refactor(e2e): simplify test infrastructure by removing duplication and dead code

e2e_test/bindings_go/conftest.py

@@ -9,10 +9,7 @@
 
 import logging
 import os
-import signal
-import socket
 import subprocess
-import time
 from collections.abc import Generator
 from pathlib import Path
 from typing import TYPE_CHECKING
@@ -22,6 +19,9 @@
 if TYPE_CHECKING:
     from infra import ModelInstance, ModelPool
 
+from infra import get_open_port, release_port, terminate_process
+from infra.process_utils import wait_for_health
+
 logger = logging.getLogger(__name__)
 
 # Paths
@@ -30,25 +30,6 @@
 _GO_OAI_SERVER = _GO_BINDINGS / "examples" / "oai_server"
 
 
-def _find_free_port() -> int:
-    """Find an available port."""
-    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
-        s.bind(("", 0))
-        return s.getsockname()[1]
-
-
-def _wait_for_server(host: str, port: int, timeout: float = 30.0) -> bool:
-    """Wait for server to become available."""
-    start = time.time()
-    while time.time() - start < timeout:
-        try:
-            with socket.create_connection((host, port), timeout=1.0):
-                return True
-        except (TimeoutError, ConnectionRefusedError, OSError):
-            time.sleep(0.5)
-    return False
-
-
 @pytest.fixture(scope="session")
 def go_ffi_library() -> Path:
     """Build the Go FFI library and return its directory path."""
@@ -241,7 +222,7 @@ def go_oai_server(
     grpc_endpoint = f"grpc://localhost:{grpc_worker.port}"
 
     # Find a free port for the Go OAI server
-    oai_port = _find_free_port()
+    oai_port = get_open_port()
 
     # Set up environment - the Go OAI server uses env vars for config
     env = os.environ.copy()
@@ -261,17 +242,26 @@ def go_oai_server(
 
     cmd = [str(go_oai_binary)]
 
-    process = subprocess.Popen(
-        cmd,
-        env=env,
-        stdout=subprocess.PIPE,
-        stderr=subprocess.PIPE,
-    )
-
     try:
+        process = subprocess.Popen(
+            cmd,
+            env=env,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+        )
+
         # Wait for server to start
-        if not _wait_for_server("localhost", oai_port, timeout=30.0):
-            stdout, stderr = process.communicate(timeout=5)
+        try:
+            wait_for_health(f"http://localhost:{oai_port}", timeout=30.0, check_interval=0.5)
+        except TimeoutError:
+            try:
+                stdout, stderr = process.communicate(timeout=5)
+            except subprocess.TimeoutExpired:
+                terminate_process(process, timeout=10)
+                pytest.fail(
+                    f"Go OAI server failed to start and did not exit cleanly.\n"
+                    f"Command: {' '.join(cmd)}"
+                )
             pytest.fail(
                 f"Go OAI server failed to start.\n"
                 f"Command: {' '.join(cmd)}\n"
@@ -283,14 +273,9 @@ def go_oai_server(
         yield ("localhost", oai_port, grpc_worker.model_path)
 
     finally:
-        # Shutdown the server
         logger.info("Shutting down Go OAI server...")
-        process.send_signal(signal.SIGTERM)
-        try:
-            process.wait(timeout=10)
-        except subprocess.TimeoutExpired:
-            process.kill()
-            process.wait()
+        terminate_process(process, timeout=10)
+        release_port(oai_port)
 
 
 @pytest.fixture(scope="class")
@@ -318,7 +303,7 @@ def go_oai_server_multi(
     grpc_endpoints = ",".join(f"grpc://localhost:{w.port}" for w in grpc_workers)
 
     # Find a free port for the Go OAI server
-    oai_port = _find_free_port()
+    oai_port = get_open_port()
 
     # Set up environment
     env = os.environ.copy()
@@ -350,17 +335,26 @@ def go_oai_server_multi(
 
     cmd = [str(go_oai_binary)]
 
-    process = subprocess.Popen(
-        cmd,
-        env=env,
-        stdout=subprocess.PIPE,
-        stderr=subprocess.PIPE,
-    )
-
     try:
+        process = subprocess.Popen(
+            cmd,
+            env=env,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+        )
+
         # Wait for server to start
-        if not _wait_for_server("localhost", oai_port, timeout=60.0):
-            stdout, stderr = process.communicate(timeout=5)
+        try:
+            wait_for_health(f"http://localhost:{oai_port}", timeout=60.0, check_interval=0.5)
+        except TimeoutError:
+            try:
+                stdout, stderr = process.communicate(timeout=5)
+            except subprocess.TimeoutExpired:
+                terminate_process(process, timeout=10)
+                pytest.fail(
+                    f"Go OAI server failed to start and did not exit cleanly.\n"
+                    f"Command: {' '.join(cmd)}"
+                )
             pytest.fail(
                 f"Go OAI server failed to start.\n"
                 f"Command: {' '.join(cmd)}\n"
@@ -376,12 +370,8 @@ def go_oai_server_multi(
 
     finally:
         logger.info("Shutting down Go OAI server...")
-        process.send_signal(signal.SIGTERM)
-        try:
-            process.wait(timeout=10)
-        except subprocess.TimeoutExpired:
-            process.kill()
-            process.wait()
+        terminate_process(process, timeout=10)
+        release_port(oai_port)
 
