riemann-tools 1.0.0 → 1.2.0

data/tools/riemann-docker/bin/riemann-docker

@@ -1,217 +1,8 @@
 #!/usr/bin/env ruby
-Process.setproctitle($0)
+# frozen_string_literal: true
 
-# Reports current CPU, disk, load average, and memory use to riemann.
+Process.setproctitle($PROGRAM_NAME)
 
-require 'riemann/tools'
-
-class Riemann::Tools::DockerHealth
-  require 'docker'
-  require 'socket'
-  include Riemann::Tools
-  include Docker
-
-  opt :docker_host, "Docker Container Host (see https://github.com/swipely/docker-api#host)", :type => String, :default => nil
-  opt :cpu_warning, "CPU warning threshold (fraction of total jiffies)", :default => 0.9
-  opt :cpu_critical, "CPU critical threshold (fraction of total jiffies)", :default => 0.95
-  opt :disk_warning, "Disk warning threshold (fraction of space used)", :default => 0.9
-  opt :disk_critical, "Disk critical threshold (fraction of space used)", :default => 0.95
-  opt :memory_warning, "Memory warning threshold (fraction of RAM)", :default => 0.85
-  opt :memory_critical, "Memory critical threshold (fraction of RAM)", :default => 0.95
-  opt :host_hostname, "Suffix of host", :type => String, :default => nil
-  opt :checks, "A list of checks to run.", :type => :strings, :default => ['cpu', 'memory', 'disk', 'basic']
-
-  def get_containers
-    Docker::Container.all
-  end
-
-  def get_container_name(container)
-    container.json['Name'][1..-1]
-  end
-
-  def initialize
-
-    if (opts[:docker_host] != nil)
-      Docker.url = opts[:docker_host]
-    end
-
-    @hostname = opts[:host_hostname]
-    if (@hostname.nil? || !(@hostname.is_a? String) || @hostname.empty?)
-      @hostname = Socket.gethostname
-    end
-
-    @cpu_coefficient = 1000 * 1000 * 1000
-
-    @limits = {
-      :cpu => {:critical => opts[:cpu_critical], :warning => opts[:cpu_warning]},
-      :disk => {:critical => opts[:disk_critical], :warning => opts[:disk_warning]},
-      :memory => {:critical => opts[:memory_critical], :warning => opts[:memory_warning]}
-    }
-
-    @last_cpu_reads = Hash.new
-    @last_uptime_reads = Hash.new
-
-    opts[:checks].each do |check|
-      case check
-      when 'disk'
-        @disk_enabled = true
-      when 'cpu'
-        @cpu_enabled = true
-      when 'memory'
-        @memory_enabled = true
-      when 'basic'
-        @basic_inspection_enabled = true
-      end
-    end
-  end
-
-  def alert(container, service, state, metric, description)
-
-    opts = {  :service => service.to_s,
-              :state => state.to_s,
-              :metric => metric.to_f,
-              :description => description }
-
-    if (container != nil)
-      opts[:host] = "#{@hostname}-#{container}"
-    else
-      opts[:host] = @hostname
-    end
-
-    report(opts)
-  end
-
-  def report_pct(container, service, fraction, report = '', name = nil)
-    if fraction
-
-      if (name == nil)
-        name = service
-      end
-
-      if fraction > @limits[service][:critical]
-        alert container, name, :critical, fraction, "#{sprintf("%.2f", fraction * 100)}% #{report}"
-      elsif fraction > @limits[service][:warning]
-        alert container, name, :warning, fraction, "#{sprintf("%.2f", fraction * 100)}% #{report}"
-      else
-        alert container, name, :ok, fraction, "#{sprintf("%.2f", fraction * 100)}% #{report}"
