Script to scrape Nagios and update Cachet status page

monitor.py

#!/usr/bin/env python

# Import required modules
import os
import requests
import time
import getpass
from bs4 import BeautifulSoup
import re

# Set some useful variables
nagios_username = input('Nagios Username: ')
nagios_password = getpass.getpass('Nagios Password: ')
api_key = input('Cachet API key: ') # This is cachet API key
poll_interval = 2 * 60 # In seconds

# We will check only for these services
# Add service name and corresponding status code
# Status code description:
#   1 = Operational
#   2 = Performance Issues
#   3 = Partial Outage
#   4 = Major Outage
scores = {
	'CPU Load': 2,
	'Mem Usage': 2,
	'DNS': 4,
	'HTTP': 3,
	'HTTPS': 3,
	'SMTP': 4,
	'SMTPS': 4,
	'POP3': 4,
	'POP3S': 4,
	'IMAP': 4,
	'IMAPS': 4,
	'Postfix Email Queue': 2,
	'Squid': 4
}

# Get list of all added components in Cachet
def getServersList():
	response = requests.get('https://status.iiit.ac.in/api/v1/components')
	response = response.json()
	response = response['data']
	return response

# Ping for given host
# We use this as a fallback mechanism
def checkHost(host):
	if os.system("ping -c 1 " + host + " 2>/dev/null 1>/dev/null") is 0:
		return 1
	return 4

# Scrape and nagios
def checkNagios(service_name):
	# Start with status code and update based on severity
  status_code = 1

   # Authenticate Nagios session
	nagios_session = requests.Session()
	nagios_session.auth = (nagios_username, nagios_password)
	nagios_session.post('https://nagios.iiit.ac.in')

  # Get host details and extract available services on the host
	nagios_page = nagios_session.get('https://nagios.iiit.ac.in/nagios/cgi-bin//status.cgi?host=' + service_name + '&limit=0')
	host_table = BeautifulSoup(nagios_page.text, 'html.parser')
	host_table = host_table.findAll('td', { 'class': 'statusEven', 'align': 'left' })
	del host_table[0]

	# Check if host is down
	url = 'https://nagios.iiit.ac.in/nagios/cgi-bin//extinfo.cgi?type=1&host='
	url += service_name
	page = nagios_session.get(url)
	if re.search('hostDOWN', page.text):
		return 4
  # Todo check for packet loss and return 3

	# Check services
	for service in host_table:
		if service.text in scores:
			url = 'https://nagios.iiit.ac.in/nagios/cgi-bin//extinfo.cgi?type=2&host='
			url += service_name + '&service='
			url += '+'.join(service_name.split())
			page = nagios_session.get(url)
      # We check only for critical status
			if re.search('serviceCRITICAL', page.text):
				tmp_code = scores[service.text]
				status_code = tmp_code if status_code < tmp_code else status_code

	return status_code

# Update cachet
# We monitor things based on the tag values of each component in cachet
# Two values are required in tag field:
#   1) service type   2) ip/host
# Right now, we just ping and return 1 or 4 code if there is nagios involved
# IMP: THERE SHOULD BE EITHER 'nagios,nagios_hostname' or 'ping,ip_or_domain' IN THE TAGS FIELD!
def updateMonitor(component):
	if 'ping' in component['tags']:
		del component['tags']['ping']
		host = list(component['tags'].values())[0]
		status_code = checkHost(host)

	elif 'nagios' in component['tags']:
		del component['tags']['nagios']
		service_name = list(component['tags'])[0]
		status_code = checkNagios(service_name)

	else:
		status_code = 4

	headers = {
		'X-Cachet-Token': api_key
	}

	payload = {
		'status': status_code
	}

	url = 'https://status.iiit.ac.in/api/v1/components/' + str(component['id'])
	requests.put(url, data = payload, headers = headers)

# Loop for all components in cachet
def monitor():
	servers = getServersList()
	for server in servers:
		updateMonitor(server)

# Run the monitor forever
# For cron job, remove the below part and run monitor()
if __name__ == '__main__':
	while True:
		monitor()
		time.sleep(poll_interval)

requirements.txt

beautifulsoup4==4.6.0
certifi==2018.4.16
chardet==3.0.4
idna==2.6
requests==2.18.4
urllib3==1.22