fusionkit-server 0.2.0__tar.gz → 0.3.1__tar.gz

{fusionkit_server-0.2.0 → fusionkit_server-0.3.1}/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.3
 Name: fusionkit-server
-Version: 0.2.0
+Version: 0.3.1
 Summary: OpenAI-compatible HTTP server for fusionkit.
 Requires-Dist: fastapi>=0.124.4
-Requires-Dist: fusionkit-core==0.2.0
+Requires-Dist: fusionkit-core==0.3.1
 Requires-Dist: uvicorn>=0.38.0
 Requires-Dist: uvicorn[standard]>=0.38.0 ; extra == 'server'
 Requires-Python: >=3.11

{fusionkit_server-0.2.0 → fusionkit_server-0.3.1}/pyproject.toml

@@ -1,11 +1,11 @@
 [project]
 name = "fusionkit-server"
-version = "0.2.0"
+version = "0.3.1"
 description = "OpenAI-compatible HTTP server for fusionkit."
 requires-python = ">=3.11"
 dependencies = [
   "fastapi>=0.124.4",
-  "fusionkit-core==0.2.0",
+  "fusionkit-core==0.3.1",
   "uvicorn>=0.38.0",
 ]
 

{fusionkit_server-0.2.0 → fusionkit_server-0.3.1}/src/fusionkit_server/openai_endpoint.py

@@ -153,13 +153,19 @@ def make_handler(endpoint: ModelEndpoint) -> type[BaseHTTPRequestHandler]:
                 )
 
                 async def run() -> Any:
+                    # Build and close the client within this request's event
+                    # loop so the SDK's HTTP connection pool is released instead
+                    # of leaking a socket/file descriptor per request.
                     client = build_client(endpoint)
-                    return await client.chat(
-                        messages,
-                        sampling,
-                        tools=tools,
-                        tool_choice=tool_choice if isinstance(tool_choice, str) else None,
-                    )
+                    try:
+                        return await client.chat(
+                            messages,
+                            sampling,
+                            tools=tools,
+                            tool_choice=tool_choice if isinstance(tool_choice, str) else None,
+                        )
+                    finally:
+                        await client.aclose()
 
                 started = time.perf_counter()
                 response = asyncio.run(run())