The server implementation which supports both HTTP/1 and HTTP/2

This server is not intended for production use. I implement this server to understand how low-level web works, low-level HTTP protocol works and Etc.

VERSION

• 0.0.0 : skeleton. (HTTP/1, HTTP/2 with h2c protocol)
• 0.0.1
  • Support TLS.
  • Support HTTP/2 with h2 protocol
  • Support Virtual server by port and domain by.
  • Change interface of using route rule.
  • Fixes bug that can't response completely when there is no route rule at all.
• 0.0.2
  • Support Server Sent Event Feature with HTTP/1.1.
• 0.0.3
  • Support Connection Context and Request Context to inject parameters to endpoint methods.
  • Internal Refactoring.
• 0.0.4
  • Support expect: 100-continue.
  • Improve error handling when server receive invalid HTTP/1.1 request.
  • Internal Refactoring.

How to use

import asyncio
import time

from http_2.server import Server, AsyncServerExecutor
from http_2.ssl_object import SSLConfig
from http_2.common_http_object import GenericHttpRequest, GenericHttpResponse
from http_2.http1_response import StreamingResponse, GeneralHttp1Response

async def main():

    localhost_ssl_config = SSLConfig()
    localhost_ssl_config.add_tls_with_hostname(
        hostname='localhost',
        certfile_path='...',
        keyfile_path='...')

    example_com_ssl_config = SSLConfig()
    example_com_ssl_config.add_tls_with_hostname(
        hostname='example.com',
        certfile_path='...',
        keyfile_path='...')

    http_localhost_server = Server(8080)
    localhost_server = Server(443, hostname='localhost', ssl_config=localhost_ssl_config)
    example_com_server = Server(443, hostname='example.com', ssl_config=example_com_ssl_config)

    @http_localhost_server.get_mapping(path='/sse')
    async def server_sent_event(http_request: GenericHttpRequest, http_response: GenericHttpResponse):
        async def gen_events():
            for _ in range(5):
                await asyncio.sleep(1)
                yield f"data: 현재 시간은 {time.strftime('%Y-%m-%d %H:%M:%S')}\n"

        return StreamingResponse(gen_events())

    @http_localhost_server.route(path='/', methods=['POST', 'GET'])
    @example_com_server.route(path='/', methods=['POST', 'GET'])
    @localhost_server.route(path='/', methods=['POST', 'GET'])
    async def common(http_request: GenericHttpRequest, http_response: GenericHttpResponse):
        return GeneralHttp1Response(body='hello')

    executor = AsyncServerExecutor()
    executor.add_server(http_localhost_server)
    executor.add_server(localhost_server)
    executor.add_server(example_com_server)

    async with executor:
        await executor.execute_forever()

if __name__ == '__main__':
    asyncio.run(main())

How to add your custom filters which is deal with expect: 100-continue

from http_2.public.expect_continue_chains import CONTINUE_CHAINS, Continue100Chain
from http_2.public.expect_continue_exception import UnauthorizedException

@CONTINUE_CHAINS.add
class ValidateUnauthorizedChain(Continue100Chain):

  def __call__(self,
               method: str,
               uri: str,
               protocol: str,
               headers: dict[str, str]) -> None:
    if 'jwt-token' in headers.keys() and headers['jwt-token'].lower() != 'abcdefg':
      raise UnauthorizedException(f'Invalid JWT Token.')

How to test

For Unit and Integration Test

$ py.test ./tests/

For HTTP/2 Spec test

$ brew install h2spec
$ cd http_h2
$ python main.py
...

$ h2spec -p 8080 -h localhost 
>>
Generic tests for HTTP/2 server
  1. Starting HTTP/2
    ✔ 1: Sends a client connection preface
    ...

Hypertext Transfer Protocol Version 2 (HTTP/2)
  3. Starting HTTP/2
    3.5. HTTP/2 Connection Preface
      ✔ 1: Sends client connection preface
      ✔ 2: Sends invalid connection preface
    ...

HPACK: Header Compression for HTTP/2
  2. Compression Process Overview
    ...

Finished in 0.2501 seconds
146 tests, 145 passed, 1 skipped, 0 failed

Performance HTTP/1

• HTTP/1 Condition
  • Local M1 Mac.
  • Locust load test.
  • FastAPI with uvicorn.
  • Custom Server.
  • Using keep-alive.

RPS

Server Type	Type	Name	# Requests	# Fails	Average (ms)	Min (ms)	Max (ms)	Average size (bytes)	RPS	Failures/s
CUSTOM SERVER	GET	/hello/ballo	659853	0	69.16	0	1189	1	1461.36	0
FAST API	GET	/hello/ballo	552122	0	244.37	0	6732	1	1220.07

Response Time

Server Type	Method	Name	50%ile (ms)	60%ile (ms)	70%ile (ms)	80%ile (ms)	90%ile (ms)	95%ile (ms)	99%ile (ms)	100%ile (ms)
CUSTOM SERVER	GET	/hello/ballo	35	49	69	100	160	230	640	1200
FAST API	GET	/hello/ballo	93	190	310	440	650	820	1300	6700

Performance HTTP/2

• HTTP/2 Condition
  • Local M1 Mac.
  • Locust load test. (h2 client was integrated to locust).
  • FastAPI with hypercorn (Note that uvicorn seems to have not HTTP/2 compatibility.
  • Custom Server

RPS

Server Type	Type	Name	# Requests	# Fails	Average (ms)	Min (ms)	Max (ms)	Average size (bytes)	RPS	Failures/s
Custom Server	GET	/hello/ballo	705092	0	15.71	0	290	0	1564.11	0
FAST API	GET	/hello/ballo	500745	0	429.87	0	3893	0	1111.84	0

Response Time

Server Type	Method	Name	50%ile (ms)	60%ile (ms)	70%ile (ms)	80%ile (ms)	90%ile (ms)	95%ile (ms)	99%ile (ms)	100%ile (ms)
Custom Server	GET	/hello/ballo	9	12	16	21	33	56	120	290
FAST API	GET	/hello/ballo	200	360	730	960	1200	1300	1400	3900