Distributed processing using celery in python

Celery is an asynchronous task queue based on distributed message passing. tasks are executed concurrently on a single or more worker servers using multiprocessing, Eventlet or gevent. Tasks can execute asynchronously (in the background) or synchronously (wait until ready).

Architecture:

Fig1 : Celery architecture

The main part of this architecture is broker(transporter), which handles all the task processing.

The client sends task to broker, broker uses round robin to distribute those tasks to workers.

Setup and Installation:

1. Broker/Distributor:

We can use RabbitMQ or Raddis as a broker, celery has full support for these two broker.

But here I am going to use RabbitMQ for setup.

First of all, on broker machine install RabbitMQ.

Rabbit comes for both platform Linux and Windows, you can use anyone.

After installation, we have to configure broker machine so that clients and workers can connect to it.

Configuration:

- Using rabbitmqctl command, run the below commands

  [In windows, you can find it here C:\Program Files\RabbitMQ Server\rabbitmq_server-3.6.6\sbin]

- rabbitmqctl add_user <user_name>

- rabbitmqctl add_vhost <vhost_name>

- rabbitmqctl set_permissions -p <vhost_name> ".*" ".*" ".*"

- rabbitmqctl restart

The above commands will setup a user and virtual host with full permissions on the broker machine.

Now, Install Celery package in python on broker machine.

You can directly install celery using below command.

pip install celery

Your broker machine is ready, let's move towards worker machines setup.

2. Workers:

Install celery package on worker machines as well.
Now, we need to implement connectivity between workers and broker.
Here, I am following the below project structure for worker machine.
distributor [name of project directory]
               |----- celery.py
               |----- tasks.py

You can add below celery.py in project directory (distributor) for the connectivity.

from __future__ import absolute_import, unicode_literals
from celery import Celery
app = Celery('distributor',
             broker='amqp://<user>:<password>@<ip>/<vhost>'),
             backend='amqp://',
             include=['distributor.tasks'])
app.conf.update(result_expires=3600,)
if __name__ == '__main__':
    app.start()

Here,
user - user name of broker machine that we have created
password - password of broker user
ip - IP of broker machine
vhost - virtual host of broker machine
distributor.tasks - distributor is the name of our project directory and tasks is name of python file where we are going to add our tasks.

In tasks.py, we have following simple celery task.

from __future__ import absolute_import, unicode_literals
from .celery import app
from celery.utils.log import get_task_logger
from celery.backends.amqp import AMQPBackend
import time

log = get_task_logger(__name__)
@app.task(backend=AMQPBackend(app, url='amqp://'))
def add(x,y):
 log.info('Calling task add(%d, %d)'%(x,y))
 print('I am in task add')
 z = sum(x,y)
 return z
def sum(x, y):
 time.sleep(10)
 return(x+y)

You can add any python function as a celery task which you want to execute on multiple worker machines at the same time.

We have created woker configuration and worker tasks, now it is time to start the worker.
To start the worker, we need to execute following command from project directory.

celery -A distributor worker -l info

Here,
celery - command
distributor - project directory
-l info - for logging option

After execution of this command, worker will connect with broker and always ready for the task execution.
You can setup multiple woker on the same machine by following command.

worker1(cmd) : celery -A distributor worker -l info -n worker1%n
worker2(cmd) : celery -A distributor worker -l info -n worker2%n

If your tasks are difficult or takes lot of time then it is better to have single worker of single machine.

Now we have broker and woker connectivity.
You can add any number of wokers (depends on your tasks) to broker with the same procedure and same configuration explained above.

3. Client:

We have ready to use distributed architecture with broker and workers.
You can run client at any machine to send tasks to distributed system (broker and workers).
For the demo purpose, I have below client.
Add this script as client.py in project directory (distributor) and run it.

from distributor.tasks import add
import time
task_ids = []
for i in range(20):
 print('Running : %d'%i)
 id = add.delay(5, i)
 task_ids.append(id)
for i in range(len(task_ids)):
 while not (task_ids[i].state == 'SUCCESS'):
  continue
 print(str(i) + ' : ' + str(task_ids[i].get()))

That's it.
If you are having a big task which is running on single machine and takes too much time then you can make it distributed and reduce lot of time.

Monitoring:

For real-time monitoring of distributed architecture, tasks distributions, active running workers, active broker, you can install flower python package.
Flower is a web based tool for monitoring celery clusters, you can install it using simple following command.

pip install flower

After installation, you can run the server using following command.

flower -A proj --port=5555

Now, you can visit this URL http://localhost:5555 for monitoring celery clusters.

Fig2 : Celery Flower

Read More