bundlewrap/bundles/sshmon/files/check_cpu_stats

69 lines
1.7 KiB
Text
Raw Permalink Normal View History

#!/usr/bin/env python3
from re import findall
from subprocess import check_output
from sys import exit
try:
top_output = None
for line in check_output(['top', '-b', '-n1', '-d1']).decode('UTF-8').splitlines():
if line.lower().strip().startswith('%cpu'):
top_output = line.lower().split(':', 2)[1]
break
if not top_output:
print('%cpu not found in top output')
exit(3)
cpu_usage = {}
for value, identifier in findall('([0-9\.\,]{3,5}) ([a-z]{2})', top_output):
cpu_usage[identifier] = float(value.replace(',', '.'))
warn = set()
crit = set()
print(top_output)
# steal
if cpu_usage['st'] > 10:
crit.add('CPU steal is {}% (>10%)'.format(cpu_usage['st']))
elif cpu_usage['st'] > 5:
warn.add('CPU steal is {}% (>5%)'.format(cpu_usage['st']))
# iowait
if cpu_usage['wa'] > 60:
crit.add('IOwait is {}% (>60%)'.format(cpu_usage['wa']))
elif cpu_usage['wa'] > 30:
warn.add('IOwait is {}% (>30%)'.format(cpu_usage['wa']))
total_usage = cpu_usage['us'] + cpu_usage['sy']
if total_usage > 90:
crit.add('Total CPU usage is {:.1f}% ({}% user, {}% system, >90%)'.format(
total_usage,
cpu_usage['us'],
cpu_usage['sy'],
))
elif total_usage > 80:
warn.add('Total CPU usage is {:.1f}% ({}% user, {}% system > 80%)'.format(
total_usage,
cpu_usage['us'],
cpu_usage['sy'],
))
for line in sorted(crit):
print(line)
for line in sorted(warn):
print(line)
if crit:
exit(2)
elif warn:
exit(1)
else:
exit(0)
except Exception as e:
print(repr(e))
exit(3)