— gavinmhackeling.com/blog

In Part 1 of this series I (incompletely) described the process for creating an EC2 instance and installing an Ubuntu/Nginx/Tornado/Supervisor software stack with the NLTK library for Python. In Part 2 I described configuration settings for Nginx and Supervisor and wrote a basic Tornado application that uses routes, templates and NLTK. In this post I describe how to forward a domain name to your EC2 instance’s address, and run multiple instances on your application on multiple processor threads.

Forwarding a domain name to your EC2 instance

First you must map a public IP address to your EC2 private address. In the EC2 tab of the AWS Management Console, select “Elastic IP” from “My Resources” and allocate a new IP address. Then associate the IP address to your EC2 instance. Note that doing so will change your ec2-n-n-n-n.compute-1.amazonaws.com address to reflect the elastic IP. Caution: Elastic IP addresses that are not associated to an EC2 instance will cost you $0.01 ($0.02?) per hour that the address is not associated.

Next, configure your domain name with your registrar to forward the domain to the elastic IP address. It should take only a few minutes for the association of your elastic IP and instance to propagate, and your domain-forwarding should take effect immediately. (Question: if your domain registrar supports forwarding to a domain name rather than to an IP address, can you forward to your EC2 public domain without associating an elastic IP?)

Running multiple instances of Tornado

Pre-forking in Tornado:
This will automatically create a Tornado thread for each CPU core. I have not tried this method yet.

Using Nginx as a reverse proxy for multiple Tornado processes:
I create multiple instances of my Tornado process (one for each core) and make each process listen to a different port. The port is passed as an argument:

python app.py --port=8005

Nginx works as a load-balancing reverse-proxy for the several Tornado instances.

The following is the app.py that is Nginx forwards requests to:

import os.path
import tornado.ioloop
import tornado.web
import nltk

class MainHandler(tornado.web.RequestHandler):
def post(self):
text = self.get_argument("rawtext")
entities = []
postags = nltk.pos_tag(nltk.word_tokenize(text))
#print postags
for chunk in nltk.ne_chunk(postags):
if hasattr(chunk, 'node'):
tmp_tree = nltk.Tree(chunk.node, [(' '.join(c[0] for c in chunk.leaves()))])
entities.append(tmp_tree)
#print entities
self.render("home_post.html", text=text, entities=entities)
def get(self):
self.render("home_get.html")

handlers = [
(r"/", MainHandler),
]

settings = dict(
template_path=os.path.join(os.path.dirname(__file__), "templates"),
)

application = tornado.web.Application(handlers, **settings)

if __name__ == "__main__":
application.listen(8000)
tornado.ioloop.IOLoop.instance().start()