Add SMART alert scheduler

config.example.yaml

@@ -17,6 +17,10 @@ alerts:
   interval_sec: 60
   cooldown_sec: 900
   notify_recovery: true
+  smart_enabled: true
+  smart_interval_sec: 3600
+  smart_cooldown_sec: 21600
+  smart_temp_warn: 50
 
 docker:
   # If true, discover containers by name/label

handlers/callbacks.py

@@ -150,3 +150,57 @@ async def logs_options(cb: CallbackQuery):
         return
 
     await cb.answer("Bad request")
+
+
+@dp.callback_query(F.data.startswith("wdrestart:"))
+async def watchdog_restart_request(cb: CallbackQuery):
+    if cb.from_user.id != ADMIN_ID:
+        return
+
+    _, alias = cb.data.split(":", 1)
+    if alias not in DOCKER_MAP:
+        await cb.answer("Container not found")
+        return
+
+    kb = InlineKeyboardMarkup(
+        inline_keyboard=[[
+            InlineKeyboardButton(
+                text="✅ Confirm restart",
+                callback_data=f"wdconfirm:{alias}"
+            ),
+            InlineKeyboardButton(
+                text="✖ Cancel",
+                callback_data="wdcancel"
+            ),
+        ]]
+    )
+    await cb.message.answer(
+        f"⚠️ Confirm restart `{alias}`?",
+        reply_markup=kb,
+        parse_mode="Markdown",
+    )
+    await cb.answer()
+
+
+@dp.callback_query(F.data == "wdcancel")
+async def watchdog_restart_cancel(cb: CallbackQuery):
+    await cb.answer("Cancelled")
+
+
+@dp.callback_query(F.data.startswith("wdconfirm:"))
+async def watchdog_restart_confirm(cb: CallbackQuery):
+    if cb.from_user.id != ADMIN_ID:
+        return
+
+    _, alias = cb.data.split(":", 1)
+    real = DOCKER_MAP.get(alias)
+    if not real:
+        await cb.answer("Container not found")
+        return
+
+    await cb.answer("Restarting…")
+    rc, out = await docker_cmd(["restart", real])
+    await cb.message.answer(
+        f"🔄 **{alias} restarted**\n```{out}```",
+        parse_mode="Markdown",
+    )

main.py

@@ -4,7 +4,7 @@ from datetime import datetime
 from app import bot, dp, cfg, ADMIN_ID
 from keyboards import menu_kb
 from services.docker import discover_containers, docker_watchdog
-from services.alerts import monitor_resources
+from services.alerts import monitor_resources, monitor_smart
 from services.notify import notify
 import state
 import handlers.menu
@@ -32,6 +32,8 @@ async def main():
         asyncio.create_task(docker_watchdog(state.DOCKER_MAP, notify, bot, ADMIN_ID))
     if cfg.get("alerts", {}).get("enabled", True):
         asyncio.create_task(monitor_resources(cfg, notify, bot, ADMIN_ID))
+    if cfg.get("alerts", {}).get("smart_enabled", True):
+        asyncio.create_task(monitor_smart(cfg, notify, bot, ADMIN_ID))
     await notify_start()
     await dp.start_polling(bot)
 

services/alerts.py

@@ -1,6 +1,7 @@
 import asyncio
 import time
 import psutil
+from system_checks import list_disks, smart_health, disk_temperature
 from services.system import worst_disk_usage
 
 
@@ -52,3 +53,39 @@ async def monitor_resources(cfg, notify, bot, chat_id):
             state["load_high"] = False
 
         await asyncio.sleep(interval)
+
+
+async def monitor_smart(cfg, notify, bot, chat_id):
+    alerts_cfg = cfg.get("alerts", {})
+    interval = int(alerts_cfg.get("smart_interval_sec", 3600))
+    cooldown = int(alerts_cfg.get("smart_cooldown_sec", 6 * 3600))
+    temp_warn = int(alerts_cfg.get("smart_temp_warn", 50))
+
+    last_sent = {}
+
+    while True:
+        for dev in list_disks():
+            health = smart_health(dev)
+            temp = disk_temperature(dev)
+
+            key = f"{dev}:{health}:{temp}"
+            now = time.time()
+            if last_sent.get(key, 0) + cooldown > now:
+                continue
+
+            if "FAILED" in health:
+                await notify(bot, chat_id, f"🔴 SMART FAIL {dev}: {health}, 🌡 {temp}")
+                last_sent[key] = now
+                continue
+
+            if temp != "n/a":
+                try:
+                    t = int(temp.replace("°C", ""))
+                except ValueError:
+                    t = None
+                if t is not None and t >= temp_warn:
+                    await notify(bot, chat_id, f"🟡 SMART HOT {dev}: {health}, 🌡 {temp}")
+                    last_sent[key] = now
+                    continue
+
+        await asyncio.sleep(interval)

services/docker.py

@@ -1,6 +1,7 @@
 import asyncio
 from datetime import datetime, timezone
 from typing import Dict
+from aiogram.types import InlineKeyboardMarkup, InlineKeyboardButton
 from services.runner import run_cmd
 
 
@@ -109,6 +110,21 @@ async def docker_watchdog(container_map, notify, bot, chat_id):
                 state = "error"
             state = state.strip()
             if last.get(alias) != state:
+                if state != "running":
+                    kb = InlineKeyboardMarkup(
+                        inline_keyboard=[[
+                            InlineKeyboardButton(
+                                text="🔄 Restart",
+                                callback_data=f"wdrestart:{alias}"
+                            )
+                        ]]
+                    )
+                    await bot.send_message(
+                        chat_id,
+                        f"🐳 {alias}: {state}",
+                        reply_markup=kb,
+                    )
+                else:
                     await notify(bot, chat_id, f"🐳 {alias}: {state}")
                 last[alias] = state
         await asyncio.sleep(120)