CODEWITHSUNDEEP
pythongunicorneventlet
Nikos Vita Topiko
Nikos Vita Topiko

Reputation: 551

What's the difference between gunicorn and starting an wsgi server programmatically?

I would like to know what is the difference between starting a gunicorn WSGI server with eventlet workers 

gunicorn --workers=2 -k eventlet test:app

and starting a wsgi server programmatically like 

from eventlet import wsgi
import eventlet

def hello_world(env, start_response):
    start_response('200 OK', [('Content-Type', 'text/plain')])
    return ['Hello, World!\r\n']

wsgi.server(eventlet.listen(('', 8090)), hello_world)

I guess gunicorn is more performant but why? What are the differences? If I have to do it programmatically would I have performance issues?

Upvotes: 1

Views: 6941

Answers (3)

Joseph Sheedy
Joseph Sheedy

Reputation: 6736

One big difference is that gunicorn forks worker processes. In your example you pass the argument --workers=2 which will spawn 2 worker processes. Since Python has the Global Interpreter Lock (GIL), a single process running Eventlet can not take advantage of the performance of multiple processor cores while the multiple worker processes spawned by gunicorn can. If your application is CPU intensive, this might make a large improvement in performance. Of course, you could just launch multiple processes like your eventlet.wsgi example yourself.

The other answers here linked to some good resources for learning about the options.

Upvotes: 5

Martijn Pieters
Martijn Pieters

Reputation: 1123740

Both gunicorn and the wsgi module are implementations of a standard for hosting Python applications. That standard is called the Web Server Gateway Interface, or WSGI. There are many more such implementations.

Which one you pick depends on your specific needs, depending on your deployment target (what can be installed there), your memory and CPU constraints, performance metrics of the specific implementation, specific features the implementations offer, project security record (does the project support in-production deployments, number of security issues raised, response time to security issues), etc. We can't make that choice for you.

Also see:

  • this overview of WSGI servers, which recommends Gunicorn as a good starting point but has a shortlist of more options for you to pick from.
  • This two-parter overview of WSGI servers (make sure to read part 2 for the performance analysis), which covers a different subset of servers. I must note however that bjoern has not been updated in a while and as a project is going to require more Python knowledge to set up correctly.

Upvotes: 3

panchicore
panchicore

Reputation: 11932

Indeed its performance and the way the system manages the process, the key metrics you should consider evaluating eventlet.wsgi vs gunicorn are:

  • requests served
  • latency
  • ram usage
  • cpu usage
  • error rate

you can read more about some python wsgi servers benchmark here and find out someone who managed to compare the ones you want to define the one you need.

also, related to this, you should consider process control when deploying on production.

Upvotes: 1

Related Questions