check_systemd_timer: parse cli output instead, more perfdta

This commit is contained in:
Cyberes 2024-03-07 14:09:42 -07:00
parent d32092c62f
commit efc3c010f9
3 changed files with 90 additions and 21 deletions

View File

@ -2,18 +2,61 @@
import argparse
import sys
import time
import traceback
from datetime import datetime, timedelta
import humanize
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
import re
import subprocess
from datetime import datetime
import humanfriendly
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]*)\s*(([0-9]*[a-z]*\s)*left)\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]*)\s*([0-9A-Za-z\s]*\sago)\s*([A-Za-z\-_]*.timer)\s*([A-Za-z\-_]*.service)')
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:
parts = re.search(SYSTEMCTL_TIMERS_RE, line)
try:
datetime_object = datetime.strptime(parts.group(1), '%a %Y-%m-%d %H:%M:%S %Z')
except ValueError as e:
return None, e
time_left = parse_systemctl_time_delta(parts.group(2))
if isinstance(time_left, humanfriendly.InvalidTimespan):
return None, humanfriendly.InvalidTimespan
time_passed = parse_systemctl_time_delta(parts.group(5))
if isinstance(time_passed, humanfriendly.InvalidTimespan):
return None, humanfriendly.InvalidTimespan
timer_info = {
'next': datetime_object,
'left': time_left,
'last': parts.group(4),
'passed': time_passed,
'unit': parts.group(6),
'activates': parts.group(7)
}
return timer_info, None
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'):
@ -28,31 +71,32 @@ def check_timer(timer_name):
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 = timer_properties.Get('org.freedesktop.systemd1.Timer', 'NextElapseUSecRealtime')
next_elapse, err = get_next_elapse(timer_name)
if err:
quit_check(f'{err}', nagios.STATE_UNKNOWN)
try:
local_timezone_offset_seconds = time.localtime().tm_gmtoff
local_timezone_offset = timedelta(seconds=local_timezone_offset_seconds)
local_datetime = datetime.utcfromtimestamp(int(next_elapse) / 1e6) + local_timezone_offset
next_elapse_str = local_datetime.strftime("%m-%d-%Y %H:%M")
except ValueError as v_err:
if 'is out of range' in repr(v_err):
# This occurs whenever the timer resets.
# "year 586524 is out of range"
print(f'OK - {timer_name} has triggered.')
sys.exit(nagios.STATE_OK)
print(next_elapse)
current_time = time.time() * 1e6 # convert current time to microseconds
remaining_time_sec = int((next_elapse - current_time) / 1e6) # convert remaining time to seconds
remaining_time_human = str(humanize.naturaltime(datetime.now() + timedelta(seconds=remaining_time_sec))).strip(' from now')
if 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)
local_tz = tz.tzlocal()
next_elapse_str = next_elapse['next'].replace(tzinfo=local_tz).strftime('%a %Y-%m-%d %H:%M %Z')
remaining_time_human = humanfriendly.format_timespan(next_elapse['left'])
passed_time_human = humanfriendly.format_timespan(next_elapse['passed'])
perfdata_dict = {
'remaining_time': {
'value': remaining_time_sec,
'value': int(next_elapse['left']),
'unit': 's',
'min': 0
},
'passed_time': {
'value': int(next_elapse['passed']),
'unit': 's',
'min': 0
}
}
quit_check(f'{timer_name} is active. Trigger time: {next_elapse_str}. Remaining time: {remaining_time_human}.', nagios.STATE_OK, perfdata_dict)
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:

24
checker/humanfriendly.py Normal file
View File

@ -0,0 +1,24 @@
import re
import humanfriendly
SYSTEMCTL_TIMESPAN_RE = re.compile(r'([0-9]+((\s[a-z]*)|([a-z]+)))')
def parse_systemctl_time_delta(time_str: str):
"""
https://humanfriendly.readthedocs.io/en/latest/api.html?highlight=parse_timespan#humanfriendly.parse_timespan
"""
time_str = time_str.replace(' left', '').replace(' ago', '')
spans = []
for part in re.findall(SYSTEMCTL_TIMESPAN_RE, time_str):
delta = part[0]
if 'months' in delta:
# humanfriendly does not support "months" so we convert it to weeks
num = int(delta.split(' ')[0])
delta = f'{num * 4} weeks'
try:
spans.append(humanfriendly.parse_timespan(delta))
except humanfriendly.InvalidTimespan as e:
return e
return sum(spans)

View File

@ -10,4 +10,5 @@ cf_speedtest==0.1.7
zfslib==0.11.0
hurry.filesize==0.9
dateparser==1.2.0
humanize==4.9.0
humanize==4.9.0
humanfriendly==10.0