webhookの文字化け対策

oproxy.py

@@ -1,5 +1,5 @@
 # バージョン情報: Python 3.12+ / FastAPI 0.115.0 / uvicorn 0.30.0 / httpx 0.28.0
-# [2026-02-14] 3060(12GB)戦術支援：Mattermost Webhook連携・遅延送信実装版
+# [2026-02-14] 3060(12GB)戦術支援：Mattermost連携・文字化け完全対策・全ロジック非短縮版
 import argparse
 import asyncio
 import json
@@ -53,40 +53,44 @@ def pulse(char, color=C_RESET):
     print(f"{color}{char}{C_RESET}", end="", flush=True)
 
 
-# --- 🛰️ Mattermost 送信エンジン (非同期・遅延送信) ---
+# --- 🛰️ Mattermost 送信エンジン (文字化け対策済み) ---
 async def post_to_mattermost(model_name, path, req_body, resp_text):
-    """通信完了後にバックグラウンドでMattermostへログを飛ばす"""
+    """通信完了後にバックグラウンドでMattermostへログを飛ばす。生日本語を保持する。"""
     if not CONFIG["webhook_url"] or not CONFIG["enable_webhook"]:
         return
     try:
-        req_obj = json.loads(req_body)
-        req_clean = json.dumps(req_obj, indent=2, ensure_ascii=False)
-    except:
-        req_clean = req_body
+        # リクエストのデコード（生日本語を維持）
+        try:
+            req_obj = json.loads(req_body)
+            req_clean = json.dumps(req_obj, indent=2, ensure_ascii=False)
+        except:
+            req_clean = req_body.decode('utf-8', errors='replace')
 
-    resp_display = (
-        resp_text if len(resp_text) < CONFIG["max_log_len"] 
-        else resp_text[:CONFIG["max_log_len"]] + "\n\n*(Truncated due to length)*"
-    )
-
-    payload = {
-        "username": f"oproxy:{model_name}",
-        "icon_url": "https://ollama.com/public/ollama.png",
-        "text": (
-            f"### 🛡️ LLM Traffic Log: `{path}`\n"
-            f"- **Model:** `{model_name}` | **Time:** {datetime.now().strftime('%H:%M:%S')}\n"
-            f"#### 📥 Request\n```json\n{req_clean[:800]}\n```\n"
-            f"#### 📤 Response\n{resp_display}"
+        # レスポンスのカットオフ
+        resp_display = (
+            resp_text if len(resp_text) < CONFIG["max_log_len"] 
+            else resp_text[:CONFIG["max_log_len"]] + "\n\n*(長文のため中略...)*"
         )
-    }
-    try:
-        async with httpx.AsyncClient(timeout=10.0) as client:
-            await client.post(CONFIG["webhook_url"], json=payload)
+
+        payload = {
+            "username": f"oproxy:{model_name}",
+            "icon_url": "https://ollama.com/public/ollama.png",
+            "text": (
+                f"### 🛡️ LLM Traffic Log: `{path}`\n"
+                f"- **Model:** `{model_name}` | **Time:** {datetime.now().strftime('%H:%M:%S')}\n"
+                f"#### 📥 Request preview\n```json\n{req_clean[:800]}\n```\n"
+                f"#### 📤 Response\n{resp_display}"
+            )
+        }
+        async with httpx.AsyncClient(timeout=15.0) as client:
+            # ensure_ascii=False で Mattermost に生の日本語を届ける
+            encoded_payload = json.dumps(payload, ensure_ascii=False).encode('utf-8')
+            await client.post(CONFIG["webhook_url"], content=encoded_payload, headers={"Content-Type": "application/json"})
     except Exception as e:
         print(f"\n{C_RED}[Webhook Fail] {e}{C_RESET}")
 
 
-# --- 🧠 戦略적モデル分析 (2行分割レイアウト) ---
+# --- 🧠 戦略的モデル分析 (2行分割レイアウト・非短縮版) ---
 async def fetch_detailed_models():
     """Ollamaの内部情報を深掘りし、メタデータを完全取得する"""
     try:
@@ -230,7 +234,7 @@ async def sticky_proxy(path: str, request: Request):
     print(f"{get_ts()} {C_WHITE}/{path: <10}{C_RESET} ", end="", flush=True)
     body = await request.body()
     
-    # モデル名の特定
+    # 送信時にどのモデルか判別するための抽出
     model_name = "unknown"
     try:
         model_name = json.loads(body).get("model", "unknown")
@@ -239,7 +243,7 @@ async def sticky_proxy(path: str, request: Request):
 
     if do_dump:
         print(
-            f"\n{C_YELLOW}{'=' * 60}\n[DUMP REQUEST: {request.method} /{path}]\n{'=' * 60}{C_RESET}"
+            f"\n{C_YELLOW}{'=' * 60}\n[DUMP REQUEST: {model_name}]\n{'=' * 60}{C_RESET}"
         )
         try:
             print(json.dumps(json.loads(body), indent=2, ensure_ascii=False))
@@ -270,16 +274,19 @@ async def sticky_proxy(path: str, request: Request):
                     
                     full_text_buffer = []
                     async for chunk in response.aiter_bytes():
+                        # チャンクから正規表現で高速テキスト抽出
                         raw_data = chunk.decode(errors="ignore")
                         matches = RE_CONTENT.findall(raw_data)
                         for m in matches:
                             try:
-                                text = m.encode().decode("unicode_escape", errors="ignore")
+                                # 【最重要】Unicodeエスケープを戻し、化けないようにUTF-8で再構成
+                                text = m.encode('utf-8').decode('unicode_escape').encode('latin1').decode('utf-8')
                                 full_text_buffer.append(text)
                                 if do_dump:
                                     print(f"{C_WHITE}{text}{C_RESET}", end="", flush=True)
                             except:
-                                pass
+                                # 万が一デコードに失敗した場合はマッチした文字列をそのまま保持
+                                full_text_buffer.append(m)
 
                         pulse("v", C_GREEN)
                         yield chunk
@@ -288,7 +295,7 @@ async def sticky_proxy(path: str, request: Request):
                     if full_text_buffer:
                         combined_text = "".join(full_text_buffer)
                         asyncio.create_task(post_to_mattermost(
-                            model_name, path, body.decode(errors='ignore'), combined_text
+                            model_name, path, body, combined_text
                         ))
 
                     if do_dump: