Pyramid Tasks

Pyramid and Celery are both fantastic projects that compliment each other well: Pyramid processes synchronous web requests, while Celery performs asynchronous tasks in the background. Unfortunately, due to differences in structure and configuration, it's very difficult to integrate the two together. Configuration, clients, etc. available in Pyramid views may be unavailable in tasks, or may need to be accessed in a different way. Configuration and functionality may have to be duplicated in order to be shared between Pyramid and Celery. Pyramid Tasks aims to bridge this gap by creating a full Pyramid application in the Celery worker and providing a request object to every task. You can use the same configuration for Celery that you do Pyramid, including a Paste-style ini file. Bringing parity to Pyramid and Celery means you can write code for Pyramid and have the code just work in Celery.

To see Pyramid Tasks in action, check out the sample app.

Getting Started

To use Pyramid Tasks, you should first be familiar with Pyramid and Celery.

You can install Pyramid Tasks from PyPI:

pip install pyramid-tasks

Include Pyramid Tasks in your application using config.include, or add it to pyramid.includes in your ini file.

config.include('pyramid_tasks')

Configuring Celery

When you import Pyramid Tasks into your application, a new Celery application is created. All settings prefixed with celery. are put into Celery's configuration. As settings from a .ini file are all strings, values are coerced as necessary. Nested settings are supported by chaining dots, e.g. celery.broker_transport_options.queue_name_prefix.

For example, the following simple celeryconf.py:

broker_url = 'redis://'
broker_transport_options = {
    'visibility_timeout': 3600,
}
result_backend = 'redis://'

Is equivalent to the following .ini file:

celery.broker_url = redis://
celery.broker_transport_options.visibility_timeout = 3600
celery.result_backend = redis://

Running a Worker

If you're running Pyramid via Paste (i.e. an ini file and possibly pserve), you can run a Celery worker using the same ini file.

celery -A pyramid_tasks --ini config.ini

This will create a Pyramid app via the same process pserve does, allowing you to share configuration between the two environments.

You can also create a Celery app using config.make_celery_app(), just like you use config.make_wsgi_app(). If you add app = config.make_celery_app() to celery.py in your project's package, you can invoke celery -A myproject worker to boot a worker.

To see both methods of running a worker in action, take a look at the sample app.

Registering Tasks

To register a new task, call config.register_task with the task function. You can also use the pyramid_tasks.task decorator as long as you run a scan (config.scan()) on the package, just like Pyramid's view_config decorator. register_task and @task take the same arguments as Celery.task.

For example, a simple Pyramid app with a task might look like the following:

from pyramid.config import Configurator


def add(request, x, y):
    return x + y


with Configurator() as config:
    config.register_task(add, name='add')

Invoking a Task

Once a task is registered, you can add it to the work queue using request.defer_task. This takes the task function or a string of the name of the task as the first argument. The remaining arguments (positional and keyword) will be passed to the task. When the task is invoked by a Celery worker, a request object will be created and passed as the first argument.

This request object will share the same configuration as requests in the Pyramid application. This means it will have the same or similar methods, registry, etc. However, it is not the request that invoked the task and properties such as url, GET, etc. will not be present. To use these values in your task, pass them in as arguments.

Let's take our simple Pyramid app and add a view that invokes the task.

from pyramid.config import Configurator


def add_view(context, request):
    request.defer_task(add, int(request.GET['x']), int(request.GET['y']))
    return 'OK\n'


def add(request, x, y):
    return x + y


with Configurator() as config:
    config.add_route('root', '/')
    config.add_view(add_view, route_name='root')
    config.register_task(add, name='add')

You can also use request.defer_task_with_options to pass options into Celery. See the Celery docs for details on what options are available. For example:

def add_view(context, request):
    args = int(request.GET['x']), int(request.GET['y'])
    request.defer_task_with_options(add, args=args, countdown=5)
    return 'OK\n'

Getting Task Results

request.defer_task returns a Celery AsyncResult object. You can use this object to check if the task has completed (AsyncResult.ready()) and to get the return value of the task (AsyncResult.result). See the Celery docs for more information.

