nagios ack bot

nagbot.rb

#!/usr/bin/env ruby

# ---
# quick and dirty nagios acknowledgement bot
# [email protected]
# @oogali
# ---
#
# what do you need to run?
# - ruby 1.8+
# - $ gem install isaac eventmachine em-hiredis em-http-request
# - my nagios mail to redis bridge
# - a working irc server
# - a working redis installation
# - a working nagios installation (daemon, web+cgi, mail)
#
# how does it work?
# a) nagios generates an alert, and e-mails it to [email protected]
#    e.g. oogali+notifications@localhost
#
# b) in ~some/.forward, you have my nagios-to-redis publisher:
#    $ cat ~oogali/.forward+notifications 
#    | /home/oogali/nagios-mail-to-redis.rb
#
# c) you're running the irc bot, and it's connected to both irc and redis
#    09:20:35 -!- nagbot [[email protected]] has joined #systems
#    09:23:36 < nagbot> [001] Service PROBLEM: Layer 7: UDP/12345 Goofy Healthcheck @ oogali.lab.blip.tv
#
# d) you acknowledge the alert, and wait for the update
#    09:36:31 <@oo> !ack 1 
#    09:36:54 < nagbot> [***] Service ACKNOWLEDGEMENT: Layer 7: UDP/12345 Goofy Healthcheck @ oogali.lab.blip.tv
#
#    you can optionally acknowledge with a message, that will be recorded in nagios ack history:
#    10:13:22 <@niels> !ack 13 the rent is too damn high
#    10:13:27 < nagbot> [***] Service ACKNOWLEDGEMENT: Layer 7: UDP/5678 Stealth Server Healthcheck @ niels.lab.blip.tv
#

require 'rubygems'
require 'isaac'
require 'em-hiredis'
require 'em-http-request'

# irc bot/network settings
configure do |c|
  c.nick = 'nagbot'
  c.server = 'not.your.favorite.efnet.server'
  c.port = 6697
  c.ssl = true
end

# define our program settings as constants
REDIS_URI = 'redis://localhost:6379'
ALERT_TTL = 300
NAGIOS_INSTANCE = 'http://localhost/nagios/'
NAGIOS_USER= 'ircuser'
NAGIOS_PASSWORD = 'too many secrets'
CHANNEL = '#systems'
#KEY = 'key to your channel'

## You shouldn't need to change anything below here ##

# define some globals
msgnum = 0
redis = nil
pubsub = nil

# connect to irc network
on :connect do
  join "#{CHANNEL} #{KEY}"

  # connect regular redis handle
  redis = EM::Hiredis.connect REDIS_URI
  redis.errback do |code|
    msg CHANNEL, "could not connect to redis: #{code}"
  end
  
  # connect redis pubsub handle
  pubsub = EM::Hiredis.connect REDIS_URI
  pubsub.errback do |code|
    msg CHANNEL, "could not connect to redis (pubsub): #{code}"
  end
  
  # subscribe to nagios alerts, and set up callback
  pubsub.subscribe('nagios::alerts')
  pubsub.on(:message) do |channel, msg|
    m = msg.match /\*\*\s+(\S+)\s+(\S+)\s+(\S+):\s+(\S+)\/(.+)\s+is\s+(.+)\s+\*\*/
    
    if !m
      msg '#systems', msg
    else
      if m[1] == 'ACKNOWLEDGEMENT'
        msg CHANNEL, "[***] #{m[2]} #{m[1]}: #{m[5]} @ #{m[4]}"
      elsif m[1] == 'RECOVERY'
        msg CHANNEL, "[***] #{m[2]} #{m[1]}: #{m[5]} @ #{m[4]} is #{m[6]}"
      else
        # msgnum is a continually incrementing counter
        msgnum += 1
        msg CHANNEL, "[#{'%03d' % msgnum}] #{m[2]} #{m[1]}: #{m[5]} @ #{m[4]}"
    
        # everytime a message is logged to the cache, it's set with a 30 second expiry
        # you have 30 seconds to ack a message
        key = "nagios::alerts::#{'%03d' % msgnum}"
        redis.set(key, "#{m[4]};#{m[5]}") do
          redis.expire key, ALERT_TTL
        end
      end
    end
  end
end

on :channel, /^!ack\s+(\d+)\s*(.*)/ do
  # create key name
  key = "nagios::alerts::#{'%03d' % match[0].to_i}"
  
  # get host/service pair from redis
  redis.get(key) do |alert|
    if alert
      # get the host and service, then send a ack request to http instance
      host, service = alert.split(/;/, 2)
      h = EventMachine::HttpRequest.new("#{NAGIOS_INSTANCE}/cgi-bin/cmd.cgi").post({
        :head => {
          'Authorization' => [ NAGIOS_USER, NAGIOS_PASSWORD ]
        },
        :body => {
          :cmd_typ => 34,
          :cmd_mod => 2,
          :host => host,
          :service => service,
          :send_notification => 1,
          :com_author => NAGIOS_USER,
          :com_data => "By #{nick}: #{match[1] || 'No message given'}" 
        }
      })
      
      # XXX: should only delete key if previous http request comes back without errors
      redis.delete(key)
    end
  end
end

On what versions of Nagios have you tested this?