-      end
-    end
-  end
-
-
-  def cpu(id, name, stats)
-
-    current = stats['precpu_stats']['cpu_usage']['total_usage'] / stats['precpu_stats']['cpu_usage']['percpu_usage'].count
-
-    unless current
-      alert name, :cpu, :unknown, nil, 'no total usage found in docker remote api stats'
-      return false
-    end
-
-    current_time = Time.parse(stats['read']);
-    if (@last_cpu_reads[id] != nil)
-      last = @last_cpu_reads[id]
-      used = (current - last[:v]) / (current_time - last[:t]) / @cpu_coefficient
-
-      report_pct name, :cpu, used
-    end
-
-    @last_cpu_reads[id] = { v: current, t: current_time }
-  end
-
-  def memory(id, name, stats)
-    memory_stats = stats['memory_stats']
-    usage = memory_stats['usage'].to_f
-    total = memory_stats['limit'].to_f
-    fraction = (usage / total)
-
-    report_pct name, :memory, fraction, "#{usage} / #{total}"
-  end
-
-  def disk
-    `df -P`.split(/\n/).each do |r|
-      f = r.split(/\s+/)
-      next if f[0] == 'Filesystem'
-      next unless f[0] =~ /\// # Needs at least one slash in the mount path
-
-      # Calculate capacity
-      x = f[4].to_f/100
-      report_pct(nil, :disk, x, "#{f[3].to_i / 1024} mb left", "disk #{f[5]}")
-    end
-  end
-
-  def basic_inspection(id, name, inspection)
-
-    state = inspection['State']
-    json_state = JSON.generate(state)
-
-    running = state['Running']
-
-    alert(name, "status",
-          running ? "ok" : "critical",
-          running ? 1 : 0,
-          json_state)
-
-    if (running)
-      start_time = DateTime.rfc3339(state['StartedAt']).to_time.utc.to_i
-      now = DateTime.now.to_time.utc.to_i
-      uptime = now - start_time
-
-      if (@last_uptime_reads[id] != nil)
-        last = @last_uptime_reads[id]
-        restarted = start_time != last
-        alert(name, "uptime",
-              restarted ? "critical" : "ok",
-              uptime,
-              "last 'StartedAt' measure was #{last} (#{Time.at(last).utc.to_s}), " +
-              "now it's #{start_time} (#{Time.at(start_time).utc.to_s})")
-      end
-
-      @last_uptime_reads[id] = start_time
-    end
-  end
-
-  def tick
-
-    # Disk is the same in every container
-    if @disk_enabled
-      disk()
-    end
-
-    # Get CPU, Memory and Load of each container
-    containers = get_containers()
-    threads = []
-
-    containers.each do |ctr|
-      threads << Thread.new(ctr) do |container|
-
-        id = container.id
-        name = get_container_name(container)
-
-        stats = Docker::Util.parse_json(container.connection.get("/containers/#{id}/stats", {stream:false}))
-
-        if @basic_inspection_enabled
-          inspection = Docker::Util.parse_json(container.connection.get("/containers/#{id}/json"))
-           basic_inspection(id, name, inspection)
-          end
-          if @cpu_enabled
-            cpu(id, name, stats)
-        end
-        if @memory_enabled
-          memory(id, name, stats)
-        end
-      end
-    end
-
-    threads.each do |thread|
-      begin
-        thread.join
-      rescue => e
-        $stderr.puts "#{e.class} #{e}\n#{e.backtrace.join "\n"}"
-      end
-    end
-  end
-end
-
-Riemann::Tools::DockerHealth.run
+require 'riemann/tools/docker'
 
