test: add transport-wiring regression tests for #459

packages/data-designer-engine/tests/engine/models/clients/test_native_http_clients.py

@@ -23,13 +23,18 @@
 _ANTHROPIC_ENDPOINT = "https://api.anthropic.com/v1"
 _SYNC_CLIENT_PATCH = "data_designer.engine.models.clients.adapters.http_model_client.lazy.httpx.Client"
 _ASYNC_CLIENT_PATCH = "data_designer.engine.models.clients.adapters.http_model_client.lazy.httpx.AsyncClient"
+_HTTP_TRANSPORT_PATCH = "data_designer.engine.models.clients.adapters.http_model_client.lazy.httpx.HTTPTransport"
+_ASYNC_HTTP_TRANSPORT_PATCH = (
+    "data_designer.engine.models.clients.adapters.http_model_client.lazy.httpx.AsyncHTTPTransport"
+)
 
 
 def _make_openai_client(
     *,
     concurrency_mode: ClientConcurrencyMode = ClientConcurrencyMode.SYNC,
     sync_client: MagicMock | None = None,
     async_client: MagicMock | None = None,
+    **kwargs: Any,
 ) -> OpenAICompatibleClient:
     return OpenAICompatibleClient(
         provider_name=_OPENAI_PROVIDER,
@@ -38,6 +43,7 @@ def _make_openai_client(
         concurrency_mode=concurrency_mode,
         sync_client=sync_client,
         async_client=async_client,
+        **kwargs,
     )
 
 
@@ -46,6 +52,7 @@ def _make_anthropic_client(
     concurrency_mode: ClientConcurrencyMode = ClientConcurrencyMode.SYNC,
     sync_client: MagicMock | None = None,
     async_client: MagicMock | None = None,
+    **kwargs: Any,
 ) -> AnthropicClient:
     return AnthropicClient(
         provider_name=_ANTHROPIC_PROVIDER,
@@ -54,6 +61,7 @@ def _make_anthropic_client(
         concurrency_mode=concurrency_mode,
         sync_client=sync_client,
         async_client=async_client,
+        **kwargs,
     )
 
 
@@ -295,17 +303,91 @@ async def test_acompletion_lazy_initializes_async_client(
 # Connection pool size regression tests (issue #459)
 # ---------------------------------------------------------------------------
 
+_POOL_LIMITS_CASES = [
+    pytest.param(_make_openai_client, id="openai"),
+    pytest.param(_make_anthropic_client, id="anthropic"),
+]
+
+_SYNC_TRANSPORT_WIRING_CASES = [
+    pytest.param(_make_openai_client, _OPENAI_MODEL, _make_openai_chat_response(), id="openai"),
+    pytest.param(_make_anthropic_client, _ANTHROPIC_MODEL, _make_anthropic_chat_response(), id="anthropic"),
+]
+
+_ASYNC_TRANSPORT_WIRING_CASES = [
+    pytest.param(_make_openai_client, _OPENAI_MODEL, _make_openai_chat_response(), id="openai"),
+    pytest.param(_make_anthropic_client, _ANTHROPIC_MODEL, _make_anthropic_chat_response(), id="anthropic"),
+]
+
 
-def test_client_limits_respect_max_parallel_requests() -> None:
-    """Connection pool limits must reflect max_parallel_requests (regression for issue #459).
+@pytest.mark.parametrize("client_factory", _POOL_LIMITS_CASES)
+def test_client_limits_respect_max_parallel_requests(client_factory: Callable[..., Any]) -> None:
+    """Connection pool limits must reflect max_parallel_requests.
 
     pool_max = max(32, 2 * max_parallel_requests) = max(32, 600) = 600
     """
-    client = OpenAICompatibleClient(
-        provider_name=_OPENAI_PROVIDER,
-        endpoint=_OPENAI_ENDPOINT,
-        api_key="sk-test",
-        max_parallel_requests=300,
+    client = client_factory(
         concurrency_mode=ClientConcurrencyMode.SYNC,
+        max_parallel_requests=300,
     )
     assert client.limits.max_connections == 600
+
+
+@pytest.mark.parametrize(("client_factory", "model_name", "response_json"), _SYNC_TRANSPORT_WIRING_CASES)
+@patch(_HTTP_TRANSPORT_PATCH)
+@patch(_SYNC_CLIENT_PATCH)
+def test_sync_pool_limits_forwarded_to_transport(
+    mock_client_cls: MagicMock,
+    mock_transport_cls: MagicMock,
+    client_factory: Callable[..., Any],
+    model_name: str,
+    response_json: dict[str, Any],
+) -> None:
+    """Regression for #459: limits must reach HTTPTransport, not just httpx.Client.
+
+    The pre-fix code passed limits= to httpx.Client which silently ignores it
+    when a custom transport= is provided.  The fix constructs HTTPTransport
+    with the correct limits before wrapping it in RetryTransport.  This test
+    fails on the pre-fix code because HTTPTransport was never constructed
+    explicitly (assert_called_once fails).
+    """
+    mock_client_cls.return_value = MagicMock(post=MagicMock(return_value=mock_httpx_response(response_json)))
+    client = client_factory(
+        concurrency_mode=ClientConcurrencyMode.SYNC,
+        max_parallel_requests=300,
+    )
+    client.completion(_make_chat_request(model_name))
+
+    mock_transport_cls.assert_called_once()
+    limits = mock_transport_cls.call_args.kwargs["limits"]
+    assert limits.max_connections == 600
+    assert limits.max_keepalive_connections == 300
+
+
+@pytest.mark.parametrize(("client_factory", "model_name", "response_json"), _ASYNC_TRANSPORT_WIRING_CASES)
+@patch(_ASYNC_HTTP_TRANSPORT_PATCH)
+@patch(_ASYNC_CLIENT_PATCH)
+@pytest.mark.asyncio
+async def test_async_pool_limits_forwarded_to_transport(
+    mock_client_cls: MagicMock,
+    mock_transport_cls: MagicMock,
+    client_factory: Callable[..., Any],
+    model_name: str,
+    response_json: dict[str, Any],
+) -> None:
+    """Regression for #459: limits must reach AsyncHTTPTransport for async clients.
+
+    Same issue as the sync path — the pre-fix code never explicitly constructed
+    AsyncHTTPTransport, so RetryTransport created a default pool with 100
+    connections regardless of max_parallel_requests.
+    """
+    mock_client_cls.return_value = MagicMock(post=AsyncMock(return_value=mock_httpx_response(response_json)))
+    client = client_factory(
+        concurrency_mode=ClientConcurrencyMode.ASYNC,
+        max_parallel_requests=300,
+    )
+    await client.acompletion(_make_chat_request(model_name))
+
+    mock_transport_cls.assert_called_once()
+    limits = mock_transport_cls.call_args.kwargs["limits"]
+    assert limits.max_connections == 600
+    assert limits.max_keepalive_connections == 300