Run conformance test with pyvoy

.github/workflows/ci.yaml

@@ -84,7 +84,7 @@ jobs:
       - name: run conformance tests
         # TODO: Debug stdin/stdout issues on Windows
         if: ${{ !startsWith(matrix.os, 'windows-') }}
-        run: uv run pytest -rfEP ${{ matrix.coverage == 'cov' && '--cov=connectrpc --cov-report=xml' || '' }}
+        run: uv run pytest ${{ matrix.coverage == 'cov' && '--cov=connectrpc --cov-report=xml' || '' }}
         working-directory: conformance
 
       - name: run tests with minimal dependencies

conformance/test/_util.py

@@ -1,7 +1,7 @@
 import asyncio
 import sys
 
-VERSION_CONFORMANCE = "v1.0.4"
+VERSION_CONFORMANCE = "0bed9ca203aeda4e344edc442a4b2ede91726db5"
 
 
 async def create_standard_streams():

conformance/test/server.py

@@ -608,6 +608,50 @@ async def serve_hypercorn(
         await proc.wait()
 
 
+async def serve_pyvoy(
+    request: ServerCompatRequest,
+    mode: Literal["sync", "async"],
+    certfile: str | None,
+    keyfile: str | None,
+    cafile: str | None,
+    port_future: asyncio.Future[int],
+):
+    args = ["--port=0"]
+    if certfile:
+        args.append(f"--tls-cert={certfile}")
+    if keyfile:
+        args.append(f"--tls-key={keyfile}")
+    if cafile:
+        args.append(f"--tls-ca-cert={cafile}")
+
+    if mode == "sync":
+        args.append("--interface=wsgi")
+        args.append("server:wsgi_app")
+    else:
+        args.append("server:asgi_app")
+
+    proc = await asyncio.create_subprocess_exec(
+        "pyvoy",
+        *args,
+        stderr=asyncio.subprocess.STDOUT,
+        stdout=asyncio.subprocess.PIPE,
+        env=_server_env(request),
+    )
+    stdout = proc.stdout
+    assert stdout is not None
+    stdout = _tee_to_stderr(stdout)
+    try:
+        async for line in stdout:
+            if b"listening on" in line:
+                port = int(line.strip().split(b"127.0.0.1:")[1])
+                port_future.set_result(port)
+                break
+        await _consume_log(stdout)
+    except asyncio.CancelledError:
+        proc.terminate()
+        await proc.wait()
+
+
 async def serve_uvicorn(
     request: ServerCompatRequest,
     certfile: str | None,
@@ -655,7 +699,7 @@ def _find_free_port():
         return s.getsockname()[1]
 
 
-Server = Literal["daphne", "granian", "gunicorn", "hypercorn", "uvicorn"]
+Server = Literal["daphne", "granian", "gunicorn", "hypercorn", "pyvoy", "uvicorn"]
 
 
 class Args(argparse.Namespace):
@@ -728,14 +772,22 @@ async def main() -> None:
                         request, args.mode, certfile, keyfile, cafile, port_future
                     )
                 )
+            case "pyvoy":
+                serve_task = asyncio.create_task(
+                    serve_pyvoy(
+                        request, args.mode, certfile, keyfile, cafile, port_future
+                    )
+                )
             case "uvicorn":
                 if args.mode == "sync":
                     msg = "uvicorn does not support sync mode"
                     raise ValueError(msg)
                 serve_task = asyncio.create_task(
                     serve_uvicorn(request, certfile, keyfile, cafile, port_future)
                 )
+
         asyncio.get_event_loop().add_signal_handler(signal.SIGTERM, serve_task.cancel)
+
         port = await port_future
         response = ServerCompatResponse()
         response.host = "127.0.0.1"

conformance/test/test_server.py

@@ -27,7 +27,7 @@ def macos_raise_ulimit():
 
 # There is a relatively low time limit for the server to respond with a resource error
 # for this test. In resource limited environments such as CI, it doesn't seem to be enough,
-# notably it is the first request that will take the longest to process as it also sets up
+# notably it is the first message that will take the longest to process as it also sets up
 # the request. We can consider raising this delay in the runner to see if it helps.
 #
 # https://github.com/connectrpc/conformance/blob/main/internal/app/connectconformance/testsuites/data/server_message_size.yaml#L46
@@ -37,24 +37,20 @@ def macos_raise_ulimit():
 ]
 
