Add periodic resource alerts

config.example.yaml

@@ -12,6 +12,12 @@ thresholds:
   disk_warn: 80
   load_warn: 2.0
 
+alerts:
+  enabled: true
+  interval_sec: 60
+  cooldown_sec: 900
+  notify_recovery: true
+
 docker:
   # If true, discover containers by name/label
   autodiscovery: true

main.py

@@ -4,6 +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.notify import notify
 import state
 import handlers.menu
@@ -29,6 +30,8 @@ async def main():
     state.DOCKER_MAP.update(await discover_containers(cfg))
     if cfg.get("docker", {}).get("watchdog", True):
         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))
     await notify_start()
     await dp.start_polling(bot)
 

services/alerts.py

@@ -0,0 +1,54 @@
+import asyncio
+import time
+import psutil
+from services.system import worst_disk_usage
+
+
+async def monitor_resources(cfg, notify, bot, chat_id):
+    alerts_cfg = cfg.get("alerts", {})
+    interval = int(alerts_cfg.get("interval_sec", 60))
+    cooldown = int(alerts_cfg.get("cooldown_sec", 900))
+    notify_recovery = bool(alerts_cfg.get("notify_recovery", True))
+
+    disk_warn = int(cfg.get("thresholds", {}).get("disk_warn", 80))
+    load_warn = float(cfg.get("thresholds", {}).get("load_warn", 2.0))
+
+    last_sent = {"disk": 0.0, "load": 0.0, "disk_na": 0.0}
+    state = {"disk_high": False, "load_high": False, "disk_na": False}
+
+    while True:
+        now = time.time()
+
+        usage, mount = worst_disk_usage()
+        if usage is None:
+            if not state["disk_na"] or now - last_sent["disk_na"] >= cooldown:
+                await notify(bot, chat_id, "⚠️ Disk usage n/a")
+                state["disk_na"] = True
+                last_sent["disk_na"] = now
+        else:
+            if state["disk_na"] and notify_recovery:
+                await notify(bot, chat_id, f"🟢 Disk usage OK ({usage}% {mount})")
+            state["disk_na"] = False
+
+            if usage >= disk_warn:
+                if not state["disk_high"] or now - last_sent["disk"] >= cooldown:
+                    await notify(bot, chat_id, f"🟡 Disk usage {usage}% ({mount})")
+                    state["disk_high"] = True
+                    last_sent["disk"] = now
+            else:
+                if state["disk_high"] and notify_recovery:
+                    await notify(bot, chat_id, f"🟢 Disk usage OK ({usage}% {mount})")
+                state["disk_high"] = False
+
+        load = psutil.getloadavg()[0]
+        if load >= load_warn:
+            if not state["load_high"] or now - last_sent["load"] >= cooldown:
+                await notify(bot, chat_id, f"🟡 Load high: {load:.2f}")
+                state["load_high"] = True
+                last_sent["load"] = now
+        else:
+            if state["load_high"] and notify_recovery:
+                await notify(bot, chat_id, f"🟢 Load OK: {load:.2f}")
+            state["load_high"] = False
+
+        await asyncio.sleep(interval)