 
 @pytest.fixture(scope="class")

e2e_test/conftest.py

@@ -213,8 +213,6 @@ def pytest_runtest_logstart(nodeid: str, location: tuple) -> None:
 from smg_client import SmgClient
 from smg_client._errors import SmgError
 
-logger = logging.getLogger(__name__)
-
 
 @pytest.fixture
 def smg(setup_backend):

e2e_test/fixtures/__init__.py

@@ -6,9 +6,6 @@
 - setup_backend.py: Backend setup fixtures (class/function-scoped)
 - markers.py: Helper utilities for marker extraction
 
-Legacy modules (to be removed during e2e_response_api migration):
-- ports.py: Use infra.get_open_port() instead
-- router_manager.py: Use infra.Gateway instead
 """
 
 # Pytest hooks (imported by conftest.py via pytest_plugins)

e2e_test/fixtures/pool.py

@@ -16,7 +16,7 @@
 import pytest
 
 if TYPE_CHECKING:
-    from infra import ModelPool
+    from infra import ModelInstance, ModelPool
 
 from .hooks import get_pool_requirements
 
@@ -170,44 +170,60 @@ def model_pool(request: pytest.FixtureRequest) -> ModelPool:
         return _model_pool
 
 
-@pytest.fixture
-def model_client(
-    request: pytest.FixtureRequest, model_pool: ModelPool
-) -> Generator[object, None, None]:
-    """Get OpenAI client for the model specified by @pytest.mark.model().
+def _get_model_instance(
+    request: pytest.FixtureRequest, model_pool: ModelPool, fixture_name: str
+) -> ModelInstance:
+    """Extract model from marker and acquire instance from pool.
 
-    Usage:
-        @pytest.mark.model("meta-llama/Llama-3.1-8B-Instruct")
-        def test_chat(model_client):
-            response = model_client.chat.completions.create(...)
+    Args:
+        request: Pytest fixture request.
+        model_pool: The model pool.
+        fixture_name: Name of the calling fixture (for error messages).
+
+    Returns:
+        Acquired ModelInstance.
     """
-    import openai
-    from infra import PARAM_MODEL
+    from infra import PARAM_MODEL, ConnectionMode
 
     marker = request.node.get_closest_marker(PARAM_MODEL)
-    if marker is None:
+    if marker is None or not marker.args:
         pytest.fail(
             f"Test must be marked with @pytest.mark.{PARAM_MODEL}('model-id') "
-            "to use model_client fixture"
+            f"to use {fixture_name} fixture"
         )
 
     model_id = marker.args[0]
 
     try:
-        # get() auto-acquires the returned instance
-        instance = model_pool.get(model_id)
+        return model_pool.get(model_id, ConnectionMode.HTTP)
     except KeyError:
         pytest.skip(f"Model {model_id} not available in model pool")
 
+
+@pytest.fixture
+def model_client(
+    request: pytest.FixtureRequest, model_pool: ModelPool
+) -> Generator[object, None, None]:
+    """Get OpenAI client for the model specified by @pytest.mark.model().
+
+    Usage:
+        @pytest.mark.model("meta-llama/Llama-3.1-8B-Instruct")
+        def test_chat(model_client):
+            response = model_client.chat.completions.create(...)
+    """
+    import openai
+
+    instance = _get_model_instance(request, model_pool, "model_client")
+
     client = openai.OpenAI(
         base_url=f"{instance.base_url}/v1",
         api_key="not-used",
     )
 
-    yield client
-
-    # Release reference to allow eviction
-    instance.release()
+    try:
+        yield client
+    finally:
+        instance.release()
 
 
 @pytest.fixture
@@ -221,24 +237,9 @@ def model_base_url(
         def test_direct_http(model_base_url):
             response = httpx.get(f"{model_base_url}/health")
     """
-    from infra import PARAM_MODEL
-
-    marker = request.node.get_closest_marker(PARAM_MODEL)
-    if marker is None:
-        pytest.fail(
-            f"Test must be marked with @pytest.mark.{PARAM_MODEL}('model-id') "
-            "to use model_base_url fixture"
-        )
-
-    model_id = marker.args[0]
+    instance = _get_model_instance(request, model_pool, "model_base_url")
 
     try:
-        # get() auto-acquires the returned instance
-        instance = model_pool.get(model_id)
-    except KeyError:
-        pytest.skip(f"Model {model_id} not available in model pool")
-
-    yield instance.base_url
-
-    # Release reference to allow eviction
-    instance.release()
+        yield instance.base_url
+    finally:
+        instance.release()