Taskqueue is a simple queuing service dividing work among networked machines. The concept of Taskqueue is perfectly explained in the Work Queues tutorial for RabbitMQ:

The main idea behind Work Queues (aka: Task Queues) is to avoid doing a resource-intensive task immediately and having to wait for it to complete. Instead we schedule the task to be done later. We encapsulate a task as a message and send it to the queue. A worker process running in the background will pop the tasks and eventually execute the job. When you run many workers the tasks will be shared between them.

This concept is especially useful in web applications where it's impossible to handle a complex task during a short HTTP request window.

Taskqueue started as a simpler and easy to manage alternative to BOSS SkyNET thus it provides the following features:

• Code handling a particular type of tasks is installed as a plugin and doesn't require registration;
• Round robin load balancing;
• Crashed workers get restarted without the need of any external tool like daemontools;
• Task queue works as soon as dispatcher and worker pool components get installed;
• Pluggable workitem formats;
• Flexible configuration;
• Small code base (~ 700 lines of code);
• 100% code coverage with unit tests.

Task queue consists of two major components: a dispatcher (optional) and a worker pool manager. The dispatcher listens to messages in the AMQP queue "taskqueue". When a new message arrives the dispatcher parses its body, extracts the workitem and the type of worker needed to handle the workitem. Then the dispatcher resends the workitem to the queue "worker<worker_type>" which worker processes of the type "<worker_type>" listen to.

Dispatchers are not actually needed as a client can send a message to worker's queue directly. But one can find them useful in case it's simpler for a client to incapsulate info on the worker type into a workitem rather than to set a proper routing key.

For every installed plugin the worker manager starts one or more worker processes according to the config file /etc/taskqueue/config.ini. For example for the following config:

[amqp]
host = localhost
user = wfworker
passwd = wfworker
vhost = /wfworker

; Workers

[worker_worker1]
instances = 3

[worker_worker2]
subgroups = bifh1, bifh2

[worker_worker2_bifh1]
instances = 1
user = bifh1

[worker_worker2_bifh2]
instances = 1
user = bifh2

the worker manager creates 3 identical instances of the type worker1 and 2 instances of the type worker2 configured to run under users bifh1 and bifh2. The worker processes of the type worker1 listen to the queue worker_worker1 and the worker processes of the type worker2 listen to the queue worker_worker2.

If a worker process gets crashed the worker manager restarts it.

The following code in the package extpackage.workerext defines a new worker plugin:

from taskqueue.worker import BaseWorker

class Worker(BaseWorker):

    def handle_task(self, workitem):
        # put meat here
        return do_something_heavy(workitem)

The return value of Worker.handle_task() is sent back to AMQP with the routing key "results" though the key is configurable (i.e. when taskqueue is used with Ruote together then the name of the key should be "ruote_workitems").

If you want to disable reporting of results or to modify the way how task results are reported or tracked then override the method report_results of your custom Worker class:

from taskqueue.worker import BaseWorker

class Worker(BaseWorker):

    def handle_task(self, workitem):
        # put meat here
        return do_something_heavy(workitem)

    def report_results(self, channel, workitem):
        pass

The plugin needs to be registered as a pluggable resource in the egg's `setup.py`:

from setuptools import setup

setup(
    entry_points={
        'worker.plugins':
            ['workerext = extpackage.workerext:Worker.factory']
    }
)

Unpack the tarball, then:

$ cd taskqueue
$ dpkg-buildpackage -rfakeroot

The commands above will produce four packages: python-taskqueue-common, python-taskqueue-dispatcher, python-taskqueue-workerpool and python-taskqueue-extra. Install the first three packages and packages with your own plugins.

Update AMQP settings in the section amqp of the file /etc/taskqueue/config.ini.

Start the task queue with the commands:

$ sudo /etc/init.d/python-taskqueue-dispatcher start
$ sudo /etc/init.d/python-taskqueue-workerpool start

or alternatively:

$ dispatcher --foreground --config=testconfig.ini -p /tmp/disp.pid
$ worker --foreground --config=testconfig.ini -p /tmp/worker.pid

The packages python-taskqueue-dispatcher and python-taskqueue-workerpool can be installed on different hosts. It's advised to install python-taskqueue-dispatcher on at least two hosts to make the setup more reliable. And python-taskqueue-workerpool should be installed on as many hosts as possible for better load balancing.

The package python-taskqueue-extra contains example worker plugins.