h2/h3 trailing header support. Fixes #147

[jeffsawatzky]

This is an initial implementation of trailing header suppport for http2/http3. This fixes #147, in theory.

Here's the thing though...I have no clue what I am doing. I would like some tips on how to do the following:

1. Ensure that I am following the asgi spec properly
2. Ensure that I am following the HTTP/2 HTTP/3 spec properly

I assume the underlying h2 and h3 libraries take care of (2) for me, but I'm not sure.

[jeffsawatzky]

@pgjones can you provide feedback/comments on my stuff so far? Don't want to get to far into this if I'm going in the wrong direction. Thanks.

[pgjones]

Thanks for this, I've adapted it a bit and merged in d8de5f2

[pgjones]

@jeffsawatzky why allow trailers to be sent before the response?

[jeffsawatzky]

@pgjones because there may not be a response. gRPC uses the trailing headers for error responses, so if there was an error server side before any response was sent then it would just return a trailing header and no response.

[jeffsawatzky]

@pgjones if you look at this:
https://github.com/grpc/grpc/blob/master/doc/PROTOCOL-HTTP2.md
The response could be trailers only.

[pgjones]

I think the status would have to be 200 (I don't think there is a self.response to use), does this make sense to you?

[jeffsawatzky]

For my usecase with gRPC, yes that makes sense. All gRPC responses are supposed to be 200.
https://github.com/grpc/grpc/blob/master/doc/PROTOCOL-HTTP2.md#responses

Whether it makes sense generally, I think so?

[synodriver]

I did a quick test with this implementation with python-grpc-client. The asgi app just works like a grpc server, however, the client is not happy with the server and lost the connection half way. It seems that the server send trailers after EndBody been send, which closed the connection. maybe we should avoid this. I'll dig deeper tomorrow.

[pgjones]

Thanks both, hopefully fixed with d16b503

[synodriver]

Emmmm... That didn't work either. Here is my asgi app which pretend to be a grpc server(A simple grpc echo server)

async def grpcapp(scope, receive, send):
    if scope["type"] == "http":
        body = (await receive())["body"]
        await send(
            {
                "type": "http.response.start",
                "status": 200,
                "headers": [
                    (b"Content-Type", b"application/grpc+proto"),
                    (b"Cache-Control", b"no-cache"),
                    (b"Trailer", b"grpc-status")
                ],
                "trailers": True,
            }
        )

        await send({"type": "http.response.body", "body": body})
        await send({"type": "http.response.trailers", "headers": [(b"grpc-status", b"0")]})

A real grpc client would expect the same response type and value when talk to the server, but with grpclib in python, I just got

Traceback (most recent call last):
  File "D:\conda\envs\py310\lib\site-packages\grpclib\client.py", line 468, in recv_trailing_metadata
    trailers = await self._stream.recv_trailers()
  File "D:\conda\envs\py310\lib\site-packages\grpclib\protocol.py", line 351, in recv_trailers
    await self.trailers_received.wait()
  File "D:\conda\envs\py310\lib\asyncio\locks.py", line 214, in wait
    await fut
asyncio.exceptions.CancelledError

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "E:\pyproject\nonecorn\src\pclient.py", line 28, in <module>
    asyncio.run(main())
  File "D:\conda\envs\py310\lib\asyncio\runners.py", line 44, in run
    return loop.run_until_complete(main)
  File "D:\conda\envs\py310\lib\asyncio\base_events.py", line 649, in run_until_complete
    return future.result()
  File "E:\pyproject\nonecorn\src\pclient.py", line 23, in main
    reply = await greeter.SayHello(Test(id=2, data="xsndjasndsa"))
  File "D:\conda\envs\py310\lib\site-packages\grpclib\client.py", line 902, in __call__
    async with self.open(timeout=timeout, metadata=metadata) as stream:
  File "D:\conda\envs\py310\lib\site-packages\grpclib\client.py", line 563, in __aexit__
    raise exc_val
  File "D:\conda\envs\py310\lib\site-packages\grpclib\client.py", line 553, in __aexit__
    await self._maybe_finish()
  File "D:\conda\envs\py310\lib\site-packages\grpclib\client.py", line 523, in _maybe_finish
    await self.recv_trailing_metadata()
  File "D:\conda\envs\py310\lib\site-packages\grpclib\client.py", line 467, in recv_trailing_metadata
    with self._wrapper:
  File "D:\conda\envs\py310\lib\site-packages\grpclib\utils.py", line 70, in __exit__
    raise self._error
grpclib.exceptions.StreamTerminatedError: Connection lost

And here is the client side:

import asyncio
import os
import ssl
from datetime import datetime

os.environ["PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION"] = "python"
from grpclib.client import Channel

from test_grpc import TestServerStub

# generated by protoc, it doens't matter since you can generate your own
from test_pb2 import Test


async def main():
    async with Channel("127.0.0.1", 9001, ssl=False) as channel:
        greeter = TestServerStub(channel)

        reply = await greeter.SayHello(Test(id=2, data="xsndjasndsa"))
        print(reply)


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

Besides, grpcurl also failed with ERROR: Code: Internal Message: server closed the stream without sending trailers