Update app.py

app.py

@@ -1,11 +1,13 @@
 import os
 import time
+import threading
+import requests
 import gradio as gr
 import pandas as pd
 import matplotlib.pyplot as plt
 import base64
 from io import BytesIO
-import requests
+from datetime import datetime
 
 from core.dashboard import ErohaDashboard
 from core.intelligence import update_memory, summarize_context
@@ -13,66 +15,154 @@ from core.selfcheck import evaluate_answer, improve_answer
 from core.learning import analyze_user_input, adapt_answer
 from core.model_selector import choose_model
 
+
 # ==============================
-# 🔐 ENV
+# 🔐 ENVIRONMENT CONFIG
 # ==============================
 HF_TOKEN = os.getenv("HF_TOKEN", "")
 LOG_FILE = "logs/history.csv"
 os.makedirs("logs", exist_ok=True)
 dashboard = ErohaDashboard()
 
+
 # ==============================
-# 🔁 Автоматический Router с Failover
+# 🧠 Router Manager
 # ==============================
-def query_huggingface_router(prompt, model_id):
-    """
-    Универсальный запрос к Router API с fallback на прямую модель.
-    """
-    headers = {"Authorization": f"Bearer {HF_TOKEN}"}
-    payload = {
-        "inputs": prompt,
-        "parameters": {"max_new_tokens": 600, "temperature": 0.7},
-    }
-
-    try:
-        # 1️⃣ Основной Router API
-        response = requests.post(
-            "https://router.huggingface.co/models/" + model_id,
-            headers=headers,
-            json=payload,
-            timeout=30,
-        )
-        if response.status_code == 200:
-            return response.json()[0]["generated_text"]
-        elif response.status_code == 410:
-            raise RuntimeError("Old endpoint rejected (410)")
-        else:
-            raise RuntimeError(f"Router error: {response.status_code}")
+class RouterManager:
+    def __init__(self, token: str):
+        self.token = token
+        self.router_url = "https://router.huggingface.co"
+        self.headers = {"Authorization": f"Bearer {self.token}"}
+        self.is_router_healthy = True
+        self.last_check = None
+        self.reconnect_attempts = 0
+        self.last_token_validation = 0
+
+    def validate_token(self):
+        """Проверяет, действителен ли токен"""
+        try:
+            r = requests.get(self.router_url + "/status", headers=self.headers, timeout=10)
+            if r.status_code == 200:
+                return True
+            elif r.status_code == 401:
+                print("🔒 Ошибка: недействительный токен Hugging Face!")
+                self.is_router_healthy = False
+                return False
+            return False
+        except Exception:
+            return False
+
+    def check_router_health(self):
+        """Проверяет доступность Router API"""
+        try:
+            start = time.time()
+            r = requests.post(self.router_url, headers=self.headers, json={"inputs": "ping"})
+            latency = round(time.time() - start, 2)
+            if r.status_code == 200:
+                self.is_router_healthy = True
+                self.last_check = datetime.now()
+                self.reconnect_attempts = 0
+                print(f"✅ Router API активен ({latency}s)")
+            else:
+                self.is_router_healthy = False
+                print(f"⚠️ Router API ответил {r.status_code}")
+        except Exception as e:
+            self.is_router_healthy = False
+            print("❌ Router недоступен:", str(e))
+
+    def auto_reconnect(self):
+        """Автоматическое восстановление при потере связи"""
+        self.reconnect_attempts += 1
+        delay = min(30, 2 ** self.reconnect_attempts)
+        print(f"🔁 Попытка переподключения через {delay}s...")
+        time.sleep(delay)
+        self.check_router_health()
+
+    def request(self, model_id: str, prompt: str):
+        """Отправляет запрос с failover и auto-heal"""
+        payload = {
+            "model": model_id,
+            "inputs": prompt,
+            "parameters": {"max_new_tokens": 600, "temperature": 0.7},
+        }
 
