sensu-plugins-elasticsearch-boutetnico 1.0.4

data/bin/metrics-es-node-graphite.rb

@@ -0,0 +1,354 @@
+#! /usr/bin/env ruby
+#
+#   es-node-graphite
+#
+# DESCRIPTION:
+#   This check creates node metrics from the elasticsearch API
+#
+# OUTPUT:
+#   metric data
+#
+# PLATFORMS:
+#   Linux, Windows, BSD, Solaris, etc
+#
+# DEPENDENCIES:
+#   gem: sensu-plugin
+#   gem: rest-client
+#
+# USAGE:
+#   #YELLOW
+#
+# NOTES:
+#   2014/04
+#   Modifid by Vincent Janelle @randomfrequency http://github.com/vjanelle
+#   Add more metrics, fix es 1.x URLs, translate graphite stats from
+#   names directly
+#
+#   2012/12 - Modified by Zach Dunn @SillySophist http://github.com/zadunn
+#   To add more metrics, and correct for new versins of ES. Tested on
+#   ES Version 0.19.8
+#
+# LICENSE:
+#   Copyright 2013 Vincent Janelle <randomfrequency@gmail.com>
+#   Copyright 2012 Sonian, Inc <chefs@sonian.net>
+#   Released under the same terms as Sensu (the MIT license); see LICENSE
+#   for details.
+#
+
+require 'sensu-plugin/metric/cli'
+require 'rest-client'
+require 'json'
+require 'base64'
+
+#
+# ES Node Graphite Metrics
+#
+class ESNodeGraphiteMetrics < Sensu::Plugin::Metric::CLI::Graphite
+  option :scheme,
+         description: 'Metric naming scheme, text to prepend to queue_name.metric',
+         short: '-s SCHEME',
+         long: '--scheme SCHEME',
+         default: "#{Socket.gethostname}.elasticsearch"
+
+  option :server,
+         description: 'Elasticsearch server host.',
+         short: '-h HOST',
+         long: '--host HOST',
+         default: 'localhost'
+
+  option :port,
+         description: 'Elasticsearch port.',
+         short: '-p PORT',
+         long: '--port PORT',
+         proc: proc(&:to_i),
+         default: 9200
+
+  option :timeout,
+         description: 'Request timeout to elasticsearch',
+         short: '-t TIMEOUT',
+         long: '--timeout TIMEOUT',
+         proc: proc(&:to_i),
+         default: 30
+
+  option :disable_jvm_stats,
+         description: 'Disable JVM statistics',
+         long: '--disable-jvm-stats',
+         boolean: true,
+         default: false
+
+  option :disable_os_stats,
+         description: 'Disable OS Stats',
+         long: '--disable-os-stat',
+         boolean: true,
+         default: false
+
+  option :disable_process_stats,
+         description: 'Disable process statistics',
+         long: '--disable-process-stats',
+         boolean: true,
+         default: false
+
+  option :disable_thread_pool_stats,
+         description: 'Disable thread-pool statistics',
+         long: '--disable-thread-pool-stats',
+         boolean: true,
+         default: false
+
+  option :disable_fs_stats,
+         description: 'Disable filesystem statistics',
+         long: '--disable-fs-stats',
+         boolean: true,
+         default: false
+
+  option :user,
+         description: 'Elasticsearch User',
+         short: '-u USER',
+         long: '--user USER'
+
+  option :password,
+         description: 'Elasticsearch Password',
+         short: '-P PASS',
+         long: '--password PASS'
+
+  option :https,
+         description: 'Enables HTTPS',
+         short: '-e',
+         long: '--https'
+
+  option :cert_file,
+         description: 'Cert file to use',
+         long: '--cert-file CERT_FILE'
+
+  def get_es_resource(resource)
+    headers = {}
+    if config[:user] && config[:password]
+      auth = 'Basic ' + Base64.strict_encode64("#{config[:user]}:#{config[:password]}").chomp
+      headers = { 'Authorization' => auth }
+    end
+
+    protocol = if config[:https]
+                 'https'
+               else
+                 'http'
+               end
+
+    r = if config[:cert_file]
+          RestClient::Resource.new("#{protocol}://#{config[:host]}:#{config[:port]}#{resource}",
+                                   ssl_ca_file: config[:cert_file].to_s,
+                                   timeout: config[:timeout],
+                                   headers: headers)
+        else
+          RestClient::Resource.new("#{protocol}://#{config[:server]}:#{config[:port]}#{resource}?pretty",
+                                   timeout: config[:timeout],
