add temp monitoring

src/lego_monitoring/__init__.py

@@ -1,12 +1,16 @@
 import argparse
 import asyncio
+import datetime
 import logging
 import signal
 import time
 
+from . import checks
 from .alerting import alerts
+from .checks.temp.sensors import print_readings
+from .config import load_config
 from .core import cvars
-from .core.config import load_config
+from .core.checkers import interval_checker
 
 stopping = False
 
@@ -27,11 +31,21 @@ async def async_main():
         prog="lego-monitoring",
         description="Lego-monitoring service",
     )
-    parser.add_argument("-c", "--config", required=True)
+    parser.add_argument("-c", "--config", help="config file")
+    parser.add_argument("--print-temp", help="print temp sensor readings and exit", action="store_true")
+    args = parser.parse_args()
 
-    config_path = parser.parse_args().config
-    config = load_config(config_path)
-    cvars.config.set(config)
+    if args.config:
+        config_path = parser.parse_args().config
+        config = load_config(config_path)
+        cvars.config.set(config)
+
+    if args.print_temp:
+        print_readings()
+        raise SystemExit
+
+    if not args.config:
+        raise RuntimeError("--config must be specified in standard operating mode")
 
     tg_client = await alerts.get_client()
     cvars.tg_client.set(tg_client)
@@ -41,10 +55,11 @@ async def async_main():
             alerts.send_start_alert(),
         ],
         "stop": [],  # this is checked later
+        "temp": [interval_checker(checks.temp_check, datetime.timedelta(minutes=5))],
     }
 
     checkers = []
-    for enabled_set in config.enabled_checker_sets:
+    for enabled_set in config.enabled_check_sets:
         for checker in checker_sets[enabled_set]:
             checkers.append(checker)
 
@@ -57,7 +72,7 @@ async def async_main():
             checker_tasks.add(task)
         while True:
             if stopping:
-                if "stop" in config.enabled_checker_sets:
+                if "stop" in config.enabled_check_sets:
                     await alerts.send_stop_alert()
                 await tg_client.disconnect()
                 raise SystemExit

src/lego_monitoring/alerting/alerts.py

@@ -57,7 +57,7 @@ async def send_start_alert() -> None:
     await send_alert(
         Alert(
             alert_type=AlertType.BOOT,
-            message=f"Service running with enabled checkers: {', '.join(config.enabled_checker_sets)}",
+            message=f"Service running with enabled checks: {', '.join(config.enabled_check_sets)}",
             severity=Severity.INFO,
         )
     )

src/lego_monitoring/alerting/enum.py

@@ -3,11 +3,11 @@ from enum import StrEnum
 
 class AlertType(StrEnum):
     BOOT = "BOOT"
+    TEMP = "TEMP"
     TEST = "TEST"
     # ERROR = "ERROR"
     # RAM = "RAM"
     # CPU = "CPU"
-    # TEMP = "TEMP"
     # VULN = "VULN"
     # LOGIN = "LOGIN"
     # SMART = "SMART"  # TODO

src/lego_monitoring/checks/__init__.py

@@ -0,0 +1 @@
+from .temp import temp_check

src/lego_monitoring/checks/temp/__init__.py

@@ -0,0 +1,29 @@
+from lego_monitoring.alerting import alerts
+from lego_monitoring.alerting.enum import AlertType, Severity
+
+from . import sensors
+
+IS_TESTING = False
+
+
+def temp_check() -> list[alerts.Alert]:
+    alert_list = []
+    temps = sensors.get_readings()
+    for sensor, readings in temps.items():
+        for r in readings:
+            if r.critical_temp is not None and (IS_TESTING or r.current_temp > r.critical_temp):
+                alert = alerts.Alert(
+                    alert_type=AlertType.TEMP,
+                    message=f"{sensor} {r.label}: {r.current_temp}°C > {r.critical_temp}°C",
+                    severity=Severity.CRITICAL,
+                )
+            elif r.warning_temp is not None and (IS_TESTING or r.current_temp > r.warning_temp):
+                alert = alerts.Alert(
+                    alert_type=AlertType.TEMP,
+                    message=f"{sensor} {r.label}: {r.current_temp}°C > {r.warning_temp}°C",
+                    severity=Severity.WARNING,
+                )
+            else:
+                continue
+            alert_list.append(alert)
+    return alert_list

