chrony_status

#!/usr/bin/env python3

"""Munin plugin to monitor chrony NTP daemon status.

=head1 NAME

chrony_status - monitor chrony NTP daemon status

=head1 APPLICABLE SYSTEMS

Systems with chrony installed.

=head1 CONFIGURATION

Needs to be run as the user running chronyd (or root) in order to access the
Unix domain socket which chronyc uses to communicate with chronyd. Example
/etc/munin/plugin-conf.d/chrony_status.conf:

    [chrony_status]
    user _chrony

=head1 INTERPRETATION

Monitor Chrony's stratum value (with warning), time offset, network delay, clock
frequency, packets received, and packets dropped. It would be very easy to
monitor all of Chrony's values, but IMHO they aren't needed. The most important
information is stratum, giving "synced" / "not synced" value.

=head1 AUTHOR

Kim B. Heino <b@bbbs.net>

=head1 LICENSE

GPLv2

=head1 MAGIC MARKERS

 #%# family=auto
 #%# capabilities=autoconf

=cut
"""

import os
import subprocess
import sys

GRAPHS = [
    'stratum',
    'systime',
    'delay',
    'frequency',
    'serverstats',
]

SERVERSTATS = [
    'received',
    'dropped',
    'command_received',
    'command_dropped',
    'client_log_records_dropped',
    'nts_ke_connections_accepted',
    'nts_ke_connections_dropped',
    'authenticated_packets',
]

FIELDS = {
    'stratum': 'Stratum',
    'systime': 'System time',
    'delay': 'Root delay',
    'frequency': 'Frequency',
    'received': 'NTP packets received',
    'dropped': 'NTP packets dropped',
    'command_received': 'Command packets received',
    'command_dropped': 'Command packets dropped',
    'client_log_records_dropped': 'Client log records dropped',
    'nts_ke_connections_accepted': 'NTS-KE connections accepted',
    'nts_ke_connections_dropped': 'NTS-KE connections dropped',
    'authenticated_packets': 'Authenticated NTP packets',
}


def get_values():
    """Run `chronyc tracking` and `chronyc serverstats` and parse their output.

    Return: list of (label, value, description)
    """
    try:
        output = subprocess.run(['chronyc', '-m', 'tracking', 'serverstats'],
                                stdout=subprocess.PIPE, check=False,
                                encoding='utf-8', errors='ignore')
    except FileNotFoundError:
        return {}
    lines = output.stdout.splitlines()
    ret = {}
    for label in FIELDS:
        for line in lines:
            if line.startswith(FIELDS[label]):
                value = line.split(':', 1)[1].split()[0]
                if ' slow' in line:
                    value = -float(value)
                ret[label] = value
    return ret


def config():
    """Print plugin config."""
    print('multigraph chrony_stratum')
    print('graph_title Chrony stratum')
    print('graph_vlabel stratum')
    print('graph_category time')
    print('graph_args --base 1000 --lower-limit 0')
    print('stratum.label Stratum')
    print('stratum.warning 1:9')
    # Use long unknown_limit to allow server reboot without Munin warning.
    # Clock doesn't drift fast so there's no hurry with warning.
    print('stratum.unknown_limit 15')

    print('multigraph chrony_systime')
    print('graph_title Chrony system time offset')
    print('graph_vlabel seconds')
    print('graph_category time')
    print('graph_args --base 1000')
    print('systime.label System time offset to NTP time')

    print('multigraph chrony_delay')
    print('graph_title Chrony network delay')
    print('graph_vlabel seconds')
    print('graph_category time')
    print('graph_args --base 1000')
    print('delay.label Network path delay')

    print('multigraph chrony_frequency')
    print('graph_title Chrony clock frequency error')
    print('graph_vlabel ppm')
    print('graph_category time')
    print('graph_args --base 1000')
    print('frequency.label Local clock frequency error')

    print('multigraph chrony_serverstats')
    print('graph_title Chrony server statistics')
    print('graph_vlabel Packets/${graph_period}')
    print('graph_category time')
    print('graph_args --base 1000')
    print('received.label Packets received')
    print('received.type DERIVE')
    print('received.min 0')
    print('dropped.label Packets dropped')
    print('dropped.type DERIVE')
    print('dropped.min 0')
    print('command_received.label Command packets received')
    print('command_received.type DERIVE')
    print('command_received.min 0')
    print('command_dropped.label Command packets dropped')
    print('command_dropped.type DERIVE')
    print('command_dropped.min 0')
    print('client_log_records_dropped.label Client log records dropped')
    print('client_log_records_dropped.type DERIVE')
    print('client_log_records_dropped.min 0')
    print('nts_ke_connections_accepted.label NTS-KE connections accepted')
    print('nts_ke_connections_accepted.type DERIVE')
    print('nts_ke_connections_accepted.min 0')
    print('nts_ke_connections_dropped.label NTS-KE connections dropped')
    print('nts_ke_connections_dropped.type DERIVE')
    print('nts_ke_connections_dropped.min 0')
    print('authenticated_packets.label Authenticated NTP packets')
    print('authenticated_packets.type DERIVE')
    print('authenticated_packets.min 0')

    if os.environ.get('MUNIN_CAP_DIRTYCONFIG') == '1':
        fetch()


def fetch():
    """Print values."""
    values = get_values()
    for graph in GRAPHS:
        print('multigraph chrony_{}'.format(graph))
        if graph == 'stratum' and values[graph] == 0:
            print('{}.value U'.format(graph))
        elif graph == 'serverstats':
            for stat in SERVERSTATS:
                if stat in values:
                    print('{}.value {}'.format(stat, values[stat]))
        else:
            print('{}.value {}'.format(graph, values[graph]))


if __name__ == '__main__':
    if len(sys.argv) > 1 and sys.argv[1] == 'autoconf':
        print('yes' if get_values() else 'no (chrony is not running)')
    elif len(sys.argv) > 1 and sys.argv[1] == 'config':
        config()
    else:
        fetch()