tremolo 0.5.2

Tremolo is a stream-oriented, asynchronous, programmable HTTP server written in pure Python. It can also serve as an ASGI server.

Tremolo is a stream-oriented, asynchronous, programmable HTTP server written in pure Python. It can also serve as an ASGI server for other ASGI web frameworks.

Being built with a stream in mind, Tremolo tends to use yield instead of return in route handlers.

@app.route('/hello')
async def hello_world(**server):
    yield b'Hello '
    yield b'World!'

But that's not the only pattern. You can go straight to the Basics to start building your web application with Tremolo.

Features

Tremolo is only suitable for those who value minimalism and stability over features.

With only 3k lines of code, with no dependencies other than the Python Standard Library, it gives you:

• A production-ready HTTP/1.x server rather than just a development server,
• of course with WebSocket support
• Keep-Alive connections with configurable limit
• Stream chunked uploads
• Stream multipart uploads with per-part streaming
• Download/upload speed throttling, prevent network bandwidth saturation by a single client
• Resumable downloads
• Framework features; routing, CBV, async/sync handlers, middleware, etc.
• ASGI server implementation
• PyPy compatible

All built-in in a single, portable folder module tremolo, in a very compact way like a Swiss Army knife.

Installation

python3 -m pip install --upgrade tremolo

Example

Here is a complete hello world example in case you missed the usual return.

from tremolo import Application

app = Application()

@app.route('/hello')
async def hello_world(**server):
    return 'Hello World!', 'latin-1'


if __name__ == '__main__':
    app.run('0.0.0.0', 8000, debug=True)

Well, latin-1 on the right side is not required. The default is utf-8.

You can save it as hello.py and just run it with python3 hello.py. And your first hello world page with Tremolo will be at http://localhost:8000/hello.

ASGI Server

Tremolo is an HTTP Server framework. You can build abstractions on top of it, say an ASGI server.

In fact, Tremolo already has ASGI server (plus WebSocket) implementation. So you can immediately use existing ASGI applications / frameworks, on top of Tremolo (ASGI server).

For example, If a minimal ASGI application with the name example.py:

async def app(scope, receive, send):
    assert scope['type'] == 'http'

    await send({
        'type': 'http.response.start',
        'status': 200,
        'headers': [
            (b'content-type', b'text/plain')
        ]
    })
    await send({
        'type': 'http.response.body',
        'body': b'Hello, World!'
    })

Then you can run as follows:

python3 -m tremolo --debug --bind 127.0.0.1:8000 example:app

To see more available options:

python3 -m tremolo --help

It's also possible to run the ASGI server programmatically (example with uvloop):

python3 example_uvloop.py

Testing

Just run python3 alltests.py for all tests. Or individual test_*.py in the tests/ folder, for example python3 tests/test_cli.py.

If you also want measurements with coverage:

coverage run alltests.py
coverage combine
coverage report
coverage html # to generate html reports

Benchmarking

The first thing to note is that Tremolo is a pure Python server framework.

As a pure Python server framework, it is hard to find a comparison. Because most servers/frameworks today are full of steroids like httptools, uvloop, Rust, etc.

You can try comparing with Uvicorn with the following option (disabling steroids to be fair):

uvicorn --loop asyncio --http h11 --log-level error example:app

vs

python3 -m tremolo --log-level ERROR example:app

You will find that Tremolo is reasonably fast.

If it's not, it could be due to --upload-rate or --download-rate limits, which take effect when the payload is slightly larger. Despite causing benchmarks to show poor results, it prevents any single client from monopolizing bandwidth, ensuring responsiveness under heavy load.

However, it should be noted that bottlenecks often occur on the application side. Which means that in real-world usage, throughput reflects more on the application than the server.

Misc.

Tremolo utilizes SO_REUSEPORT (Linux 3.9+) to load balance worker processes.

app.run('0.0.0.0', 8000, worker_num=2)

Tremolo can also listen to multiple ports in case you are using an external load balancer like Nginx / HAProxy.

app.listen(8001)
app.listen(8002)

app.run('0.0.0.0', 8000)

You can even get higher concurrency with PyPy or uvloop:

python3 -m tremolo --loop uvloop --log-level ERROR example:app

See: --loop

License

MIT License