-    except Exception as e:
-        print("⚠️ Router недоступен, выполняется fallback:", str(e))
-        # 2️⃣ Fallback: прямое обращение к конкретной модели
-        fallback_url = f"https://huggingface.co/api/models/{model_id}"
         try:
-            response = requests.post(fallback_url, headers=headers, json=payload, timeout=30)
+            response = requests.post(
+                self.router_url,
+                headers=self.headers,
+                json=payload,
+                timeout=60,
+            )
             if response.status_code == 200:
-                return response.json()[0]["generated_text"]
+                data = response.json()
+                if isinstance(data, list) and "generated_text" in data[0]:
+                    return data[0]["generated_text"]
+                elif isinstance(data, dict) and "generated_text" in data:
+                    return data["generated_text"]
+                else:
+                    raise RuntimeError("Неверный формат ответа")
+            elif response.status_code >= 500:
+                print("⚠️ Ошибка Router 5xx. Переподключение...")
+                self.auto_reconnect()
+                return self.request(model_id, prompt)
             else:
-                raise RuntimeError(f"Fallback error: {response.status_code}")
-        except Exception as e2:
-            return f"❌ Ошибка при обращении к API: {str(e2)}"
+                raise RuntimeError(f"Router ошибка: {response.status_code}")
+
+        except Exception as e:
+            print(f"❌ Router ошибка: {e}, активируем fallback...")
+            return self.fallback_request(prompt)
+
+    def fallback_request(self, prompt):
+        """Fallback на стабильную модель"""
+        fallback_model = "mistralai/Mixtral-8x7B-Instruct-v0.1"
+        try:
+            response = requests.post(
+                self.router_url,
+                headers=self.headers,
+                json={"model": fallback_model, "inputs": prompt},
+                timeout=60,
+            )
+            if response.status_code == 200:
+                data = response.json()
+                if isinstance(data, list) and "generated_text" in data[0]:
+                    return f"[⚠️ Fallback {fallback_model}] {data[0]['generated_text']}"
+            raise RuntimeError(f"Fallback ошибка: {response.status_code}")
+        except Exception as e:
+            return f"❌ Ошибка fallback: {e}"
+
+
+# ==============================
+# 🐕 Watchdog Thread
+# ==============================
+def watchdog_loop(router_manager: RouterManager):
+    """Поток, который следит за Router API и токеном"""
+    while True:
+        try:
+            router_manager.check_router_health()
+            now = time.time()
+            # Проверка токена каждые 10 минут
+            if now - router_manager.last_token_validation > 600:
+                router_manager.validate_token()
+                router_manager.last_token_validation = now
+            time.sleep(60)
+        except Exception as e:
+            print("⚠️ Watchdog ошибка:", str(e))
+            time.sleep(30)
+
 
 # ==============================
-# 🧠 Основная логика
+# 🧠 Основная генерация
 # ==============================
+router_manager = RouterManager(HF_TOKEN)
+
+
 def generate_response(user_input):
     try:
         start = time.time()
         prefs = analyze_user_input(user_input)
         model_id = choose_model(user_input)
 
-        result = query_huggingface_router(user_input, model_id)
-
+        result = router_manager.request(model_id, user_input)
         check = evaluate_answer(result)
         improved = improve_answer(result)
         personalized = adapt_answer(improved)
@@ -105,8 +195,9 @@ def generate_response(user_input):
     except Exception as e:
         return f"❌ Ошибка выполнения: {str(e)}"
 
+
 # ==============================
-# 📈 Визуализация аналитики
+# 📈 Аналитика
 # ==============================
 def generate_chart():
     if not os.path.exists(LOG_FILE):
@@ -130,20 +221,22 @@ def generate_chart():
     plt.close()
     return f"<img src='data:image/png;base64,{img_base64}'/>"
 
+
 def show_dashboard():
     metrics_text, df = dashboard.dashboard_ui()
     chart_html = generate_chart()
     return metrics_text, df, chart_html
 
+
 # ==============================
-# 🎨 Интерфейс Gradio
+# 🎨 Gradio UI
 # ==============================
-with gr.Blocks(title="Eroha AgentAPI v5.6 — AutoRouter Failover Core", theme="soft") as app:
-    gr.Markdown("# 🤖 Eroha AgentAPI v5.6 — Guru Edition (AutoRouter + Failover Core)")
-    gr.Markdown("**Автоматический Router + fallback + визуальная аналитика** ⚙️")
+with gr.Blocks(title="Eroha AgentAPI v5.8 — Production-Cluster Core", theme="soft") as app:
+    gr.Markdown("# 🤖 Eroha AgentAPI v5.8 — Production-Cluster Core 🚀")
+    gr.Markdown("**Полный контроль: Router-watchdog + auto-heal + token-validation + fallback.**")
 
     with gr.Tab("💬 Agent Chat"):
-        user_input = gr.Textbox(label="Введите запрос", placeholder="Например: придумай философскую сказку об ИИ.")
+        user_input = gr.Textbox(label="Введите запрос", placeholder="Например: придумай легенду о ИИ, который нашёл сознание.")
         output_box = gr.Textbox(label="Ответ", lines=15)
         submit_btn = gr.Button("🚀 Отправить")
         submit_btn.click(fn=generate_response, inputs=user_input, outputs=output_box)
@@ -155,4 +248,7 @@ with gr.Blocks(title="Eroha AgentAPI v5.6 — AutoRouter Failover Core", theme="
         refresh = gr.Button("🔄 Обновить дашборд")
         refresh.click(show_dashboard, outputs=[metrics, log_table, chart_box])
 
+# 🔥 Запуск watchdog-потока
+threading.Thread(target=watchdog_loop, args=(router_manager,), daemon=True).start()
+
 app.launch(server_name="0.0.0.0", server_port=7860)