+Riemann::Tools::Docker.run

data/tools/riemann-docker/lib/riemann/tools/docker.rb

@@ -0,0 +1,200 @@
+# frozen_string_literal: true
+
+require 'riemann/tools'
+
+# Reports current CPU, disk, load average, and memory use to riemann.
+module Riemann
+  module Tools
+    class Docker
+      require 'docker'
+      require 'socket'
+      include Riemann::Tools
+      include ::Docker
+
+      opt :docker_host, 'Docker Container Host (see https://github.com/swipely/docker-api#host)', type: String,
+                                                                                                  default: nil
+      opt :cpu_warning, 'CPU warning threshold (fraction of total jiffies)', default: 0.9
+      opt :cpu_critical, 'CPU critical threshold (fraction of total jiffies)', default: 0.95
+      opt :disk_warning, 'Disk warning threshold (fraction of space used)', default: 0.9
+      opt :disk_critical, 'Disk critical threshold (fraction of space used)', default: 0.95
+      opt :memory_warning, 'Memory warning threshold (fraction of RAM)', default: 0.85
+      opt :memory_critical, 'Memory critical threshold (fraction of RAM)', default: 0.95
+      opt :host_hostname, 'Suffix of host', type: String, default: nil
+      opt :checks, 'A list of checks to run.', type: :strings, default: %w[cpu memory disk basic]
+
+      def containers
+        Docker::Container.all
+      end
+
+      def get_container_name(container)
+        container.json['Name'][1..]
+      end
+
+      def initialize
+        Docker.url = opts[:docker_host] unless opts[:docker_host].nil?
+
+        @hostname = opts[:host_hostname]
+        @hostname = Socket.gethostname if @hostname.nil? || !(@hostname.is_a? String) || @hostname.empty?
+
+        @cpu_coefficient = 1000 * 1000 * 1000
+
+        @limits = {
+          cpu: { critical: opts[:cpu_critical], warning: opts[:cpu_warning] },
+          disk: { critical: opts[:disk_critical], warning: opts[:disk_warning] },
+          memory: { critical: opts[:memory_critical], warning: opts[:memory_warning] },
+        }
+
+        @last_cpu_reads = {}
+        @last_uptime_reads = {}
+
+        opts[:checks].each do |check|
+          case check
+          when 'disk'
+            @disk_enabled = true
+          when 'cpu'
+            @cpu_enabled = true
+          when 'memory'
+            @memory_enabled = true
+          when 'basic'
+            @basic_inspection_enabled = true
+          end
+        end
+      end
+
+      def alert(container, service, state, metric, description)
+        opts = {
+          service: service.to_s,
+          state: state.to_s,
+          metric: metric.to_f,
+          description: description,
+        }
+
+        opts[:host] = if !container.nil?
+                        "#{@hostname}-#{container}"
+                      else
+                        @hostname
+                      end
+
+        report(opts)
+      end
+
+      def report_pct(container, service, fraction, report = '', name = nil)
+        return unless fraction
+
+        name = service if name.nil?
+
+        if fraction > @limits[service][:critical]
+          alert container, name, :critical, fraction, "#{format('%.2f', fraction * 100)}% #{report}"
+        elsif fraction > @limits[service][:warning]
+          alert container, name, :warning, fraction, "#{format('%.2f', fraction * 100)}% #{report}"
+        else
+          alert container, name, :ok, fraction, "#{format('%.2f', fraction * 100)}% #{report}"
+        end
+      end
+
+      def cpu(id, name, stats)
+        current = stats['precpu_stats']['cpu_usage']['total_usage'] / stats['precpu_stats']['cpu_usage']['percpu_usage'].count
+
+        unless current
+          alert name, :cpu, :unknown, nil, 'no total usage found in docker remote api stats'
+          return false
+        end
+
+        current_time = Time.parse(stats['read'])
+        unless @last_cpu_reads[id].nil?
+          last = @last_cpu_reads[id]
+          used = (current - last[:v]) / (current_time - last[:t]) / @cpu_coefficient
+
+          report_pct name, :cpu, used
+        end
+
+        @last_cpu_reads[id] = { v: current, t: current_time }
+      end
+
+      def memory(_id, name, stats)
+        memory_stats = stats['memory_stats']
+        usage = memory_stats['usage'].to_f
+        total = memory_stats['limit'].to_f
+        fraction = (usage / total)
+
+        report_pct name, :memory, fraction, "#{usage} / #{total}"
+      end
+
+      def disk
+        `df -P`.split(/\n/).each do |r|
+          f = r.split(/\s+/)
+          next if f[0] == 'Filesystem'
+          next unless f[0] =~ %r{/} # Needs at least one slash in the mount path
+
+          # Calculate capacity
+          x = f[4].to_f / 100
+          report_pct(nil, :disk, x, "#{f[3].to_i / 1024} mb left", "disk #{f[5]}")
+        end
+      end
+
+      def basic_inspection(id, name, inspection)
+        state = inspection['State']
+        json_state = JSON.generate(state)
+
+        running = state['Running']
+
+        alert(
+          name, 'status',
+          running ? 'ok' : 'critical',
+          running ? 1 : 0,
+          json_state,
+        )
+
+        return unless running
+
+        start_time = DateTime.rfc3339(state['StartedAt']).to_time.utc.to_i
+        now = DateTime.now.to_time.utc.to_i
+        uptime = now - start_time
+
+        unless @last_uptime_reads[id].nil?
+          last = @last_uptime_reads[id]
+          restarted = start_time != last
+          alert(
+            name, 'uptime',
+            restarted ? 'critical' : 'ok',
+            uptime,
+            "last 'StartedAt' measure was #{last} (#{Time.at(last).utc}), " \
+            "now it's #{start_time} (#{Time.at(start_time).utc})",
+          )
+        end
+
+        @last_uptime_reads[id] = start_time
+      end
+
+      def tick
+        # Disk is the same in every container
+        disk if @disk_enabled
+
+        # Get CPU, Memory and Load of each container
+        threads = []
+
+        containers.each do |ctr|
+          threads << Thread.new(ctr) do |container|
+            id = container.id
+            name = get_container_name(container)
+
+            stats = Docker::Util.parse_json(container.connection.get("/containers/#{id}/stats", { stream: false }))
+
+            if @basic_inspection_enabled
+              inspection = Docker::Util.parse_json(container.connection.get("/containers/#{id}/json"))
+              basic_inspection(id, name, inspection)
+            end
+            cpu(id, name, stats) if @cpu_enabled
+            memory(id, name, stats) if @memory_enabled
+          end
+        end
+
+        threads.each do |thread|
+          thread.join
+        rescue StandardError => e
+          warn "#{e.class} #{e}\n#{e.backtrace.join "\n"}"
+        end
+      end
+    end
+  end
+end