AsyncResult also has an id property. If you store this property somewhere, such as a client session, you can use request.get_task_result(id) to return a new AsyncResult object.

pyramid_tm Integration

pyramid_tm is the recommended way of adding transaction management to Pyramid. For example, the Pyramid cookiecutter uses pyramid_tm and zope.sqlalchemy to integrate SQLAlchemy into Pyramid.

Pyramid Tasks includes built-in support for pyramid_tm. It can be enabled by including pyramid_tasks.transaction in your project. This must be included after Pyramid Tasks, but doesn't need to be included before pyramid_tm.

To see Pyramid Tasks, pyramid_tm, and SQLAlchemy in action, check out the SQLAlchemy sample app.

Periodic Tasks

Pyramid Tasks supports Celery Beat for running periodic tasks. After registering a task, use config.add_periodic_task to schedule the task. The arguments mirror Celery.add_periodic_task:

config.add_periodic_task(
    5.0,  # Run every five seconds
    'mytask',
    ('foo', 'bar'),  # Position arguments passed to task
    {'fizz': 'buzz'},  # Keyword arguments passed to task
)

You can also use celery.schedules.crontab as the first argument to use crontab-style scheduling.

You can run the Beat scheduler the same way you run the Celery worker.

celery -A pyramid_tasks beat --ini config.ini

To see Celery Beat in action, check out the beat sample app.

Extending Tasks: Task Derivers

Task Derivers are analogous to Pyramid's View Derivers. They allow you to transform the task before registering it, such as wrapping the task in a transaction or adding metric collection.

A task deriver is a callable that takes two arguments: A task function and an info object. The deriver should return a task callable. The info object has the following attributes:

registry — The registry for the current Pyramid application.
package — The package where the configuration statement was found.
name — The name of the task.
options — The options passed in to the register task action.
original_func — The original task function.

You can register a new task deriver with the Configurator.add_task_deriver method. The first argument is the task deriver. The second argument is the name. If omitted, the name of the task deriver function will be used. It also optionally takes over and under arguments, which work the same as with Pyramid's view deriver.

For example, here's a simple task deriver that wraps the task in a database transaction:

def transaction_deriver(task, info):
    def wrapped(request, *args, **kwargs):
        with request.db:
            task(request, *args, **kwargs)

    return wrapped

def includeme(config):
    config.add_task_deriver(transaction_deriver)

You can pass in options when registering the task to configure your task derivers. For example, here's the same transaction task deriver as above, but now will only wrap the task if the in_transaction option is set.

def transaction_deriver(task, info):
    def wrapped(request, *args, **kwargs):
        with request.db:
            task(request, *args, **kwargs)

    if info.options.get('in_transaction', False):
		    return wrapped
    else:
        return task

Celery will accept any keyword arguments passed in, so no configuration is necessary to use your own options. All options will be set as attributes on the task object.

Extending Tasks: Events

Pyramid Task also fires events using Pyramid's event system. Currently the only event is pyramid_task.events.BeforeDeferTask, which will fire when calling defer_task or defer_task_with_options. The event contains the following attributes:

request — The current request.
task — The task being deferred.
args — The arguments being passed to the task.
kwargs — The keyword arguments being passed to the task.
options — The options being passed into Task.apply_async.

You can modify options in-place and the changes will be reflected in the apply_async call.

For example, here's an event subscriber that adds the current user ID to the headers.

def add_headers(event):
    headers = event.options.setdefault('headers', {})
    headers.setdefault('user_id', event.request.authenticated_userid)

def includeme(config):
    config.add_subscriber(add_headers, BeforeDeferTask)

The user ID will now be accessible from request.current_task.request.user_id.

Fork Safety

Celery by default uses a pre-fork worker model, meaning the application will be initialized and then forked to launch the desired number of workers. This can cause issues with some libraries, especially ones utilizing file descriptors such as database connections. For example, SQLAlchemy requires disposing connections on fork. You can do this by subscribing to the pyramid_task.events.CeleryWorkerProcessInit event.

config.add_subscriber(lambda _: engine.pool.recreate(), CeleryWorkerProcessInit)

The event includes the current application registry in the registry property.

CeleryWorkerProcessInit is triggered by Celery's worker_process_init signal, so use it in the same situations you would that signal.

Acknowledgements

Pyramid Tasks is heavily inspired by the code of PyPA's Warehouse project.

luhn/pyramid-tasks