asgi-replay

A tool for recording and replaying requests using ASGI.

Very early prototype

This is an early prototype. I have used this tool exactly once.

Installation

pip install asgi-replay

Usage

The tool provides two commands - one for recording logs of ASGI requests and one for running a server to replay them.

Recording

asgi-replay record https://www.example.com log.json --port 8000

This runs a local server on port 8000 that proxies content through to www.example.com - and logs each request to log.json.

The log.json file will be over-written for each request. This tool is mainly designed for capturing a single HTTP interaction.

If you add --increment a new file will be written for each request, starting with log-0.json and then log-1.json and so on.

Replaying

To replay a single log - such that any HTTP request to the local server will replay that exact response - run this:

asgi-replay replay log.json --port 8000

(8000 is the default port for both commands.)

What I built this for

I wanted to build a tiny application that could simulate the streaming response from an OpenAI completion endpoint.

The app I built is here: https://github.com/simonw/openai-canned-completion

You can hit it using LLM like this:

OPENAI_API_KEY='x' \
OPENAI_API_BASE='https://openai-canned-completion.vercel.app/v1' \
llm 'hello'

I first ran the following on my own machine:

asgi-replay record https://api.openai.com log.json --port 8000

Then:

OPENAI_API_BASE='http://localhost:8000/v1' \
llm 'hello'

This ran a real request against OpenAI and saved the response in log.json.

Then I ran the replay like this:

asgi-replay record https://api.openai.com log.json --port 8000

And when I ran the llm command again I got the same response, without hitting the OpenAI API:

OPENAI_API_BASE='http://localhost:8000/v1' \
llm 'hello'

Hello! How can I assist you today?

I copied the saved ASGI logs into this ASGI app: https://github.com/simonw/openai-canned-completion/blob/main/canned.py