sensu-plugins-rabbitmq 0.0.1

data/bin/check-rabbitmq-messages.rb

@@ -0,0 +1,128 @@
+#!/usr/bin/env ruby
+#  encoding: UTF-8
+#
+# Check RabbitMQ Messages
+# ===
+#
+# DESCRIPTION:
+# This plugin checks the total number of messages queued on the RabbitMQ server
+#
+# PLATFORMS:
+#   Linux, BSD, Solaris
+#
+# DEPENDENCIES:
+#   RabbitMQ rabbitmq_management plugin
+#   gem: sensu-plugin
+#   gem: carrot-top
+#
+# LICENSE:
+# Copyright 2012 Evan Hazlett <ejhazlett@gmail.com>
+# Copyright 2015 Tim Smith <tim@cozy.co> and Cozy Services Ltd.
+#
+# Released under the same terms as Sensu (the MIT license); see LICENSE
+# for details.
+
+require 'sensu-plugin/check/cli'
+require 'socket'
+require 'carrot-top'
+
+# main plugin class
+class CheckRabbitMQMessages < Sensu::Plugin::Check::CLI
+  option :host,
+         description: 'RabbitMQ management API host',
+         long: '--host HOST',
+         default: 'localhost'
+
+  option :port,
+         description: 'RabbitMQ management API port',
+         long: '--port PORT',
+         proc: proc(&:to_i),
+         default: 15_672
+
+  option :user,
+         description: 'RabbitMQ management API user',
+         long: '--user USER',
+         default: 'guest'
+
+  option :password,
+         description: 'RabbitMQ management API password',
+         long: '--password PASSWORD',
+         default: 'guest'
+
+  option :ssl,
+         description: 'Enable SSL for connection to the API',
+         long: '--ssl',
+         boolean: true,
+         default: false
+
+  option :warn,
+         short: '-w NUM_MESSAGES',
+         long: '--warn NUM_MESSAGES',
+         description: 'WARNING message count threshold',
+         default: 250
+
+  option :critical,
+         short: '-c NUM_MESSAGES',
+         long: '--critical NUM_MESSAGES',
+         description: 'CRITICAL message count threshold',
+         default: 500
+
+  option :queuelevel,
+         short: '-q',
+         long: '--queuelevel',
+         description: 'Monitors that no individual queue is above the thresholds specified'
+
+  option :excluded,
+         short: '-e queue_name',
+         long: '--excludedqueues queue_name',
+         description: 'Comma separated list of queues to exclude when using queue level monitoring',
+         proc: proc { |q| q.split(',') },
+         default: []
+
+  def generate_message(status_hash)
+    message =  []
+    status_hash.each_pair do |k, v|
+      message << "#{k}: #{v}"
+    end
+    message.join(', ')
+  end
+
+  def acquire_rabbitmq_info
+    begin
+      rabbitmq_info = CarrotTop.new(
+        host: config[:host],
+        port: config[:port],
+        user: config[:user],
+        password: config[:password],
+        ssl: config[:ssl]
+      )
+    rescue
+      warning 'Could not connect to rabbitmq'
+    end
+    rabbitmq_info
+  end
+
+  def run
+    rabbitmq = acquire_rabbitmq_info
+
+    # monitor counts in each queue or monitor the total number of messages in the system
+    if config[:queuelevel]
+      warn_queues = {}
+      crit_queues = {}
+      rabbitmq.queues.each do |queue|
+        next if config[:excluded].include?(queue['name'])
+        (crit_queues["#{queue['name']}"] = queue['messages']; next) if queue['messages'] >= config[:critical].to_i # rubocop: disable Style/Semicolon
+        (warn_queues["#{queue['name']}"] = queue['messages']; next) if queue['messages'] >= config[:warn].to_i # rubocop: disable Style/Semicolon
+      end
+      message crit_queues.empty? ? generate_message(warn_queues) : generate_message(crit_queues)
+      critical unless crit_queues.empty?
+      warning unless warn_queues.empty?
+    else
+      total = rabbitmq.overview['queue_totals']['messages']
+      message "#{total}"
+      critical if total > config[:critical].to_i
+      warning if total > config[:warn].to_i
+    end
+    ok
+  end
+end

data/bin/check-rabbitmq-network-partitions.rb

@@ -0,0 +1,82 @@
+#!/usr/bin/env ruby
+#  encoding: UTF-8
+#
+# RabbitMQ Network Partitions Check
+# ===
+#
+# DESCRIPTION:
+# This plugin checks if a RabbitMQ network partition has occured.
+# https://www.rabbitmq.com/partitions.html
+#
+# PLATFORMS:
+#   Linux, BSD, Solaris
+#
+# DEPENDENCIES:
+#   RabbitMQ rabbitmq_management plugin
+#   gem: sensu-plugin
+#   gem: carrot-top
+#
+# LICENSE:
+# Copyright 2015 Ed Robinson <ed@reevoo.com> and Reevoo LTD.
+#
+# Released under the same terms as Sensu (the MIT license); see LICENSE
+# for details.
+
+require 'sensu-plugin/check/cli'
+require 'carrot-top'
+
+# main plugin class
+class CheckRabbitMQPartitions < Sensu::Plugin::Check::CLI
+  option :host,
+         description: 'RabbitMQ management API host',
+         short: '-w',
+         long: '--host HOST',
+         default: 'localhost'
+
+  option :port,
+         description: 'RabbitMQ management API port',
+         long: '--port PORT',
+         proc: proc(&:to_i),
+         default: 15_672
+
+  option :username,
+         description: 'RabbitMQ management API user',
+         short: '-u',
+         long: '--username USERNAME',
+         default: 'guest'
+
+  option :password,
+         description: 'RabbitMQ management API password',
+         short: '-p',
+         long: '--password PASSWORD',
+         default: 'guest'
+
+  option :ssl,
+         description: 'Enable SSL for connection to the API',
+         long: '--ssl',
+         boolean: true,
+         default: false
+
+  def run
+    critical 'network partition detected' if partition?
+    ok 'no network partition detected'
+  rescue Errno::ECONNREFUSED => e
+    critical e.message
+  rescue => e
+    unknown e.message
+  end
+
+  def partition?
+    rabbitmq_management.nodes.map { |node| node['partitions'] }.any?(&:any?)
+  end
+
+  def rabbitmq_management
+    CarrotTop.new(
+      host: config[:host],
+      port: config[:port],
+      user: config[:username],
+      password: config[:password],
+      ssl: config[:ssl]
+    )
+  end
+end

