icinga2-checks/check_systemd_timer.py

#!/usr/bin/env python3
import argparse
import re
import subprocess
import sys
import traceback
from datetime import datetime

import humanfriendly
from dateutil import tz

from checker import nagios
from checker.humanfriendly import parse_systemctl_time_delta
from checker.result import quit_check

sys.path.insert(0, "/usr/lib/python3/dist-packages")
import dbus

SYSTEMCTL_TIMERS_RE = re.compile(
    r'^(([A-Za-z]*\s[0-9]{4}-[0-9]{2}-[0-9]{2}\s*[0-9]{2}:[0-9]{2}:[0-9]{2}\s[A-Z]*)|n\/a|-)\s*((([0-9]*[a-z]*\s)*left)|n\/a|-)\s*(([A-Za-z]*\s[0-9]{4}-[0-9]{2}-[0-9]{2}\s[0-9]{2}:[0-9]{2}:[0-9]{2}\s[A-Z]*)|n\/a|-)\s*(([0-9A-Za-z\s]*\sago)|n\/a|-)\s*([A-Za-z\-_]*.timer)\s*(([A-Za-z\-_]*.service)|\s*)')


def get_next_elapse(timer_name):
    try:
        output = subprocess.check_output(["systemctl", "list-timers", "--all"], universal_newlines=True)
        lines = output.split('\n')
        for line in lines:
            if timer_name in line:
                try:
                    parts = re.search(SYSTEMCTL_TIMERS_RE, line)

                    datetime_object = None
                    if parts.group(2):
                        try:
                            datetime_object = datetime.strptime(parts.group(2), '%a %Y-%m-%d %H:%M:%S %Z')
                        except ValueError as e:
                            return None, e

                    time_left = 'n/a'
                    if parts.group(4):
                        time_left = parse_systemctl_time_delta(parts.group(4))
                        if isinstance(time_left, humanfriendly.InvalidTimespan):
                            return None, humanfriendly.InvalidTimespan

                    time_passed = 'n/a'
                    if parts.group(9):
                        time_passed = parse_systemctl_time_delta(parts.group(9))
                        if isinstance(time_passed, humanfriendly.InvalidTimespan):
                            return None, humanfriendly.InvalidTimespan

                    timer_info = {
                        'next': datetime_object,
                        'left': time_left,
                        'last': parts.group(7),
                        'passed': time_passed,
                        'unit': parts.group(10),
                        'activates': parts.group(12)
                    }
                    return timer_info, None
                except Exception:
                    print(output)
                    traceback.print_exc()
                    sys.exit(nagios.STATE_UNKNOWN)
        return None, ValueError('Timer not found')
    except subprocess.CalledProcessError as e:
        return None, e


def check_timer(timer_name):
    if not timer_name.endswith('.timer'):
        timer_name = timer_name + '.timer'

    try:
        system_bus = dbus.SystemBus()
        systemd1 = system_bus.get_object('org.freedesktop.systemd1', '/org/freedesktop/systemd1')
        manager = dbus.Interface(systemd1, 'org.freedesktop.systemd1.Manager')
        timer_unit_path = manager.GetUnit(timer_name)
        timer_unit = system_bus.get_object('org.freedesktop.systemd1', timer_unit_path)
        timer_properties = dbus.Interface(timer_unit, 'org.freedesktop.DBus.Properties')
        active_state = timer_properties.Get('org.freedesktop.systemd1.Unit', 'ActiveState')
        if active_state == 'active':
            next_elapse, err = get_next_elapse(timer_name)
            if err:
                quit_check(f'{err}', nagios.STATE_UNKNOWN)

            if (next_elapse['left'] != 'n/a' and next_elapse['passed'] != 'n/a') and (next_elapse['left'] < 0 or next_elapse['passed'] < 0):
                quit_check(f'Timer is negative??? Left: {next_elapse["left"]}. Passed: {next_elapse["passed"]}', nagios.STATE_UNKNOWN)

            if next_elapse['next']:
                local_tz = tz.tzlocal()
                next_elapse_str = next_elapse['next'].replace(tzinfo=local_tz).strftime('%a %Y-%m-%d %H:%M %Z')
            else:
                next_elapse_str = 'n/a'

            if next_elapse['left'] != 'n/a':
                remaining_time_human = humanfriendly.format_timespan(next_elapse['left'])
            else:
                remaining_time_human = 'n/a'
            if next_elapse['passed'] != 'n/a':
                passed_time_human = humanfriendly.format_timespan(next_elapse['passed'])
            else:
                passed_time_human = 'n/a'

            perfdata_dict = {
                'remaining_time': {
                    'value': int(next_elapse['left']) if next_elapse['left'] != 'n/a' else -1,
                    'unit': 's',
                    'min': 0
                },
                'passed_time': {
                    'value': int(next_elapse['passed']) if next_elapse['passed'] != 'n/a' else -1,
                    'unit': 's',
                    'min': 0
                }
            }
            quit_check(f'{timer_name} is active. Trigger time: {next_elapse_str}. Remaining time: {remaining_time_human}. Time since last trigger: {passed_time_human}.', nagios.STATE_OK, perfdata_dict)
        else:
            quit_check(f'{timer_name} is not active.', nagios.STATE_CRIT)
    except dbus.exceptions.DBusException:
        quit_check(f'{timer_name} could not be found.', nagios.STATE_UNKNOWN)


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('-t', '--timer', required=True, help='The name of the timer to check.')
    parser.add_argument('-l', '--last-ran-delta', help='The associated service should have been triggered at least this many seconds ago.')
    args = parser.parse_args()

    try:
        check_timer(args.timer)
    except Exception as e:
        print(f'UNKNOWN - exception "{e}"')
        traceback.print_exc()
        sys.exit(nagios.STATE_UNKNOWN)