163 lines
4.0 KiB
Python
Executable File
163 lines
4.0 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
# pylint: disable=invalid-name
|
|
# pylint: enable=invalid-name
|
|
# pylint: disable=consider-using-f-string
|
|
|
|
"""Munin plugin to monitor systemd service status.
|
|
|
|
=head1 NAME
|
|
|
|
systemd_status - monitor systemd service status, including normal services,
|
|
mounts, hotplugs and socket activations
|
|
|
|
=head1 APPLICABLE SYSTEMS
|
|
|
|
Linux systems with systemd installed.
|
|
|
|
=head1 CONFIGURATION
|
|
|
|
No configuration is required for this plugin.
|
|
|
|
Warning level for systemd "failed" state is set to 0:0. If any of the services
|
|
enters "failed" state, Munin will emit warning.
|
|
|
|
Single service can be ignored by configuring it's value to "0":
|
|
|
|
[systemd_status]
|
|
env.fwupd_refresh_service 0
|
|
|
|
Normally this plugin monitors global system services. User services can be
|
|
monitored by manually adding softlink and config:
|
|
|
|
ln -s /usr/share/munin/plugins/systemd_status \
|
|
/etc/munin/plugins/systemd_status_b
|
|
|
|
[systemd_status_b]
|
|
user b
|
|
env.monitor_user yes
|
|
|
|
=head1 AUTHOR
|
|
|
|
Kim B. Heino <b@bbbs.net>
|
|
|
|
=head1 LICENSE
|
|
|
|
GPLv2
|
|
|
|
=head1 MAGIC MARKERS
|
|
|
|
#%# family=auto
|
|
#%# capabilities=autoconf
|
|
|
|
=cut
|
|
|
|
"""
|
|
|
|
import os
|
|
import pwd
|
|
import re
|
|
import subprocess
|
|
import sys
|
|
import unicodedata
|
|
|
|
|
|
STATES = (
|
|
'failed',
|
|
'dead',
|
|
'running',
|
|
'exited',
|
|
'active',
|
|
'listening',
|
|
'waiting',
|
|
'plugged',
|
|
'mounted',
|
|
)
|
|
|
|
USER = os.environ.get('monitor_user') in ('yes', '1')
|
|
|
|
|
|
def safename(name):
|
|
"""Return safe variable name."""
|
|
# Convert ä->a as isalpha('ä') is true
|
|
value = unicodedata.normalize('NFKD', name)
|
|
value = value.encode('ASCII', 'ignore').decode('utf-8')
|
|
|
|
# Remove non-alphanumeric chars
|
|
return ''.join(char.lower() if char.isalnum() else '_' for char in value)
|
|
|
|
|
|
def config():
|
|
"""Autoconfig values."""
|
|
if USER:
|
|
print('graph_title systemd services for {}'.format(
|
|
pwd.getpwuid(os.geteuid())[0]))
|
|
else:
|
|
print('graph_title systemd services')
|
|
print('graph_vlabel Services')
|
|
print('graph_category processes')
|
|
print('graph_args --base 1000 --lower-limit 0')
|
|
print('graph_scale no')
|
|
print('graph_info Number of services in given activation state.')
|
|
for state in STATES:
|
|
print('{state}.label Services in {state} state'.format(state=state))
|
|
print('failed.warning 0:0')
|
|
if os.environ.get('MUNIN_CAP_DIRTYCONFIG') == '1':
|
|
fetch()
|
|
|
|
|
|
def fetch():
|
|
"""Print runtime values."""
|
|
# Get data
|
|
try:
|
|
output = subprocess.check_output(['systemctl', 'list-units'] +
|
|
(['--user'] if USER else []),
|
|
encoding='utf-8')
|
|
except (OSError, subprocess.CalledProcessError):
|
|
return
|
|
|
|
# Parse data
|
|
states = {state: 0 for state in STATES}
|
|
for line in output.splitlines():
|
|
token = line.split()
|
|
if token and len(token[0]) < 3: # Skip failed-bullet
|
|
token = token[1:]
|
|
if len(token) < 4:
|
|
continue
|
|
service = token[0]
|
|
state = token[3]
|
|
if service.endswith('.scope'):
|
|
continue # Ignore scopes
|
|
if re.match(r'user.*@\d+\.service', service):
|
|
continue # These fail randomly in older systemd
|
|
if state in states:
|
|
value = 1
|
|
if state == 'failed':
|
|
value = int(os.environ.get(safename(service), 1))
|
|
states[state] += value
|
|
|
|
# Output
|
|
for state in STATES:
|
|
print('{}.value {}'.format(state, states[state]))
|
|
|
|
|
|
def fix_dbus():
|
|
"""Fix DBus address for user service."""
|
|
if not USER:
|
|
return
|
|
euid = '/{}/'.format(os.geteuid())
|
|
if euid in os.environ.get('DBUS_SESSION_BUS_ADDRESS', ''):
|
|
return
|
|
os.putenv('DBUS_SESSION_BUS_ADDRESS',
|
|
'unix:path=/run/user{}bus'.format(euid))
|
|
|
|
|
|
if __name__ == '__main__':
|
|
fix_dbus()
|
|
if len(sys.argv) > 1 and sys.argv[1] == 'autoconf':
|
|
print('yes' if os.path.exists('/run/systemd/system') else
|
|
'no (systemd is not running)')
|
|
elif len(sys.argv) > 1 and sys.argv[1] == 'config':
|
|
config()
|
|
else:
|
|
fetch()
|