 
-@pytest.mark.parametrize("server", ["granian", "gunicorn", "hypercorn"])
+@pytest.mark.parametrize("server", ["gunicorn", "pyvoy"])
 def test_server_sync(server: str) -> None:
+    if server == "pyvoy" and sys.platform == "win32":
+        pytest.skip("pyvoy not supported on Windows")
+
     args = maybe_patch_args_with_debug(
         [sys.executable, _server_py_path, "--mode", "sync", "--server", server]
     )
     opts = [
-        # While Hypercorn and Granian supports HTTP/2 and WSGI, they both have simple wrappers
-        # that reads the entire request body before running the application, which does not work for
-        # full duplex. There are no other popular WSGI servers that support HTTP/2, so in practice
-        # it cannot be supported. It is possible in theory following hyper-h2's example code in
-        # https://python-hyper.org/projects/hyper-h2/en/stable/wsgi-example.html though.
+        # TODO: Enable full-duplex in pyvoy
         "--skip",
         "**/bidi-stream/full-duplex/**",
     ]
     match server:
-        case "granian" | "hypercorn":
-            # granian and hypercorn seem to have issues with concurrency
-            opts += ["--parallel", "1"]
         case "gunicorn":
             # gunicorn doesn't support HTTP/2
             opts = ["--skip", "**/HTTPVersion:2/**"]
@@ -78,17 +74,14 @@ def test_server_sync(server: str) -> None:
         check=False,
     )
     if result.returncode != 0:
-        if server == "granian":
-            # Even with low parallelism, some tests are flaky. We'll need to investigate further.
-            print(  # noqa: T201
-                f"Granian server tests failed, see output below, not treating as failure:\n{result.stdout}\n{result.stderr}"
-            )
-            return
         pytest.fail(f"\n{result.stdout}\n{result.stderr}")
 
 
-@pytest.mark.parametrize("server", ["daphne", "granian", "hypercorn", "uvicorn"])
+@pytest.mark.parametrize("server", ["daphne", "pyvoy", "uvicorn"])
 def test_server_async(server: str) -> None:
+    if server == "pyvoy" and sys.platform == "win32":
+        pytest.skip("pyvoy not supported on Windows")
+
     args = maybe_patch_args_with_debug(
         [sys.executable, _server_py_path, "--mode", "async", "--server", server]
     )
@@ -104,9 +97,6 @@ def test_server_async(server: str) -> None:
                 "--skip",
                 "**/full-duplex/**",
             ]
-        case "granian" | "hypercorn":
-            # granian and hypercorn seem to have issues with concurrency
-            opts = ["--parallel", "1"]
         case "uvicorn":
             # uvicorn doesn't support HTTP/2
             opts = ["--skip", "**/HTTPVersion:2/**"]
@@ -115,6 +105,7 @@ def test_server_async(server: str) -> None:
             "go",
             "run",
             f"connectrpc.com/conformance/cmd/connectconformance@{VERSION_CONFORMANCE}",
+            "-v",
             "--conf",
             _config_path,
             "--mode",
@@ -129,10 +120,4 @@ def test_server_async(server: str) -> None:
         check=False,
     )
     if result.returncode != 0:
-        if server == "granian":
-            # Even with low parallelism, some tests are flaky. We'll need to investigate further.
-            print(  # noqa: T201
-                f"Granian server tests failed, see output below, not treating as failure:\n{result.stdout}\n{result.stderr}"
-            )
-            return
         pytest.fail(f"\n{result.stdout}\n{result.stderr}")

pyproject.toml

@@ -42,7 +42,7 @@ dev = [
     "daphne==4.2.1",
     "httpx[http2]==0.28.1",
     "hypercorn==0.17.3",
-    "granian==2.5.5",
+    "granian==2.5.7",
     "gunicorn==23.0.0",
     "just-bin==1.42.4; sys_platform != 'win32'",
     "mkdocs==1.6.1",
@@ -52,6 +52,8 @@ dev = [
     "pytest==8.4.2",
     "pytest-asyncio==1.2.0",
     "pytest-cov==7.0.0",
+    "pytest-timeout==2.4.0",
+    "pyvoy==0.1.1; sys_platform != 'win32'",
     "ruff~=0.13.2",
     "uvicorn==0.37.0",
     # Needed to enable HTTP/2 in daphne
@@ -72,6 +74,7 @@ module-name = "connectrpc"
 
 [tool.pytest.ini_options]
 testpaths = ["test"]
+timeout = 1800       # 30 min
 
 [tool.ruff.format]
 skip-magic-trailing-comma = true