Zabbix Ceph Monitor Cron

zabbix-ceph-mon.py

#!/usr/bin/env python
'''
Script to send Zabbix Agent active data to the Zabbix Server.

It runs the ceph command locally to get information from the running
OSDs and send the aggegrated data back to Zabbix.
'''

import sys
import subprocess
import socket
import tempfile
import os
import json

hostname = None
zbxactive = None
zbxsender = '/usr/bin/zabbix_sender'

def spawn(command, shell=True):
    p = subprocess.Popen(command, stdout=subprocess.PIPE, shell=shell)
    (result, err) = p.communicate()
    return result.strip()

stats = {
          'num_osds': 0,
          'num_up_osds': 0,
          'num_in_osds': 0,
          'num_pgs': 0,
          'num_pg_backfill': 0,
          'num_pg_backfilling': 0,
          'num_pg_incomplete': 0,
          'num_pg_degraded': 0,
          'num_pg_unclean': 0,
          'num_pg_inactive': 0,
          'num_pg_recovering': 0,
          'overall_status': 0,
          'slow_requests': 0,
          'blocked_requests': 0
        }

perf_dump = json.loads(spawn(['ceph', '-s', '--format=json'], False))

stats['num_osds'] = perf_dump['osdmap']['osdmap']['num_osds']
stats['num_up_osds'] = perf_dump['osdmap']['osdmap']['num_up_osds']
stats['num_in_osds'] = perf_dump['osdmap']['osdmap']['num_in_osds']
stats['num_pgs'] = perf_dump['pgmap']['num_pgs']

status = perf_dump['health']['overall_status']

if status == "HEALTH_OK":
    stats['overall_status'] = 1
elif status == "HEALTH_WARN":
    stats['overall_status'] = 2
elif status == "HEALTH_ERR":
    stats['overall_status'] = 3

for sum in perf_dump['health']['summary']:
    if 'slow' in sum['summary']:
        stats['slow_requests'] = sum['summary'].split()[0]

    if 'blocked' in sum['summary']:
        stats['blocked_requests'] = sum['summary'].split()[0]

    if sum['summary'].endswith('backfill'):
        stats['num_pg_backfill'] = sum['summary'].split()[0]

    if sum['summary'].endswith('backfilling'):
        stats['num_pg_backfilling'] = sum['summary'].split()[0]

    if sum['summary'].endswith('incomplete'):
        stats['num_pg_incomplete'] = sum['summary'].split()[0]

    if sum['summary'].endswith('unclean'):
        stats['num_pg_unclean'] = sum['summary'].split()[0]

    if sum['summary'].endswith('recovering'):
        stats['num_pg_recovering'] = sum['summary'].split()[0]

    if sum['summary'].endswith('inactive'):
        stats['num_pg_inactive'] = sum['summary'].split()[0]

    if sum['summary'].endswith('degraded'):
        stats['num_pg_degraded'] = sum['summary'].split()[0]

for zbxserver in zbxactive:
    command="%s -z %s -s %s -vv -i -" % (zbxsender, zbxserver, hostname)
    zbxproc = subprocess.Popen(command, shell=True, stdin=subprocess.PIPE, stdout=subprocess.PIPE)

    try:
        for key in stats:
            zbxproc.stdin.write("%s ceph.mon.%s %d\n" % (hostname, key, int(stats[key])))
            print "%s ceph.mon.%s %d" % (hostname, key, int(stats[key]))
    except Exception as e:
        print "Failed to fetch Ceph statistics: %s" % str(e)

    output = zbxproc.communicate()[0]
    print output