添加流式 Copilot 请求处理及审计中间件优化

app/infra/audit/middleware.py

@@ -61,12 +61,24 @@ class AuditMiddleware(BaseHTTPMiddleware):
         "/api/v1/openproject/",
         "/openproject/",
         "/api/v1/copilot/chat/",
+        "/api/v1/copilot/chat/stream",
     }
+    EXCLUDED_PATH_PREFIXES = (
+        "/api/v1/copilot/chat/",
+        "/copilot/chat/",
+    )
 
     async def dispatch(self, request: Request, call_next: Callable) -> Response:
         # 提取开始时间
         start_time = time.time()
 
+        # 流式 Copilot 请求前置排除，避免读取/改写 body 影响 SSE 生命周期
+        if self._is_excluded_path(request.url.path):
+            response = await call_next(request)
+            process_time = time.time() - start_time
+            response.headers["X-Process-Time"] = str(process_time)
+            return response
+
         # 1. 预判是否需要读取Body (针对写操作)
         # 注意：我们暂时移除早期的 return，因为需要等待路由匹配后才能检查 Tag
         should_capture_body = request.method in ["POST", "PUT", "PATCH"]
@@ -75,13 +87,24 @@ class AuditMiddleware(BaseHTTPMiddleware):
         if should_capture_body:
             try:
                 # 注意：读取 body 后需要重新设置，避免影响后续处理
+                original_receive = request._receive
                 body = await request.body()
                 if body:
                     request_data = json.loads(body.decode())
 
-                # 重新构造请求以供后续使用
+                # 重新构造请求以供后续使用：仅回放一次，后续回落原始 receive
+                body_sent = False
+
                 async def receive():
-                    return {"type": "http.request", "body": body}
+                    nonlocal body_sent
+                    if not body_sent:
+                        body_sent = True
+                        return {
+                            "type": "http.request",
+                            "body": body,
+                            "more_body": False,
+                        }
+                    return await original_receive()
 
                 request._receive = receive
             except Exception as e:
@@ -91,7 +114,7 @@ class AuditMiddleware(BaseHTTPMiddleware):
         response = await call_next(request)
 
         # 3. 决定是否审计
-        if request.url.path in self.EXCLUDED_PATHS:
+        if self._is_excluded_path(request.url.path):
             process_time = time.time() - start_time
             response.headers["X-Process-Time"] = str(process_time)
             return response
@@ -151,6 +174,11 @@ class AuditMiddleware(BaseHTTPMiddleware):
 
         return response
 
+    def _is_excluded_path(self, path: str) -> bool:
+        if path in self.EXCLUDED_PATHS:
+            return True
+        return any(path.startswith(prefix) for prefix in self.EXCLUDED_PATH_PREFIXES)
+
     def _resolve_project_id(self, request: Request) -> UUID | None:
         project_header = request.headers.get("X-Project-Id")
         if not project_header:

copilot-sidecar/server.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import asyncio
 import json
+import logging
 import os
 import time
 import uuid
@@ -9,7 +10,8 @@ from dataclasses import dataclass
 from typing import Any, Optional
 
 from fastapi import FastAPI, Request
-from fastapi.responses import JSONResponse, StreamingResponse
+from fastapi.responses import StreamingResponse
+from pydantic import BaseModel, Field, ConfigDict
 from copilot import CopilotClient, PermissionHandler
 
 
@@ -23,11 +25,21 @@ class SessionHolder:
     last_used_at: float
 
 
-app = FastAPI(title="TJWater Copilot Python Sidecar")
+app = FastAPI(title="TJWater Copilot Sidecar")
 client: Optional[CopilotClient] = None
 sessions: dict[str, SessionHolder] = {}
 session_ttl_seconds = int(os.getenv("COPILOT_SESSION_TTL_SECONDS", "1800"))
-model = os.getenv("COPILOT_MODEL", "gpt-5.3-codex")
+model = os.getenv("COPILOT_MODEL", "gpt-4.1")
+logger = logging.getLogger("copilot_sidecar")
+
+
+class ChatStreamRequest(BaseModel):
+    message: str = Field(..., min_length=1, max_length=10000)
+    conversation_id: Optional[str] = Field(
+        default=None, alias="conversationId", max_length=128
+    )
+    user_id: Optional[str] = Field(default=None, alias="userId", max_length=128)
+    model_config = ConfigDict(populate_by_name=True)
 
