Nagios DNS resolution health check

check_dns_resolution.py

#!/usr/bin/env python
"""
Nagios check to test DNS resolution data pulled from Graphite
"""

import optparse
import logging
import requests
import sys

OK = 0
WARNING = 1
CRITICAL = 2

LOGGER = logging.getLogger(__name__)
LOGGER.addHandler(logging.StreamHandler(sys.stdout))
LOGGER.setLevel(logging.INFO)

parse = optparse.OptionParser()
parse.add_option('-s', '--server', type='string', dest='server',
                 help='hostname of the server being checked')
parse.add_option('-g', '--graphite', type='string', dest='graphite',
                 help='hostname of the graphite server to poll')
parse.add_option('-w', '--warning', type='int', dest='warning_threshold',
                 help='threshold at which to warn')
parse.add_option('-c', '--critical', type='int', dest='critical_threshold',
                 help='threshold at which to go critical')
parse.add_option('-u', '--url', type='string', dest='url',
                 help='url of the site being resolved')
parse.add_option('-p', '--points', type='int', dest='points', default='10',
                 help='number of historical data points to average')

(opts, args) = parse.parse_args()

# Require at least one argument
if not opts.server:
    parse.print_help()
    exit(OK)

# Define the metric url on the Graphite server
graphite_url = 'http://{0}/render/?from=-1days&target=servers.{1}.dns.{2}' \
    '&rawData'.format(opts.graphite, opts.server, opts.url.replace('.', '_'))

try:
    # Grab Graphite data and parse out the data points
    data = requests.get(graphite_url).text.split('|')[-1].rstrip('\n')

    # Replace None with zeros
    data = [d.replace('None', '0.0') for d in data.split(',')]

    # Convert everything from strings -> floats
    data = [float(d) for d in data[-opts.points:]]

    # Determine the average response time over the past X points
    average = float(sum(data) / opts.points)

    if not average:
        LOGGER.info('CRITICAL - DNS Resolution is failing!')
        exit(CRITICAL)
    elif average > opts.critical_threshold:
        LOGGER.info('CRITICAL - DNS Resolution took %s ms' % average)
        exit(CRITICAL)
    elif average > opts.warning_threshold:
        LOGGER.info('WARNING - DNS Resolution took %s ms' % average)
        exit(WARNING)
    else:
        LOGGER.info('OK - DNS Resolution took %s ms' % average)
        exit(OK)
except requests.ConnectionError:
    LOGGER.critical('CRITICAL - There was an error connecting to graphite')
    exit(CRITICAL)