lego-monitoring/src/lego_monitoring/checks/cpu.py

44 lines
1.4 KiB
Python

from socket import gethostname
from psutil import cpu_percent
from lego_monitoring.alerting.alert import Alert
from lego_monitoring.alerting.enum import AlertType, Severity
from lego_monitoring.core import cvars
from .utils import format_for_healthchecks_slug
IS_TESTING = False
def cpu_check() -> list[Alert]:
percentage = cpu_percent()
config = cvars.config.get().checks.cpu
slug = f"{format_for_healthchecks_slug(gethostname())}-cpu"
if config.critical_percentage and (IS_TESTING or percentage >= config.critical_percentage):
return [
Alert(
alert_type=AlertType.CPU,
message=f"CPU load: {percentage:.2f}% >= {config.critical_percentage:.2f}%",
severity=Severity.CRITICAL,
healthchecks_slug=slug,
)
]
elif config.warning_percentage and (IS_TESTING or percentage >= config.warning_percentage):
return [
Alert(
alert_type=AlertType.CPU,
message=f"CPU load: {percentage:.2f}% >= {config.warning_percentage:.2f}%",
severity=Severity.WARNING,
healthchecks_slug=slug,
)
]
else:
return [
Alert(
alert_type=AlertType.CPU,
message=f"CPU load: {percentage:.2f}% (nominal)",
severity=Severity.OK,
healthchecks_slug=slug,
)
]