+                                   headers: headers)
+        end
+    ::JSON.parse(r.get)
+  rescue Errno::ECONNREFUSED
+    warning 'Connection refused'
+  rescue RestClient::RequestTimeout
+    warning 'Connection timed out'
+  end
+
+  def acquire_es_version
+    info = get_es_resource('/')
+    info['version']['number']
+  end
+
+  def run
+    # invert various stats depending on if some flags are set
+    os_stat = !config[:disable_os_stats]
+    process_stats = !config[:disable_process_stats]
+    jvm_stats = !config[:disable_jvm_stats]
+    tp_stats = !config[:disable_thread_pool_stats]
+    fs_stats = !config[:disable_fs_stats]
+
+    es_version = Gem::Version.new(acquire_es_version)
+
+    if es_version >= Gem::Version.new('3.0.0')
+      stats_query_array = %w[indices http transport]
+      stats_query_array.push('jvm') if jvm_stats == true
+      stats_query_array.push('os') if os_stat == true
+      stats_query_array.push('process') if process_stats == true
+      stats_query_array.push('thread_pool') if tp_stats == true
+      stats_query_array.push('fs') if fs_stats == true
+      stats_query_string = stats_query_array.join(',')
+    elsif es_version >= Gem::Version.new('1.0.0')
+      stats_query_array = %w[indices http network transport thread_pool]
+      stats_query_array.push('jvm') if jvm_stats == true
+      stats_query_array.push('os') if os_stat == true
+      stats_query_array.push('process') if process_stats == true
+      stats_query_array.push('tp_stats') if tp_stats == true
+      stats_query_array.push('fs_stats') if fs_stats == true
+      stats_query_string = stats_query_array.join(',')
+    else
+      stats_query_string = [
+        'clear=true',
+        'indices=true',
+        'http=true',
+        "jvm=#{jvm_stats}",
+        'network=true',
+        "os=#{os_stat}",
+        "process=#{process_stats}",
+        "thread_pool=#{tp_stats}",
+        'transport=true',
+        'thread_pool=true',
+        "fs=#{fs_stats}"
+      ].join('&')
+    end
+
+    stats = if es_version >= Gem::Version.new('3.0.0')
+              get_es_resource("/_nodes/_local/stats/#{stats_query_string}")
+            elsif es_version >= Gem::Version.new('1.0.0')
+              get_es_resource("/_nodes/_local/stats?#{stats_query_string}")
+            else
+              get_es_resource("/_cluster/nodes/_local/stats?#{stats_query_string}")
+            end
+
+    timestamp = Time.now.to_i
+    node = stats['nodes'].values.first
+
+    metrics = {}
+
+    if os_stat
+      if es_version >= Gem::Version.new('2.0.0')
+        metrics['os.load_average']                  = node['os']['load_average']
+      else
+        metrics['os.load_average']                  = node['os']['load_average'][0]
+        metrics['os.load_average.1']                = node['os']['load_average'][0]
+        metrics['os.load_average.5']                = node['os']['load_average'][1]
+        metrics['os.load_average.15']               = node['os']['load_average'][2]
+        metrics['os.cpu.sys']                       = node['os']['cpu']['sys']
+        metrics['os.cpu.user']                      = node['os']['cpu']['user']
+        metrics['os.cpu.idle']                      = node['os']['cpu']['idle']
+        metrics['os.cpu.usage']                     = node['os']['cpu']['usage']
+        metrics['os.cpu.stolen']                    = node['os']['cpu']['stolen']
+        metrics['os.uptime']                        = node['os']['uptime_in_millis']
+      end
+      metrics['os.mem.free_in_bytes']             = node['os']['mem']['free_in_bytes']
+    end
+
+    if process_stats
+      metrics['process.cpu.percent']              = node['process']['cpu']['percent']
+      metrics['process.mem.resident_in_bytes']    = node['process']['mem']['resident_in_bytes'] if node['process']['mem']['resident_in_bytes']
+    end
+
+    if jvm_stats
+      metrics['jvm.mem.heap_used_in_bytes']       = node['jvm']['mem']['heap_used_in_bytes']
+      metrics['jvm.mem.non_heap_used_in_bytes']   = node['jvm']['mem']['non_heap_used_in_bytes']
+      metrics['jvm.mem.max_heap_size_in_bytes']   = 0
+
+      node['jvm']['mem']['pools'].each do |k, v|
+        metrics["jvm.mem.#{k.tr(' ', '_')}.max_in_bytes"] = v['max_in_bytes']