src/lego_monitoring/checks/temp/sensors.py

@@ -0,0 +1,66 @@
+from dataclasses import dataclass
+from typing import Optional
+
+from psutil import sensors_temperatures
+
+from lego_monitoring.config.checks.temp import TempSensorConfig
+from lego_monitoring.core import cvars
+
+
+@dataclass
+class TemperatureReading:
+    label: str
+    current_temp: float
+    warning_temp: Optional[float]
+    critical_temp: Optional[float]
+
+
+def print_readings():
+    sensor_readings = get_readings()
+    for sensor, readings in sensor_readings.items():
+        print(f"*** Sensor {sensor}***\n")
+        for r in readings:
+            print(f"Label: {r.label}")
+            print(f"Current temp: {r.current_temp}")
+            print(f"Warning temp: {r.warning_temp}")
+            print(f"Critical temp: {r.critical_temp}\n")
+
+
+def get_readings() -> dict[str, list[TemperatureReading]]:
+    try:
+        config = cvars.config.get().checks.temp.sensors
+    except LookupError:
+        config: dict[str, TempSensorConfig] = {}
+
+    psutil_temperatures = sensors_temperatures()
+
+    sensor_readings = {}
+    for sensor, readings in psutil_temperatures.items():
+        if sensor in config:
+            if not config[sensor].enabled:
+                continue
+            sensor_friendly_name = config[sensor].name if config[sensor].name else sensor
+        else:
+            sensor_friendly_name = sensor
+
+        sensor_readings[sensor_friendly_name] = []
+
+        for r in readings:
+            try:
+                config_r = config[sensor].readings[r.label]
+            except KeyError:
+                friendly_r = TemperatureReading(
+                    label=r.label, current_temp=r.current, warning_temp=r.high, critical_temp=r.critical
+                )
+            else:
+                if not config_r.enabled:
+                    continue
+                friendly_r = TemperatureReading(
+                    label=config_r.label if config_r.label else r.label,
+                    current_temp=r.current,
+                    warning_temp=config_r.warning_temp if config_r.warning_temp else r.high,
+                    critical_temp=config_r.critical_temp if config_r.critical_temp else r.critical,
+                )
+            sensor_readings[sensor_friendly_name].append(friendly_r)
+
+    return sensor_readings

src/lego_monitoring/config/__init__.py

@@ -3,6 +3,13 @@ from dataclasses import dataclass
 
 from alt_utils import NestedDeserializableDataclass
 
+from .checks.temp import TempCheckConfig
+
+
+@dataclass
+class ChecksConfig(NestedDeserializableDataclass):
+    temp: TempCheckConfig
+
 
 @dataclass
 class TelegramConfig:
@@ -12,7 +19,8 @@ class TelegramConfig:
 
 @dataclass
 class Config(NestedDeserializableDataclass):
-    enabled_checker_sets: list[str]
+    enabled_check_sets: list[str]
+    checks: ChecksConfig
     telegram: TelegramConfig
 
 

src/lego_monitoring/config/checks/temp.py

@@ -0,0 +1,24 @@
+from dataclasses import dataclass
+from typing import Optional
+
+from alt_utils import NestedDeserializableDataclass
+
+
+@dataclass
+class TempReadingConfig:
+    label: Optional[str]
+    enabled: bool
+    warning_temp: Optional[float]
+    critical_temp: Optional[float]
+
+
+@dataclass
+class TempSensorConfig(NestedDeserializableDataclass):
+    name: Optional[str]
+    enabled: bool
+    readings: dict[str, TempReadingConfig]
+
+
+@dataclass
+class TempCheckConfig(NestedDeserializableDataclass):
+    sensors: dict[str, TempSensorConfig]

src/lego_monitoring/core/cvars.py

@@ -2,7 +2,7 @@ from contextvars import ContextVar
 
 from telethon import TelegramClient
 
-from .config import Config
+from ..config import Config
 
 config: ContextVar[Config] = ContextVar("config")
 tg_client: ContextVar[TelegramClient] = ContextVar("tg_client")