chpbench

Benchmarking utilities and scripts for configurable-http-proxy.

This consists of two pieces, a worker and a runner.

running the benchmark

install requirements:

python3 -m pip install -r requirements.txt

run a test:

python3 runner.py -w 10 -c 10 -n 1000

usage: runner.py [-h] [--ws] [--msgs MSGS] [--size SIZE] [--delay DELAY]
                 [-n N] [-c C] [-w WORKERS] [--plot]

optional arguments:
  -h, --help            show this help message and exit
  --ws                  Run websocket test instead of http.
  --msgs MSGS           Number of messages per websocket test.
  --size SIZE           Size of each websocket message (or http reply).
  --delay DELAY         Artificial delay to add.
  -n N                  Number of requests to make.
  -c C                  Number of concurrent requests.
  -w WORKERS, --workers WORKERS
                        Number of worker processes.
  --plot                Show a plot of the results after running.

Worker

worker.py is a simple tornado server that handles HTTP and websocket requests. On startup, it adds itself to the proxy via the proxy's REST API.

For HTTP requests, url parameters can govern behavior:

• size sets the size (in bytes) of the body of the reply (default: 0).
• delay adds an artificial delay (in seconds) before sending the reply (default: 0)

Any URL ending with websocket messages are parsed as JSON and echoed back without modification. If the message contains a delay field, an artificial delay will be added before replying to the message.

Runner

runner.py contains some utilities and a script for performing a test. It:

1. starts a proxy
2. starts w workers behind the proxy and adds them to the proxy
3. runs a test, making requests to random endpoints

The test is run twice, once with raw URLs bypassing the proxy, and again via the proxy, so the proxy's contribution can be measured.

The last step should probably be run with something like apache-bench instead of our own runner.

TODO

• Run on Docker Swarm / kubernetes for real-world multi-node testing.
• dump results to a file