Job Queue


In request-response model server usually has very short time to run transactions, and thus it doesn't have enough time to handle time-consuming jobs. For example, the web server needs to respond in less than 30 seconds, while an API importing his/her data might need to take 5 minutes.

Job Queue is a general solution to solve the problem.


A Job queue, sometimes task queue or batch queue, is a data structure containing jobs to run outside the common request-response cycle.

A job is a data structure containing running context, and usually encoded as strings or byte streams in job queue.

A job producer initializes jobs, serializes them and sends them out. The producer can be the HTTP process, desktop GUI main process, or anything that is required to be responsive.

A job consumer receives jobs from the job queue, deserializes them and runs jobs in isolated environment.


Common Pattern

Below involves a set of popular job queue frameworks like Python-RQ, Celery, and Sidekiq. By comparing their APIs and design philosophies, we could have a deeper understanding of job queue.

Setup Message Queue

First, run a Message Queue software. It can be Redis, RabbitMQ, ZeroMQ, Disque, etc. Without the Message Queue running, the publisher can't publish jobs and the consumer would very likely crash. However, Job Queue won't launch Message Queue for you usually, meaning you need to manually bring up Message Queue instance before running a Job Queue.

Setup Queue and Connection

Job Queue usually requires you to define a queue object. The setup function usually needs the connection parameters of Message Queue and return the job queue object. This queue object encapsulates all functionalities a producer needs.

For example, in Python-RQ,

from redis import Redis
from rq import Queue

rq = Queue(connection=Redis())

For example, in Celery,

from celery import Celery

celery = Celery('tasks', broker='pyamqp://guest@localhost//')

Note that some frameworks many have already created a Job Queue so you only need to configure it. For example, in Sidekiq,

require 'sidekiq'

Sidekiq.configure_server do |config|
  config.redis = { url: 'redis://redis.svc:6379/0' }

Publish the job

Almost all framework try to simplify the effort of publishing the job. The API has two common kinds:

! Enqueue Plain Function

Define plain function, and use the queue to enqueue function call.

For example, in Python-RQ,

def count_words_at_url(url):
    return len(requests.get(url).content.split())

rq.enqueue(count_words_at_url, '')

! Make Plain Function Asynchronous

Decorate plain function with the queue to make it asynchronous.

For example, in Celery,

def count_words_at_url(url):
    return len(requests.get(url).content.split())


For example, in Sidekiq,

class CountWordsAtURLWorker
  include Sidekiq::Worker
  def perform(url)
    open(url) do |r|


Dedicated Worker as Consumer

Most Job Queue softwares take over the worker completely, meaning you only need to start the workers and let them run. The worker will start executing the job in the background on fetching one from MQ. You might want to run it as a daemon or under process supervisors.

For example, in Python-RQ,

$ pipenv run rq worker

For example, in Celery,

$ pipenv run celery -A tasks worker

For example, in Sidekiq,

$ bundle exec sidekiq


Since Job Queue process runs in the background, so monitoring is a battery-included feature. It often provides you some job statistics like below:

The Job Queue framework often provides two kinds of interfaces:

Job Context

Below lists some common seen job context data:

Result Storage

Job functions may return values. Some Job Queue frameworks support storing results into a backend. Some frameworks simply have no support for result storage for it might increase the complexity of implementation. Anyway, since we have the job id in the context, it's quite easy to store the id-result into key-value database for the Job Queue. It requires setting a storage backend when the Job Queue is setup.

Error Handling

Job Queue has below solutions for failed jobs usually:

When job failed, it's unrealistic to keep trying running the job. It's usually the maintainer's responsibility to intervene, either to purge the job or fix the code and re-run the job.


Below lists popular Job Queue frameworks.

Other solutions:


Job queue is an essential component to extend request-response model for handling time-consuming jobs. Choose a Job Queue framework that has API and features you like, and make sure that you have solutions to overcome the disadvantages.