lemunozm / message-io

Fast and easy-to-use event-driven network library.
Apache License 2.0
1.11k stars 74 forks source link

Improving `Decoder` performance using `Read` trait. #75

Open lemunozm opened 3 years ago

lemunozm commented 3 years ago

The Decoder is used by the FramedTcp transport to transform a stream-based protocol (TCP) into a packet-based protocol that fits really well with the concept of message.

The Decoder collects data from the stream until it can be considered a message. In that process, each chunk of data received from the network is written in a temporal buffer. If that data is not yet a message, then, de Decoder copies from that buffer to its internal buffer in order to wait for more chunks.

This last copy can be avoided if we are able to read directly into the decoder. To get this, the decoder could expose its buffer in order to allow the stream.read() dumping its data directly into the decoder, or even better, the Decoder can receive a Read trait object (that would be the socket) from which extract the data. Something similar to:

Decoder::decode_from(&self mut, reader: &dyn Read, impl decoded_callback: impl FnMut(&[u8]) -> Result<()>

Note that since it works in a non-blocking way, several calls to read must be performed inside this function until receiving a WouldBlock io error.

hasanhaja commented 3 years ago

Hi, I'm new to the code base and I don't have any experience contributing to open source but I want to start and offer my help. Would you be able to give me some pointers on how I can get familiar with the code base enough to tackle this issue?

lemunozm commented 3 years ago

Hi @hasanhaja, thanks for your help!

This improvement is quite localized in the library and only two files should be updated:

To make this change. it is important to be familiar with the Read trait.

Do not hesitate to ask any doubt or any new ideas to tackle the problem. 😃

hasanhaja commented 3 years ago

Hi @lemunozm, thank you for the pointers! I'm looking into the code and getting a feel for what's going on now, and I'll circle back with questions soon.

hasanhaja commented 3 years ago

Task management

@hasanhaja Hi @lemunozm, thank you for the pointers! I'm looking into the code and getting a feel for what's going on now, and I'll circle back with questions soon.

Todo

NOTE: I will add more todos as my exploration develops.

seonWKim commented 7 months ago

Hi @lemunozm , is this still relevant issue? 🤔

lemunozm commented 7 months ago

Hi @seonwoo960000,

It's just a matter of performance improvement. I'm not sure how much it can pump up the performance in real scenarios, to be honest. As far as I know, there is no work done on it right now