 
 @app.on_event("startup")
@@ -43,8 +55,8 @@ async def shutdown_event() -> None:
         for holder in sessions.values():
             try:
                 await holder.session.disconnect()
-            except Exception:
-                pass
+            except Exception as exc:
+                logger.warning("Failed to disconnect session during shutdown: %s", exc)
         sessions.clear()
         await client.stop()
 
@@ -62,8 +74,8 @@ async def _cleanup_sessions() -> None:
             continue
         try:
             await holder.session.disconnect()
-        except Exception:
-            pass
+        except Exception as exc:
+            logger.warning("Failed to disconnect expired session %s: %s", sid, exc)
 
 
 async def _get_or_create_session(conversation_id: str):
@@ -76,11 +88,9 @@ async def _get_or_create_session(conversation_id: str):
         raise RuntimeError("Copilot client is not initialized")
 
     session = await client.create_session(
-        {
-            "model": model,
-            "streaming": True,
-            "on_permission_request": PermissionHandler.approve_all,
-        }
+        model=model,
+        streaming=True,
+        on_permission_request=PermissionHandler.approve_all,
     )
     sessions[conversation_id] = SessionHolder(session=session, last_used_at=time.time())
     return session
@@ -92,42 +102,40 @@ async def health() -> dict[str, Any]:
 
 
 @app.post("/chat/stream")
-async def chat_stream(request: Request):
-    payload = await request.json()
-    message = payload.get("message")
-    conversation_id = payload.get("conversationId")
-    if not isinstance(message, str) or not message.strip():
-        return JSONResponse(status_code=400, content={"message": "message is required"})
-
+async def chat_stream(payload: ChatStreamRequest, request: Request):
     conv_id = (
-        conversation_id.strip()
-        if isinstance(conversation_id, str) and conversation_id.strip()
+        payload.conversation_id.strip()
+        if isinstance(payload.conversation_id, str) and payload.conversation_id.strip()
         else f"conv-{uuid.uuid4().hex[:10]}"
     )
+    message = payload.message.strip()
 
     async def event_generator():
-        session = None
         queue: asyncio.Queue[tuple[str, dict[str, Any]]] = asyncio.Queue()
         done = asyncio.Event()
-        error_emitted = False
+        saw_message_delta = False
 
         def on_event(event):
-            nonlocal error_emitted
+            nonlocal saw_message_delta
             event_type = getattr(event.type, "value", str(event.type))
             data = getattr(event, "data", None)
             if event_type == "assistant.message_delta":
                 content = getattr(data, "delta_content", "") or ""
                 if content:
-                    queue.put_nowait(("token", {"conversationId": conv_id, "content": content}))
-            elif event_type == "assistant.message":
+                    saw_message_delta = True
+                    queue.put_nowait(
+                        ("token", {"conversationId": conv_id, "content": content})
+                    )
+            elif event_type == "assistant.message" and not saw_message_delta:
                 content = getattr(data, "content", "") or ""
                 if content:
-                    queue.put_nowait(("token", {"conversationId": conv_id, "content": content}))
+                    queue.put_nowait(
+                        ("token", {"conversationId": conv_id, "content": content})
+                    )
             elif event_type == "session.idle":
                 queue.put_nowait(("done", {"conversationId": conv_id}))
                 done.set()
             elif event_type == "error":
-                error_emitted = True
                 queue.put_nowait(
                     (
                         "error",
@@ -144,22 +152,27 @@ async def chat_stream(request: Request):
             session = await _get_or_create_session(conv_id)
             unsubscribe = session.on(on_event)
             try:
-                await session.send({"prompt": message})
+                await session.send(message)
                 while not done.is_set() or not queue.empty():
                     if await request.is_disconnected():
+                        logger.info(
+                            "Client disconnected during stream: conversation=%s",
+                            conv_id,
+                        )
                         return
                     try:
-                        event_name, event_data = await asyncio.wait_for(queue.get(), timeout=0.2)
+                        event_name, event_data = await asyncio.wait_for(
+                            queue.get(), timeout=0.2
+                        )
                         yield _sse(event_name, event_data)
                     except asyncio.TimeoutError:
                         continue
-                if not error_emitted:
-                    yield _sse("done", {"conversationId": conv_id})
             finally:
                 unsubscribe()
                 if conv_id in sessions:
                     sessions[conv_id].last_used_at = time.time()
         except Exception as exc:
+            logger.exception("Copilot generation failed for %s: %s", conv_id, exc)
             yield _sse(
                 "error",
                 {

