grpc: trailing metadata can't be passed when aborting an AIO server RPC

[cookiefission]

Describe your environment

Python: 3.12.0
opentelemetry-distro[otlp]==0.42b0
opentelemetry-instrumentation-grpc==0.42b0
grpcio==1.59.2

Steps to reproduce
Full replication can be found here.

server.py:

import asyncio
import logging

import grpc
import helloworld_pb2
import helloworld_pb2_grpc


class Greeter(helloworld_pb2_grpc.GreeterServicer):
    async def SayHello(
        self,
        request: helloworld_pb2.HelloRequest,
        context: grpc.aio.ServicerContext,
    ) -> helloworld_pb2.HelloReply:
        metadata = (
            ("this", "should"),
            ("work", "nicely")
        )
        await context.abort(code=grpc.StatusCode.ABORTED, details="This is the actual error message", trailing_metadata=metadata)


async def serve() -> None:
    server = grpc.aio.server()
    helloworld_pb2_grpc.add_GreeterServicer_to_server(Greeter(), server)
    listen_addr = "[::]:50051"
    server.add_insecure_port(listen_addr)
    logging.info("Starting server on %s", listen_addr)
    await server.start()
    await server.wait_for_termination()


if __name__ == "__main__":
    logging.basicConfig(level=logging.INFO)
    asyncio.run(serve())

This server is just adopted from the grpc examples. The helloworld protobuf artefacts can be found there.

For a client, the async_greeter_client.py from grpc examples can be used.

This only happens when trailing_metadata is passed as an argument to context.abort.

What is the expected behavior?

Running the server without opentelemetry, the (expected) error from running the client is:

grpc.aio._call.AioRpcError: <AioRpcError of RPC that terminated with:
        status = StatusCode.ABORTED
        details = "This is the actual error message"
        debug_error_string = "UNKNOWN:Error received from peer  {grpc_message:"This is the actual error message", grpc_status:10, created_time:"2023-11-16T15:09:03.552303+00:00"}"
>

When running the server under opentelemetry-instrument, I expect the same output as when it is run without OTEL.

What is the actual behavior?

Instead, an exception is still raised but the underlying code, details, and trailing_metadata are missing. This makes any nuanced error handling on the client side impossible.

grpc.aio._call.AioRpcError: <AioRpcError of RPC that terminated with:
        status = StatusCode.UNKNOWN
        details = "Unexpected <class 'TypeError'>: _OpenTelemetryServicerContext.abort() got an unexpected keyword argument 'trailing_metadata'"
        debug_error_string = "UNKNOWN:Error received from peer  {created_time:"2023-11-16T15:30:02.017999+00:00", grpc_status:2, grpc_message:"Unexpected <class \'TypeError\'>: _OpenTelemetryServicerContext.abort() got an unexpected keyword argument \'trailing_metadata\'"}"
>

Additional context

The abort method for grpc.ServicerContext and grpc.aio.ServicerContext is different. It's unclear why. The AIO instrumentation re-uses the _OpenTelemetryServicerContext from the non-async server instrumentation which is how this bug snuck in, despite there being a test in place for the abort.

[cookiefission]

I'm working on a PR to fix this.