+        metrics['jvm.mem.max_heap_size_in_bytes'] += v['max_in_bytes']
+      end
+
+      # This makes absolutely no sense - not sure what it's trying to measure - @vjanelle
+      # metrics['jvm.gc.collection_time_in_millis'] = node['jvm']['gc']['collection_time_in_millis'] + \
+      # node['jvm']['mem']['pools']['CMS Old Gen']['max_in_bytes']
+
+      node['jvm']['gc']['collectors'].each do |gc, gc_value|
+        gc_value.each do |k, v|
+          # this contains stupid things like '28ms' and '2s', and there's already
+          # something that counts in millis, which makes more sense
+          unless k.end_with? 'collection_time'
+            metrics["jvm.gc.collectors.#{gc}.#{k}"] = v
+          end
+        end
+      end
+
+      metrics['jvm.threads.count']                = node['jvm']['threads']['count']
+      metrics['jvm.threads.peak_count']           = node['jvm']['threads']['peak_count']
+      metrics['jvm.uptime']                       = node['jvm']['uptime_in_millis']
+    end
+
+    node['indices'].each do |type, index|
+      index.each do |k, v|
+        # #YELLOW
+        if k.end_with? 'is_throttled'
+          metrics["indices.#{type}.#{k}"] = true?(v) ? 1 : 0
+        elsif !(k =~ /(_time$)/ || v =~ /\d+/)
+          metrics["indices.#{type}.#{k}"] = v
+        end
+      end
+    end
+
+    node['transport'].each do |k, v|
+      # #YELLOW
+      unless k =~ /(_size$)/
+        metrics["transport.#{k}"] = v
+      end
+    end
+
+    metrics['http.current_open']                = node['http']['current_open']
+    metrics['http.total_opened']                = node['http']['total_opened']
+
+    if node['network']
+      metrics['network.tcp.active_opens']         = node['network']['tcp']['active_opens']
+      metrics['network.tcp.passive_opens']        = node['network']['tcp']['passive_opens']
+
+      metrics['network.tcp.in_segs']              = node['network']['tcp']['in_segs']
+      metrics['network.tcp.out_segs']             = node['network']['tcp']['out_segs']
+      metrics['network.tcp.retrans_segs']         = node['network']['tcp']['retrans_segs']
+      metrics['network.tcp.attempt_fails']        = node['network']['tcp']['attempt_fails']
+      metrics['network.tcp.in_errs']              = node['network']['tcp']['in_errs']
+      metrics['network.tcp.out_rsts']             = node['network']['tcp']['out_rsts']
+
+      metrics['network.tcp.curr_estab']           = node['network']['tcp']['curr_estab']
+      metrics['network.tcp.estab_resets']         = node['network']['tcp']['estab_resets']
+    end
+
+    if tp_stats
+      node['thread_pool'].each do |pool, stat|
+        stat.each do |k, v|
+          metrics["thread_pool.#{pool}.#{k}"] = v
+        end
+      end
+    end
+
+    if fs_stats
+      node['fs'].each do |fs, fs_value|
+        unless fs =~ /(timestamp|data)/
+          metrics_fs = hash_to_dotted_path(fs_value, "#{fs}.")
+          metrics_fs.each do |k, v|
+            metrics["fs.#{k}"] = v
+          end
+        end
+      end
+    end
+
+    metrics.each do |k, v|
+      if v.is_a? Numeric
+        output([config[:scheme], k].join('.'), v, timestamp)
+      end
+    end
+    ok
+  end
+end
+
+def hash_to_dotted_path(hash, path = '')
+  hash.each_with_object({}) do |(k, v), ret|
+    key = path + k.to_s
+    if v.is_a? Hash
+      ret.merge! hash_to_dotted_path(v, "#{key}.")
+    elsif v.is_a? Array
+      v.each do |element|
+        if element['device_name']
+          key2 = "#{key}.#{element['device_name']}"
+          ret.merge! hash_to_dotted_path(element, "#{key2}.")
+        end
+      end
+    else
+      ret[key] = v
+    end
+  end
+end
+
+def true?(obj)
+  if obj.to_s == 'true'
+    true
+  elsif obj.to_s == 'false'
+    false
+  else
+    "#{obj} is not a truthy value, please open an issue with this output so we can fix it"
+  end
+end

data/bin/metrics-es-node.rb

@@ -0,0 +1,143 @@
+#! /usr/bin/env ruby
+#
+#   es-node-metrics
+#
+# DESCRIPTION:
+#   This plugin uses the ES API to collect metrics, producing a JSON
+#   document which is outputted to STDOUT. An exit status of 0 indicates
+#   the plugin has successfully collected and produced.
