lego-monitoring/misc/checks.py

from alerting import alerts
from misc import sensors

IS_TESTING = False


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.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 (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.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]