Add dedicated RAID alert category and monitor

services/alerts.py

@@ -1,7 +1,7 @@
 import asyncio
 import time
 import psutil
-from system_checks import list_disks, smart_health, disk_temperature
+from system_checks import list_disks, smart_health, disk_temperature, list_md_arrays, md_array_status
 from services.system import worst_disk_usage
 from services.disk_report import build_disk_report
 
@@ -130,3 +130,54 @@ async def monitor_smart(cfg, notify, bot, chat_id):
                     continue
 
         await asyncio.sleep(interval)
+
+
+async def monitor_raid(cfg, notify, bot, chat_id):
+    alerts_cfg = cfg.get("alerts", {})
+    interval = int(alerts_cfg.get("raid_interval_sec", 300))
+    cooldown = int(alerts_cfg.get("raid_cooldown_sec", 1800))
+    notify_recovery = bool(alerts_cfg.get("notify_recovery", True))
+
+    last_sent: dict[str, float] = {}
+    bad_state: dict[str, bool] = {}
+
+    while True:
+        now = time.time()
+        for dev in list_md_arrays():
+            status = md_array_status(dev)
+            lower = status.lower()
+            level = None
+            key_suffix = None
+            if "inactive" in lower:
+                level = "critical"
+                key_suffix = "inactive"
+            elif "degraded" in lower:
+                level = "warn"
+                key_suffix = "degraded"
+
+            if level:
+                if not bad_state.get(dev) or (now - last_sent.get(dev, 0.0) >= cooldown):
+                    icon = "🔴" if level == "critical" else "🟡"
+                    await notify(
+                        bot,
+                        chat_id,
+                        f"{icon} RAID {dev}: {status}",
+                        level=level,
+                        key=f"raid_{key_suffix}:{dev}",
+                        category="raid",
+                    )
+                    last_sent[dev] = now
+                bad_state[dev] = True
+            else:
+                if bad_state.get(dev) and notify_recovery:
+                    await notify(
+                        bot,
+                        chat_id,
+                        f"🟢 RAID {dev}: {status}",
+                        level="info",
+                        key=f"raid_ok:{dev}",
+                        category="raid",
+                    )
+                bad_state[dev] = False
+
+        await asyncio.sleep(interval)