Merge branch 'ram-cpu-alerts' into 'main'

ram and cpu monitoring

See merge request lego/lego-monitoring!1
This commit is contained in:
Alex Tau 2024-07-27 16:46:39 +00:00
commit 56ebed516e
5 changed files with 106 additions and 9 deletions

View file

@ -11,8 +11,8 @@ from alerting.common import CREDS_FILE, ROOM_ID
class AlertType(StrEnum):
TEST = "TEST"
RAM = "RAM" # TODO
CPU = "CPU" # TODO
RAM = "RAM"
CPU = "CPU"
TEMP = "TEMP"
LOGIN = "LOGIN" # TODO
SMART = "SMART" # TODO

View file

@ -1,25 +1,67 @@
from alerting import alerts
from misc import sensors
IS_TESTING = False
def temp_check() -> set[alerts.Alert]:
def temp_check() -> list[alerts.Alert]:
alert_list = []
temps = sensors.Sensors.get_temperatures()
for _, sensor_list in temps.items():
for sensor in sensor_list:
if sensor.critical_temp is not None and sensor.current_temp > sensor.critical_temp:
if sensor.sensor_type == "nct6687":
continue # little valuable info and too low limits there, might as well ignore it
if sensor.critical_temp is not None and (IS_TESTING or sensor.current_temp > sensor.critical_temp):
alert = alerts.Alert(
alert_type=alerts.AlertType("TEMP"),
message=f"{sensor.sensor_type} {sensor.sensor_label}: {sensor.current_temp}°C > {sensor.critical_temp}°C",
severity=alerts.Severity.CRITICAL,
)
elif sensor.highest_temp is not None and sensor.current_temp > sensor.highest_temp:
elif sensor.highest_temp is not None and (IS_TESTING or sensor.current_temp > sensor.highest_temp):
alert = alerts.Alert(
alert_type=alerts.AlertType("TEMP"),
message=f"{sensor.sensor_type} {sensor.sensor_label}: {sensor.current_temp}°C > {sensor.critical_temp}°C",
message=f"{sensor.sensor_type} {sensor.sensor_label}: {sensor.current_temp}°C > {sensor.highest_temp}°C",
severity=alerts.Severity.WARNING,
)
else:
continue
alert_list.append(alert)
return alert_list
def cpu_check() -> list[alerts.Alert]:
sensor = sensors.Sensors.get_cpu()
if IS_TESTING or sensor.current_load > sensor.critical_load:
alert = alerts.Alert(
alert_type=alerts.AlertType("CPU"),
message=f"{sensor.current_load}% > {sensor.critical_load}%",
severity=alerts.Severity.CRITICAL,
)
elif IS_TESTING or sensor.current_load > sensor.highest_load:
alert = alerts.Alert(
alert_type=alerts.AlertType("CPU"),
message=f"{sensor.current_load}% > {sensor.highest_load}%",
severity=alerts.Severity.WARNING,
)
else:
return []
return [alert]
def ram_check() -> list[alerts.Alert]:
sensor = sensors.Sensors.get_ram()
if IS_TESTING or sensor.current_avail < sensor.critical_avail:
alert = alerts.Alert(
alert_type=alerts.AlertType("RAM"),
message=f"{(sensor.current_avail / 1024**3):.2f} GiB < {(sensor.critical_avail / 1024**3):.2f} GiB",
severity=alerts.Severity.CRITICAL,
)
elif IS_TESTING or sensor.current_avail < sensor.warning_avail:
alert = alerts.Alert(
alert_type=alerts.AlertType("RAM"),
message=f"{(sensor.current_avail / 1024**3):.2f} GiB < {(sensor.warning_avail / 1024**3):.2f} GiB",
severity=alerts.Severity.WARNING,
)
else:
return []
return [alert]

View file

@ -1,6 +1,6 @@
from dataclasses import dataclass
from psutil import sensors_temperatures
from psutil import cpu_percent, sensors_temperatures, virtual_memory
@dataclass
@ -12,6 +12,21 @@ class TemperatureSensor:
critical_temp: float | None = None
@dataclass
class CpuSensor:
current_load: float
highest_load: float = 90
critical_load: float = 95
@dataclass
class RamSensor:
current_avail: int
current_avail_percentage: float
warning_avail: int = 4 * 1024**3
critical_avail: int = 2 * 1024**3
class Sensors:
@staticmethod
def get_temperatures() -> dict[str, list[TemperatureSensor]]:
@ -51,9 +66,39 @@ class Sensors:
critical_temp=95.0, # hardcoded because we have R9 7900X
)
)
case "nct6687":
lables = {
"AMD TSI Addr 98h": "CPU",
"Diode 0 (curr)": "System",
"Thermistor 15": "VRM MOSFET",
"Thermistor 1": "Platform Controller Hub (Peripherals)",
"Thermistor 16": "CPU Socket",
}
for sensor in sensors[:-2]:
real_label = lables[sensor.label]
temp_sensors[s_type].append(
TemperatureSensor(
sensor_type=s_type,
sensor_label=real_label,
current_temp=sensor.current,
highest_temp=sensor.high or None,
critical_temp=sensor.critical or None,
)
)
return temp_sensors
@staticmethod
def get_cpu() -> CpuSensor:
return CpuSensor(current_load=cpu_percent())
@staticmethod
def get_ram() -> RamSensor:
ram = virtual_memory()
return RamSensor(current_avail=ram.available,
current_avail_percentage=ram.percent)
if __name__ == "__main__":
for i in Sensors.get_temperatures():

View file

@ -1,6 +1,6 @@
#!/usr/bin/env python3
from colorama import Back, Fore, Style
from colorama import Back, Style
from misc.sensors import Sensors
@ -14,6 +14,12 @@ def pretty_print():
for sensors in v:
print(f"{sensors.sensor_label}: {sensors.current_temp}°C")
print()
s = Sensors.get_cpu()
print(f"Used CPU: {s.current_load}%")
s = Sensors.get_ram()
print(f"Available RAM: {(s.current_avail / 1024**3):.2f} ({s.current_avail_percentage}%) GiB")
if __name__ == "__main__":
pretty_print()

View file

@ -39,7 +39,11 @@ async def checker(check: Callable | Coroutine, interval_secs: int, client: nio.A
async def main():
signal.signal(signal.SIGTERM, stop_gracefully)
client = await alerts.get_client()
checkers = (checker(checks.temp_check, 5 * 60, client),)
checkers = (
checker(checks.temp_check, 5 * 60, client),
checker(checks.cpu_check, 5 * 60, client),
checker(checks.ram_check, 1 * 60, client),
)
async with asyncio.TaskGroup() as tg:
checker_tasks: set[asyncio.Task] = set()
for c in checkers: