tpitale / coap_ex

CoAP Server/Client with support for Phoenix

MIT License

17 stars 2 forks source link

Split Messaging/Application layers #72

Open jeanparpaillon opened 3 years ago

jeanparpaillon commented 3 years ago

Rationale

In actual implementation, transport (UDP) / messaging / request/response layers are not explicitly separated.

Implementing the following features is therefore complex:

code testing: tests need to open real UDP ports, which may cause false negatives
different transport implementations: DTLS, websocket, TCP, test
correct implementation of timeouts: CoAP defines a lot of different timeouts, at different layers. Not implementing them correctly lead to process/memory leaking, in particular. Specs defines timeouts per layer, so should be the implementation
may help implementation message deduplication

Proposal

Following guidance: https://tools.ietf.org/id/draft-ietf-lwig-coap-05.html
Split into:
- Messaging layer (behaviour ?): may we call it Transport ?
- Request/response: Connection ?
Leverage GenStateMachine for implementing those 2 layers: provides off the shelf timeout handling and more explicit state machine modelling (see https://tools.ietf.org/id/draft-ietf-lwig-coap-05.html#rfc.section.2.5)

jeanparpaillon commented 3 years ago

@tpitale @johngiffin wdyt ?