data/bin/check-rabbitmq-node-health.rb

@@ -0,0 +1,184 @@
+#!/usr/bin/env ruby
+#  encoding: UTF-8
+#
+# RabbitMQ check node health plugin
+# ===
+#
+# DESCRIPTION:
+# This plugin checks if RabbitMQ server node is in a running state.
+#
+# The plugin is based on the RabbitMQ cluster node health plugin by Tim Smith
+#
+# PLATFORMS:
+#   Linux, Windows, BSD, Solaris
+#
+# DEPENDENCIES:
+#   RabbitMQ rabbitmq_management plugin
+#   gem: sensu-plugin
+#   gem: rest-client
+#
+# LICENSE:
+# Copyright 2012 Abhijith G <abhi@runa.com> and Runa Inc.
+# Copyright 2014 Tim Smith <tim@cozy.co> and Cozy Services Ltd.
+# Copyright 2015 Edward McLain <ed@edmclain.com> and Daxko, LLC.
+#
+# Released under the same terms as Sensu (the MIT license); see LICENSE
+# for details.
+
+require 'sensu-plugin/check/cli'
+require 'json'
+require 'rest_client'
+
+# main plugin class
+class CheckRabbitMQNodeHealth < Sensu::Plugin::Check::CLI
+  option :host,
+         description: 'RabbitMQ host',
+         short: '-w',
+         long: '--host HOST',
+         default: 'localhost'
+
+  option :username,
+         description: 'RabbitMQ username',
+         short: '-u',
+         long: '--username USERNAME',
+         default: 'guest'
+
+  option :password,
+         description: 'RabbitMQ password',
+         short: '-p',
+         long: '--password PASSWORD',
+         default: 'guest'
+
+  option :port,
+         description: 'RabbitMQ API port',
+         short: '-P',
+         long: '--port PORT',
+         default: '15672'
+
+  option :memwarn,
+         description: 'Warning % of mem usage vs high watermark',
+         short: '-m',
+         long: '--mwarn PERCENT',
+         proc: proc(&:to_f),
+         default: 80
+
+  option :memcrit,
+         description: 'Critical % of mem usage vs high watermark',
+         short: '-c',
+         long: '--mcrit PERCENT',
+         proc: proc(&:to_f),
+         default: 90
+
+  option :fdwarn,
+         description: 'Warning % of file descriptor usage vs high watermark',
+         short: '-m',
+         long: '--mwarn PERCENT',
+         proc: proc(&:to_f),
+         default: 80
+
+  option :fdcrit,
+         description: 'Critical % of file descriptor usage vs high watermark',
+         short: '-c',
+         long: '--mcrit PERCENT',
+         proc: proc(&:to_f),
+         default: 90
+
+  option :socketwarn,
+         description: 'Warning % of socket usage vs high watermark',
+         short: '-m',
+         long: '--mwarn PERCENT',
+         proc: proc(&:to_f),
+         default: 80
+
+  option :socketcrit,
+         description: 'Critical % of socket usage vs high watermark',
+         short: '-c',
+         long: '--mcrit PERCENT',
+         proc: proc(&:to_f),
+         default: 90
+
+  option :watchalarms,
+         description: 'Sound critical if one or more alarms are triggered',
+         short: '-a BOOLEAN',
+         long: '--alarms BOOLEAN',
+         default: 'true'
+
+  def run
+    res = node_healthy?
+
+    if res['status'] == 'ok'
+      ok res['message']
+    elsif res['status'] == 'warning'
+      warning res['message']
+    elsif res['status'] == 'critical'
+      critical res['message']
+    else
+      unknown res['message']
+    end
+  end
+
+  def node_healthy?
+    host     = config[:host]
+    port     = config[:port]
+    username = config[:username]
+    password = config[:password]
+
+    begin
+      resource = RestClient::Resource.new "http://#{host}:#{port}/api/nodes", username, password
+      # Parse our json data
+      nodeinfo = JSON.parse(resource.get)[0]
+
+      # Determine % memory consumed
+      pmem = format('%.2f', nodeinfo['mem_used'].fdiv(nodeinfo['mem_limit']) * 100)
+      # Determine % sockets consumed
+      psocket = format('%.2f', nodeinfo['sockets_used'].fdiv(nodeinfo['sockets_total']) * 100)
+      # Determine % file descriptors consumed
+      pfd = format('%.2f', nodeinfo['fd_used'].fdiv(nodeinfo['fd_total']) * 100)
+
+      # build status and message
+      status = 'ok'
+      message = 'Server is healthy'
+
+      # criticals
+      if pmem.to_f >= config[:memcrit]
+        message = "Memory usage is critical: #{pmem}%"
+        status = 'critical'
+      elsif psocket.to_f >= config[:socketcrit]
+        message = "Socket usage is critical: #{psocket}%"
+        status = 'critical'
+      elsif pfd.to_f >= config[:fdcrit]
+        message = "Socket usage is critical: #{pfd}%"
+        status = 'critical'
+      # warnings
+      elsif pmem.to_f >= config[:memwarn]
+        message = "Memory usage is at warning: #{pmem}%"
+        status = 'warning'
+      elsif psocket.to_f >= config[:socketwarn]
+        message = "Socket usage is at warning: #{psocket}%"
+        status = 'warning'
+      elsif pfd.to_f >= config[:fdwarn]
+        message = "Socket usage is at warning: #{pfd}%"
+        status = 'warning'
+      end
+
+      # If we are set to watch alarms then watch those and set status and messages accordingly
+      if config[:watchalarms] == 'true'
+        if nodeinfo['mem_alarm'] == true
+          status = 'critical'
+          message += ' Memory Alarm ON'
+        end
+
+        if nodeinfo['disk_free_alarm'] == true
+          status = 'critical'
+          message += ' Disk Alarm ON'
+        end
+      end
+
+      { 'status' => status, 'message' => message }
+    rescue Errno::ECONNREFUSED => e
+      { 'status' => 'critical', 'message' => e.message }
+    rescue => e
+      { 'status' => 'unknown', 'message' => e.message }
+    end
+  end
+end

