscriptions. Default: None.
  • DJANGO_CLOUD_TASKS_DELIMITER: uses this name as delimiter to the APP_NAME and the topic/subscription name. Default: --.
  • DJANGO_CLOUD_TASKS_EAGER: force the tasks to always run synchronously. Useful for local development. Default: False.
  • DJANGO_CLOUD_TASKS_URL_NAME: Django URL-name that process On Demand tasks. We provide a view for that, but if you want to create your own, set this value. Default: tasks-endpoint.
  • DJANGO_CLOUD_TASKS_SUBSCRIBERS_URL_NAME: Django URL-name that process Subscribers. We provide a view for that, but if you want to create your own, set this value. Default: subscriptions-endpoint.
  • DJANGO_CLOUD_TASKS_PROPAGATED_HEADERS: . Default: ["traceparent"].
  • DJANGO_CLOUD_TASKS_PROPAGATED_HEADERS_KEY: when propagating headers in PubSub, use a key to store the values in the Message data. Default: _http_headers.
  • DJANGO_CLOUD_TASKS_MAXIMUM_ETA_TASK: maximum time in seconds to schedule a task in the future. Default: None. In GCP documentation the maximum schedule time for a task is 30 days from current date and time.

On Demand Task

Tasks can be executed on demand, asynchronously or synchronously.

The service used is Google Cloud Tasks:

from django_cloud_tasks.tasks import Task


class MyTask(Task):
    def run(self, x, y):
        # computation goes here
        print(x**y)


MyTask.asap(x=10, y=3)  # run async (another instance will execute and print)
MyTask.sync(x=10, y=5)  # run sync (the print happens right now)

It's also possible to execute asynchronously, but not immediately:

MyTask.later(task_kwargs=dict(x=10, y=5), eta=3600)  # run async in 1 hour (int, timedelta and datetime are accepted)

MyTask.until(task_kwargs=dict(x=10, y=5), eta=3600)  # run async up to 1 hour, decided randomly (int, timedelta and datetime are accepted)

All of these call methods are wrappers to the fully customizable method push, which supports overriding queue name, headers and more:

MyTask.push(task_kwargs=dict(x=10, y=5), **kwargs)  # run async, but deeper customization is available

Queue

When executing an async task, a new job will be added to a queue in Cloud Tasks to be processed by another Cloud Run instance.

You can choose this queue's name in the following order:

  • Overriding manually when scheduling with push, until or later
  • Defining DJANGO_CLOUD_TASKS_APP_NAME in Django settings
  • otherwise, tasks will be used as queue name

It's also possible to set dynamically with:

from django_cloud_tasks.tasks import Task


class MyTask(Task):
    @classmethod
    def queue(cls) -> str:
        return "my-queue-name-here"

Troubleshooting

When a task if failing in Cloud Tasks and you want to debug locally with the same data, you can get the task ID from Cloud Task UI (the big number in the column NAME) and run the task locally with the same parameters with:

MyTask.debug(task_id="<the task number>")

Cleanup

Google Cloud Tasks will automatically discard any jobs after the max-retries.

If by any reason you need to discard jobs manually, you can provide the Task ID:

MyTask.discard(task_id="<the task number>")

Or you can batch discard many tasks at once:

MyTask.discard()

You can also provide min_retries parameter to filter the tasks that have retried at least some amount (so tasks have some chance to execute):

MyTask.discard(min_retries=5)

Periodic Task

Tasks can be executed recurrently, using a crontab syntax.

The backend used in Google Cloud Scheduler.

from django_cloud_tasks.tasks import PeriodicTask


class RecurrentTask(PeriodicTask):
    run_every = "* * 0 0"  # crontab syntax
    
    def run(self):
        # computation goes here
        ...

For these tasks to be registered in Cloud Scheduler, you must execute the setup once (in production, usually at the same momento you perform database migrations, ou collect static files)

python manage.py schedule_tasks

If you need, you can also run these tasks synchronously or asynchronously, they will behave exactly as a task on demand:

RecurrentTask.asap()  # run async
RecurrentTask.sync()  # run sync

Publisher

Messages can be sent to a Pub/Sub topic, synchronously or asynchronously.

The backend used in Google Cloud PubSub topics.

from django_cloud_tasks.tasks import PublisherTask


class MyPublisher(PublisherTask):
    @classmethod
    def topic_name(cls) -> str:
        return "potato"  # if you don't set one, we'll use the class name (ie. my-publisher)
    

MyPublisher.sync(message={"x": 10, "y": 3})  # publish synchronously
MyPublisher.asap(message={"x": 10, "y": 3})  # publish asynchronously, using Cloud Tasks

For convenience, there's also a dynamic publisher specialized in publishing Django models.

from django_cloud_tasks.tasks import ModelPublisherTask
from django.db.models import Model


class MyModelPublisherTask(ModelPublisherTask):
    @classmethod
    def build_message_content(cls, obj: Model, **kwargs) -> dict:
        return {"pk": obj.pk}  # serialize your model

    @classmethod
    def build_message_attributes(cls, obj: Model, **kwargs) -> dict[str, str]:
        return {}  # any metadata you might want to send as attributes
    
    
one_obj = MyModel.objects.first()

MyModelPublisherTask.sync(obj=one_obj) # publish synchronously
MyModelPublisherTask.asap(obj=one_obj) # publish asynchronously, using Cloud Tasks

Subscriber

Messages can be received in a Pub/Sub subscription, synchronously or asynchronously.

The backend used in Google Cloud PubSub subscriptions.

from django_cloud_tasks.tasks import SubscriberTask


class MySubscriber(SubscriberTask):
    def run(self, content: dict, attributes: dict[str, str] | None = None):
        ...  # process the message you received
    
    @classmethod
    def topic_name(cls) -> str:
        return "potato"
    
    @classmethod
    def subscription_name(cls) -> str:
        return "tomato"  # if you don't set it, we'll use the class name (eg. my-subscriber)

Contributing

When contributing to this repository, you can setup:

As an application (when contributing)

  • Install packages:
    make dependencies
  • If you have changed the package dependencies in poetry.lock:
    make update

As a package (when inside another application)

  • In the application's pyproject.toml, add the remote private repository and the package with version:
[packages]
django-google-cloud-tasks = {version="<version>"}
  • During development, if you wish to install from a local source (in order to test integration with ease):
    # inside the application
    poetry add -e /path/to/this/lib

Testing

To run tests:

make test

To fix linter issues:

make style

Version

Use Semantic versioning.