check_systemd_service

#!/usr/bin/env python3

#
# This plugin is designed as a nagios compatible check plugin to use with
# Icinga 2 and others
#

import argparse
import sys
import subprocess
from datetime import datetime, timedelta


class Service(object):
    def __init__(self, service):
        self.name = service
        self._info = self._get_properties(service)

    @staticmethod
    def _get_properties(service):
        info = {}
        output = subprocess.check_output(['systemctl', 'show', service])

        for line in output.decode('utf-8').split('\n'):
            if not line.strip():
                continue

            key, value = line.split('=', 1)
            if value and key.endswith('Timestamp') and value != 'n/a':
                value = datetime.strptime(value, '%a %Y-%m-%d %H:%M:%S %Z')
            info[key] = value

        return info

    def get_property(self, name):
        return self._info.get(name, None)

    def is_enabled(self):
        enabled_states = [
            'enabled',
            'enabled-runtime',
            'static',
            'indirect',
            'generated',
            'transient'
        ]
        if self.get_property('UnitFileState') in enabled_states:
            return True
        return False

    def is_running(self):
        if self.get_property('ActiveState') in ['active', 'activating']:
            return True
        return False

    def is_failed(self):
        if self.get_property('ActiveState') == 'failed':
            return True
        return False

def get_boot_time():
    output = subprocess.check_output(['cat', '/proc/uptime'])
    s = float(output.decode('utf-8').split()[0])
    return datetime.now() - timedelta(seconds=s)

def get_perfdata(label, value, warning, critical, minimum=0):
    return '%s=%ss;%s;%s;%s' % (label, value, warning, critical, minimum)

def get_status_msg(retval):
    if retval == 0:
        msg = 'OK'
    elif retval == 1:
        msg = 'Warning'
    elif retval == 2:
        msg = 'Critical'
    elif retval == 3:
        msg = 'Unknown'
    else:
        raise ValueError('Invalid retval %s' % retval)

    return msg

def check_status(service):
    if not service.is_enabled():
        msg = '%s is not enabled' % service.name
        return (2, msg, None)

    if service.is_running():
        msg = 'active since %s' % service.get_property('ActiveEnterTimestamp')
        retval = 0
    else:
        msg = 'inactive since %s' % service.get_property('ActiveExitTimestamp')
        retval = 2

    return (retval, msg, None)

def check_status_oneshot(service,
                         exit_time_warning=0,
                         exit_time_critical=0,
                         running_time_warning=0,
                         running_time_critical=0):
    if not service.is_enabled():
        msg = '%s is not enabled' % service.name
        return (2, msg, None)

    perfdata = []
    exit_seconds = 0
    running_seconds = 0
    now = datetime.now()
    start_timestamp = service.get_property('ExecMainStartTimestamp')
    exit_timestamp = service.get_property('ExecMainExitTimestamp')

    if service.is_running():
        msg = 'service has been running since %s' % start_timestamp
        running_seconds = int((now - start_timestamp).total_seconds())
        warning = timedelta(seconds=running_time_warning)
        critical = timedelta(seconds=running_time_critical)

        if warning > critical:
            raise ValueError('Warning threshold is higher than critical threshold')

        if critical and start_timestamp < now - critical:
            retval = 2
        elif warning and start_timestamp < now - warning:
            retval = 1
        else:
            retval = 0
    elif exit_timestamp:
        msg = 'service ran for %s' % (exit_timestamp - start_timestamp)
        exit_seconds = int((now - exit_timestamp).total_seconds())
        warning = timedelta(seconds=exit_time_warning)
        critical = timedelta(seconds=exit_time_critical)

        if warning > critical:
            raise ValueError('Warning threshold is higher than critical threshold')

        if service.is_failed():
            msg = 'service failed at %s' % exit_timestamp
            retval = 2
        else:
            if critical and exit_timestamp < now - critical:
                retval = 2
            elif warning and exit_timestamp < now - warning:
                retval = 1
            else:
                retval = 0
    else:
        msg = 'service has not run since boot'
        boot_timestamp = get_boot_time()
        warning = timedelta(seconds=exit_time_warning)
        critical = timedelta(seconds=exit_time_critical)

        if warning > critical:
            raise ValueError('Warning threshold is higher than critical threshold')

        if critical and boot_timestamp < now - critical:
            retval = 2
        elif warning and boot_timestamp < now - warning:
            retval = 1
        else:
            retval = 3

    perfdata.append(get_perfdata('running_time', running_seconds, running_time_warning,
                                 running_time_critical))
    perfdata.append(get_perfdata('time_since_exit', exit_seconds, exit_time_warning,
                                 exit_time_critical))
    return (retval, msg, ' '.join(perfdata))


def main():
    parser = argparse.ArgumentParser('Check systemd service status')
    parser.add_argument('--service', required=True)
    parser.add_argument('--exit-time-warning', type=int, default=0)
    parser.add_argument('--exit-time-critical', type=int, default=0)
    parser.add_argument('--running-time-warning', type=int, default=0)
    parser.add_argument('--running-time-critical', type=int, default=0)
    args = parser.parse_args()

    service = Service(args.service)

    if service.get_property('Type') == 'oneshot':
        retval, msg, perfdata = check_status_oneshot(service,
                                                     args.exit_time_warning,
                                                     args.exit_time_critical,
                                                     args.running_time_warning,
                                                     args.running_time_critical)
    else:
        retval, msg, perfdata = check_status(service)

    status = get_status_msg(retval)

    if perfdata:
        print('%s - %s | %s' % (status, msg, perfdata))
    else:
        print('%s - %s' % (status, msg))

    return retval

if __name__ == '__main__':
    try:
        sys.exit(main())
    except Exception as e:
        print('Error - %s' % str(e))
        sys.exit(3)