data/bin/check-rabbitmq-queue-drain-time.rb

@@ -0,0 +1,129 @@
+#!/usr/bin/env ruby
+#  encoding: UTF-8
+#
+# RabbitMQ Queue Drain Time
+# ===
+#
+# DESCRIPTION:
+# This plugin checks the time it will take for each queue on the RabbitMQ
+# server to drain based on the current message egress rate.  For example
+# if a queue has 1,000 messages in it, but egresses only 1 message a sec
+# the alert would fire as this is greater than the default critical level of 360s
+#
+# The plugin is based on the RabbitMQ Queue Metrics plugin
+#
+# PLATFORMS:
+#   Linux, BSD, Solaris
+#
+# DEPENDENCIES:
+#   RabbitMQ rabbitmq_management plugin
+#   gem: sensu-plugin
+#   gem: carrot-top
+#
+# LICENSE:
+# Copyright 2015 Tim Smith <tim@cozy.co> and Cozy Services Ltd.
+#
+# Released under the same terms as Sensu (the MIT license); see LICENSE
+# for details.
+
+require 'sensu-plugin/check/cli'
+require 'socket'
+require 'carrot-top'
+
+# main plugin class
+class CheckRabbitMQQueueDrainTime < Sensu::Plugin::Check::CLI
+  option :host,
+         description: 'RabbitMQ management API host',
+         long: '--host HOST',
+         default: 'localhost'
+
+  option :port,
+         description: 'RabbitMQ management API port',
+         long: '--port PORT',
+         proc: proc(&:to_i),
+         default: 15_672
+
+  option :user,
+         description: 'RabbitMQ management API user',
+         long: '--user USER',
+         default: 'guest'
+
+  option :password,
+         description: 'RabbitMQ management API password',
+         long: '--password PASSWORD',
+         default: 'guest'
+
+  option :filter,
+         description: 'Regular expression for filtering queues',
+         long: '--filter REGEX'
+
+  option :ssl,
+         description: 'Enable SSL for connection to the API',
+         long: '--ssl',
+         boolean: true,
+         default: false
+
+  option :warn,
+         short: '-w PROCESS_TIME_SECS',
+         long: '--warning PROCESS_TIME_SECS',
+         description: 'WARNING that messages will process at current rate',
+         default: 180
+
+  option :critical,
+         short: '-c PROCESS_TIME_SECS',
+         long: '--critical PROCESS_TIME_SECS',
+         description: 'CRITICAL time that messages will process at current rate',
+         default: 360
+
+  def acquire_rabbitmq_queues
+    begin
+      rabbitmq_info = CarrotTop.new(
+        host: config[:host],
+        port: config[:port],
+        user: config[:user],
+        password: config[:password],
+        ssl: config[:ssl]
+      )
+    rescue
+      warning 'could not get rabbitmq queue info'
+    end
+    rabbitmq_info.queues
+  end
+
+  def run
+    warn_queues = {}
+    crit_queues = {}
+    acquire_rabbitmq_queues.each do |queue|
+      if config[:filter]
+        next unless queue['name'].match(config[:filter])
+      end
+
+      # we don't care about empty queues and they'll have an infinite drain time so skip them
+      next if queue['messages'] == 0
+
+      # handle rate of zero which is an infinite time until empty
+      if queue['backing_queue_status']['avg_egress_rate'].to_f == 0
+        crit_queues[queue['name']]  = 'Infinite (drain rate = 0)'
+        next
+      end
+
+      secs_till_empty = queue['messages'] / queue['backing_queue_status']['avg_egress_rate']
+
+      # place warn / crit counts into hashes to be parsed for the alert message
+      if secs_till_empty > config[:critical].to_i
+        crit_queues[queue['name']] = secs_till_empty
+      elsif secs_till_empty > config[:warn].to_i
+        warn_queues[queue['name']] = secs_till_empty
+      end
+    end
+
+    # decide if we need to alert and build the message
+    if !crit_queues.empty?
+      critical "Drain time: #{crit_queues.map { |q, c| "#{q} #{c} sec" }.join(', ')}"
+    elsif !warn_queues.empty?
+      warning "Drain time: #{warn_queues.map { |q, c| "#{q} #{c} sec" }.join(', ')}"
+    else
+      ok "All (#{acquire_rabbitmq_queues.count}) queues will be drained in under #{config[:warn].to_i} seconds"
+    end
+  end
+end