psq 0.3.0

Installation

Install via pip:

pip install psq

Prerequisites

• A project on the Google Developers Console.

• The Google Cloud SDK installed locally.

• You will need the Cloud Pub/Sub API enabled on your project. The link will walk you through enabling the API.

• You will need to run gcloud auth before running these examples so that authentication to Google Cloud Platform services is handled transparently.

Usage

First, create a task:

def adder(a, b):
    return a + b

Then, create a pubsub client and a queue:

from gcloud import pubsub
import psq


PROJECT_ID = 'your-project-id'

client = pubsub.Client(project=PROJECT_ID)

q = psq.Queue(client)

Now you can enqueue tasks:

from tasks import adder

q.enqueue(adder)

In order to get task results, you have to configure storage:

from gcloud import pubsub
import psq


PROJECT_ID = 'your-project-id'

ps_client = pubsub.Client(project=PROJECT_ID)
ds_client = datastore.Client(project=PROJECT_ID)

q = psq.Queue(
    ps_client,
    storage=psq.DatastoreStorage(ds_client))

With storage configured, you can get the result of a task:

r = q.enqueue(adder, 5, 6)
r.result() # -> 11

You can also define multiple queues:

fast = psq.Queue(client, 'fast')
slow = psq.Queue(client, 'slow')

Things to note

Because psq is largely similar to rq, similar rules around tasks apply. You can put any Python function call on a queue, provided:

• The function is importable by the worker. This means the __module__ that the function lives in must be importable. Notably, you can’t enqueue functions that are declared in the main module - such as tasks defined in a file that is run directly with python or via the interactive interpreter.

• The function can be a string, but it must be the absolutely importable path to a function that the worker can import. Otherwise, the task will fail.

• The worker and the applications queuing tasks must share exactly the same source code.

• The function can’t depend on global context such as global variables, current_request, etc. Pass any needed context into the worker at queue time.

Running a worker

Execute psqworker in the same directory where you tasks are defined:

psqworker.py config.q

psqworker only operates on one queue at a time. If you want a server to listen to multiple queues, use something like supervisord to run multiple psqworker processes.

Broadcast queues

A normal queue will send a single task to a single worker, spreading your tasks over all workers listening to the same queue. There are also broadcast queues, which will deliver a copy of the task to every worker. This is useful in situations where you want every worker to execute the same task, such as installing or upgrading software on every server.

broadcast_q = psq.BroadcastQueue(client)

def restart_apache_task():
    call(["apachectl", "restart"])

broadcast_q.enqueue(restart_apache_task)

Broadcast queues provide an implementation of the solution described in Reliable Task Scheduling on Google Compute Engine.

Note: broadcast queues do not currently support any form of storage and do not support return values.

Retries

Raising psq.Retry in your task will cause it to be retried.

from psq import Retry

def retry_if_fail(self):
    try:
        r = requests.get('http://some.flaky.service.com')
    except Exception as e:
        logging.error(e)
        raise Retry()

Flask & other contexts

You can bind an extra context manager to the queue.

app = Flask(__name__)

q = psq.Queue(extra_context=app.app_context)

This will ensure that the context is available in your tasks, which is useful for things such as database connections, etc.:

from flask import current_app

def flasky_task():
    backend = current_app.config['BACKEND']

Ideas for improvements

• some sort of storage solution for broadcast queues.

• Memcache/redis value store.

• @task decorator that adds a delay/defer function.

• Task chaining / groups / chords.

• Late ack.

• Gevent worker.

• batch support for queueing.

Contributing changes

• See CONTRIBUTING.md

Licensing

• Apache 2.0 - See LICENSE