HTTPBenchmark

HTTPBenchmark aims to be inbetween Apache Bench (ab) and a more full blown HTTP test suite. In its simplest form it provides some basic compatibility with ab in that it supports a -n switch to specify the number of requests to make and a -c switch to specify how many concurrent users.

However, the true intended use is to subclass HTTPBenchmark and to override the worker method to implement your own test logic. In this regard you can think of HTTPBenchmark as a framework that allows you to build unit tests for your HTTP application that aim to stress a server as well. See the Worker Code section below.

Installation

Installation is easily accomplished using pip or easy_install.

pip install httpbenchmark

Basic Usage

A script named pb is provided and is intended to be a basic analog to the Apache ab script.

pb -n 1000 -c 25 http://my-host.tld/endpoint/

Worker code

By default HTTPBenchmark does nothing more than send a GET request and expect a 200 OK return code. To fully utilize the system you should create your own script that subclasses HTTPBenchmark.

Here's an incomplete and untested example that is load testing some application that deals with users and friends. It aims to illustrate the fundamentals of using HTTPBenchmark

from httpbenchmark import HTTPBenchmark
from tornado import gen
import random

USER_IDS = [1111, 2222, 3333, 4444]

_url = lambda x: ''.join(['http://my-host.tld/', x])

class MyBenchmark(HTTPBenchmark):
    @gen.coroutine
    def worker(self):
        '''
        get_worker should return a callable that will be called by the async http client
        '''
        if random.choice([True, False]):
            yield self.new_user()
        else:
            yield self.returning_user()

    @gen.coroutine
    def new_user(self):
        user_id = random.choice(USER_IDS)
        self.log.debug("New user: %s" % user_id)

        friends = yield self.open_json(_url("register?uid=%d" % user_id))
        # ... handle registration response ...
        if failure:
            self.log.error("Indicate reason for failure")
            self.log.debug("Show debugging info if you want")
            self.finish_request(False)
        else:
            yield self.next_step(user_id, friends['friendList'][0])

    @gen.coroutine
    def returning_user(self):
        user_id = random.choice(USER_IDS)
        self.log.debug("Returning user: %s" % user_id)

        def handle_login(response, friends):
        friends = yield self.open_json(_url("login?uid=%d" % user_id))
        # ... handle login response ...
        if failure:
            self.log.error("Indicate reason for failure")
            self.log.debug("Show debugging info if you want")
            self.finish_request(False)
        else:
            yield self.next_step(user_id, friends['friendList'][0])

    @gen.coroutine
    def next_step(self, user_id, friend_id):
        # ... do something else ...
        if failure:
            self.log.error("Indicate reason for failure")
            self.log.debug("Show debugging info if you want")
            self.finish_request(False)
        else:
            # success!
            self.finish_request()

if __name__ == '__main__':
    MyBenchmark().main()

Essentials

• This uses Tornado's async generator interface to achive concurrency, your functions need to be wrapped in @gen.coroutine and you should yield between them.
• worker is where your main code lives. It will be called whenever there is a free slot based on concurrency.
• yield self.get(url, code=200) is used to make a GET request. You will get the response object back when the operation completes.
• self.post(url, params={}, callback) is used to POST data. params should be a dictionary and will be sent as the POST data. It functions the same as get otherwise.
• If you're posting to a PHP backend and need to use PHP's neseted array syntax for parameters you pass php_urlencode to the self.post method with a value of True and it will encode the params accordingly.
• self.get_json(url, callback) is a shortcut for getting and parsing json data that is returned. Your callback should accept two arguments, the first is the response object and the second is the decoded json.
• self.finish_request(True/False) should be called to signal the end of a request. If everything worked as you expected pass it True, otherwise pass it False
• self.debug_response(response) is a handy function to use while you're developing your test cases. If you pass it a response object it will print out a summary of the object as well as the headers and body so you can debug the live data.

TODO

• Add some working examples