Tremolo Under The Hood

[nggit]

Stream-oriented

Streaming is a key approach at Tremolo. Both in the request and response area.

On Request, request.read() or request.stream() will receive and decode chunked encoding on-the-fly!

Not only that, request.files() parses multipart/form-data lazily and you will receive the uploaded files one by one, a.k.a. multipart streaming.

It doesn't wait for the whole body to arrive and then parse it with files = body.split(boundary), like some frameworks do.
That makes parsing easier but in a horrible way.

Slim codebase

Tremolo emphasizes the elegance of design.

...the size constraint has encouraged not only economy, but also a certain elegance of design.

-- Dennis Ritchie and Ken Thompson

This is subjective, but this will be a major concern in development. We will always try to remove rather than add something.
Simpler means fewer bugs and attack surfaces.

Regex is a sin

We don't rely on regex in the code. The only sin we commit is that our routing is powered by regex.

Not that we hate it. Regex is too powerful and may have unpredictable costs; security, maintainability.

Utilizes few asyncio APIs/methods

Tremolo is only built using a few asyncio paradigms.

Mostly from the low-level API category like loop.create_future(), loop.create_task(), loop.call_at() and a few from the high-level API like asyncio.sleep().

This has several benefits:

• Avoid unnecessary Python code underneath (Tremolo is highly optimized to be as fast as possible as pure-Python)
• Minimize breakage due to future changes/deprecation of the asyncio API
• Developers can easily jump into the code and understand it quickly

Resilient

Tremolo is small, but designed to withstand heavy loads. Workers will restart if killed, etc.

We wanted a reliable server with zero maintenance.