scripts/run_server.py

@@ -1,21 +1,83 @@
 import asyncio
-import sys
+import atexit
 import os
+import signal
+import subprocess
+import sys
+from urllib.parse import urlparse
 import uvicorn
 
 # 将项目根目录添加到 python 路径
 sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..")))
 
+
+_SIDECAR_PROCESS: subprocess.Popen | None = None
+
+
+def _parse_sidecar_target() -> tuple[str, int]:
+    sidecar_url = os.getenv("COPILOT_SIDECAR_URL", "http://127.0.0.1:8787").strip()
+    parsed = urlparse(sidecar_url)
+    host = parsed.hostname or "127.0.0.1"
+    port = parsed.port or 8787
+    return host, port
+
+
+def _stop_sidecar() -> None:
+    global _SIDECAR_PROCESS
+    proc = _SIDECAR_PROCESS
+    if proc is None:
+        return
+    if proc.poll() is None:
+        proc.terminate()
+        try:
+            proc.wait(timeout=5)
+        except subprocess.TimeoutExpired:
+            proc.kill()
+            proc.wait(timeout=3)
+    _SIDECAR_PROCESS = None
+
+
+def _start_sidecar_if_needed() -> None:
+    global _SIDECAR_PROCESS
+    sidecar_dir = os.path.abspath(
+        os.path.join(os.path.dirname(__file__), "..", "copilot-sidecar")
+    )
+
+    host, port = _parse_sidecar_target()
+    cmd = [
+        sys.executable,
+        "-m",
+        "uvicorn",
+        "server:app",
+        "--host",
+        host,
+        "--port",
+        str(port),
+        "--log-level",
+        os.getenv("COPILOT_SIDECAR_LOG_LEVEL", "warning"),
+    ]
+    _SIDECAR_PROCESS = subprocess.Popen(cmd, cwd=sidecar_dir)
+    print(f"[run_server] sidecar started at {host}:{port}.")
+
+
 if __name__ == "__main__":
     # Windows 设置事件循环策略
     if sys.platform == "win32":
         asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())
 
-    # 用 uvicorn.run 支持 workers 参数
-    uvicorn.run(
-        "app.main:app",
-        host="0.0.0.0",
-        port=8000,
-        # workers=2,  # 这里可以设置多进程
-        loop="asyncio",
-    )
+    atexit.register(_stop_sidecar)
+    signal.signal(signal.SIGTERM, lambda *_: _stop_sidecar())
+    signal.signal(signal.SIGINT, lambda *_: _stop_sidecar())
+
+    _start_sidecar_if_needed()
+    try:
+        # 用 uvicorn.run 支持 workers 参数
+        uvicorn.run(
+            "app.main:app",
+            host="0.0.0.0",
+            port=8000,
+            # workers=2,  # 这里可以设置多进程
+            loop="asyncio",
+        )
+    finally:
+        _stop_sidecar()