+#
+# OUTPUT:
+#   metric data
+#
+# PLATFORMS:
+#   Linux
+#
+# DEPENDENCIES:
+#   gem: sensu-plugin
+#   gem: rest-client
+#
+# USAGE:
+#   #YELLOW
+#
+# NOTES:
+#
+# LICENSE:
+#   Copyright 2011 Sonian, Inc <chefs@sonian.net>
+#   Released under the same terms as Sensu (the MIT license); see LICENSE
+#   for details.
+#
+
+require 'sensu-plugin/metric/cli'
+require 'rest-client'
+require 'json'
+require 'base64'
+
+#
+# ES Node Metrics
+#
+class ESMetrics < Sensu::Plugin::Metric::CLI::Graphite
+  option :scheme,
+         description: 'Metric naming scheme, text to prepend to queue_name.metric',
+         short: '-s SCHEME',
+         long: '--scheme SCHEME',
+         default: "#{Socket.gethostname}.elasticsearch"
+
+  option :host,
+         description: 'Elasticsearch server host.',
+         short: '-h HOST',
+         long: '--host HOST',
+         default: 'localhost'
+
+  option :port,
+         description: 'Elasticsearch port',
+         short: '-p PORT',
+         long: '--port PORT',
+         proc: proc(&:to_i),
+         default: 9200
+
+  option :user,
+         description: 'Elasticsearch User',
+         short: '-u USER',
+         long: '--user USER'
+
+  option :password,
+         description: 'Elasticsearch Password',
+         short: '-P PASS',
+         long: '--password PASS'
+
+  option :https,
+         description: 'Enables HTTPS',
+         short: '-e',
+         long: '--https'
+
+  option :cert_file,
+         description: 'Cert file to use',
+         long: '--cert CERT_FILE'
+
+  def acquire_es_version
+    info = get_es_resource('/')
+    info['version']['number']
+  end
+
+  def get_es_resource(resource)
+    headers = {}
+    if config[:user] && config[:password]
+      auth = 'Basic ' + Base64.strict_encode64("#{config[:user]}:#{config[:password]}").chomp
+      headers = { 'Authorization' => auth }
+    end
+
+    protocol = if config[:https]
+                 'https'
+               else
+                 'http'
+               end
+
+    r = if config[:cert_file]
+          RestClient::Resource.new("#{protocol}://#{config[:host]}:#{config[:port]}#{resource}",
+                                   ssl_ca_file: config[:cert_file].to_s,
+                                   timeout: config[:timeout],
+                                   headers: headers)
+        else
+          RestClient::Resource.new("#{protocol}://#{config[:host]}:#{config[:port]}#{resource}",
+                                   timeout: config[:timeout],
+                                   headers: headers)
+        end
+    ::JSON.parse(r.get)
+  rescue Errno::ECONNREFUSED
+    warning 'Connection refused'
+  rescue RestClient::RequestTimeout
+    warning 'Connection timed out'
+  end
+
+  def run
+    es_version = Gem::Version.new(acquire_es_version)
+
+    if es_version >= Gem::Version.new('1.0.0')
+      ln = get_es_resource('/_nodes/_local')
+      stats = get_es_resource('/_nodes/_local/stats')
+    else
+      ln = get_es_resource('/_cluster/nodes/_local')
+      stats = get_es_resource('/_cluster/nodes/_local/stats')
+    end
+
+    timestamp = Time.now.to_i
+    node = stats['nodes'].values.first
+    node['jvm']['mem']['heap_max_in_bytes'] = ln['nodes'].values.first['jvm']['mem']['heap_max_in_bytes']
+    metrics = {}
+    metrics['os.load_average'] = if es_version >= Gem::Version.new('2.0.0')
+                                   node['os']['load_average']
+                                 else
+                                   node['os']['load_average'][0]
+                                 end
+    metrics['os.mem.free_in_bytes'] = node['os']['mem']['free_in_bytes']
+    metrics['process.mem.resident_in_bytes'] = node['process']['mem']['resident_in_bytes']
+    metrics['jvm.mem.heap_used_in_bytes'] = node['jvm']['mem']['heap_used_in_bytes']
+    metrics['jvm.mem.non_heap_used_in_bytes'] = node['jvm']['mem']['non_heap_used_in_bytes']
+    metrics['jvm.gc.collection_time_in_millis'] = node['jvm']['gc']['collection_time_in_millis']
+    metrics.each do |k, v|
+      output([config[:scheme], k].join('.'), v, timestamp)
+    end
+    ok
+  end
+end