data/tools/riemann-elasticsearch/{Rakefile.rb → Rakefile}

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 require 'rubygems'
 require 'rubygems/package_task'
 require 'rdoc/task'
@@ -9,10 +11,8 @@ ENV['COPYFILE_DISABLE'] = 'true'
 
 # Gemspec
 gemspec = Gem::Specification.new do |s|
-  s.rubyforge_project = 'riemann-elasticsearch'
-
   s.name = 'riemann-elasticsearch'
-  s.version = '0.2.4'
+  s.version = '0.2.5'
   s.author = 'Gavin Sandie'
   s.email = 'beach@vicecity.co.uk'
   s.homepage = 'https://github.com/riemann/riemann-tools'
@@ -20,15 +20,14 @@ gemspec = Gem::Specification.new do |s|
   s.summary = 'Submits elasticsearch stats to riemann.'
   s.license = 'MIT'
 
-  s.add_dependency 'riemann-tools', '>= 0.2.13'
-  s.add_dependency 'faraday', '>= 0.8.5'
-  s.add_dependency 'json'
+  s.add_runtime_dependency 'riemann-tools', '~> 1.0', '>= 1.1.1'
+  s.add_runtime_dependency 'faraday', '~> 2.3', '>= 2.3.0'
+  s.add_runtime_dependency 'json', '~> 2.6', '>=2.6.2'
 
-  s.files = FileList['bin/*', 'LICENSE', 'README.md'].to_a
+  s.files = FileList['bin/*', 'lib/**/*.rb', 'LICENSE', 'README.md'].to_a
   s.executables |= Dir.entries('bin/')
-  s.has_rdoc = false
 
-  s.required_ruby_version = '>= 1.8.7'
+  s.required_ruby_version = Gem::Requirement.new('>= 2.6.0')
 end
 
 Gem::PackageTask.new gemspec do |p|

data/tools/riemann-elasticsearch/bin/riemann-elasticsearch

@@ -1,166 +1,8 @@
 #!/usr/bin/env ruby
-Process.setproctitle($0)
+# frozen_string_literal: true
 
-require 'riemann/tools'
+Process.setproctitle($PROGRAM_NAME)
 
-class Riemann::Tools::Elasticsearch
-  include Riemann::Tools
-  require 'faraday'
-  require 'json'
-  require 'uri'
+require 'riemann/tools/elasticsearch'
 
-  opt :read_timeout, 'Faraday read timeout', type: :int, default: 2
-  opt :open_timeout, 'Faraday open timeout', type: :int, default: 1
-  opt :path_prefix, 'Elasticsearch path prefix for proxied installations e.g. "els" for target http://localhost/els/_cluster/health', default: "/"
-  opt :es_host, 'Elasticsearch host', default: "localhost"
-  opt :es_port, 'Elasticsearch port', type: :int, default: 9200
-  opt :es_search_index, 'Elasticsearch index to fetch search statistics for', default: "_all"
-
-
-  # Handles HTTP connections and GET requests safely
-  def safe_get(uri)
-      # Handle connection timeouts
-      response = nil
-      begin
-        connection = Faraday.new(uri)
-        response = connection.get do |req|
-          req.options[:timeout] = options[:read_timeout]
-          req.options[:open_timeout] = options[:open_timeout]
-        end
-      rescue => e
-        report(:host => uri.host,
-          :service => "elasticsearch health",
-          :state => "critical",
-          :description => "HTTP connection error: #{e.class} - #{e.message}"
-        )
-      end
-      response
-  end
-
-  def make_es_url(path)
-    path_prefix = options[:path_prefix]
-    path_prefix[0] = '' if path_prefix[0]=='/'
-    path_prefix[path_prefix.length-1] = '' if path_prefix[path_prefix.length-1]=='/'
-    "http://#{options[:es_host]}:#{options[:es_port]}#{path_prefix.length>0?'/':''}#{path_prefix}/#{path}"
-  end
-
-  def health_url
-    make_es_url("_cluster/health")
-  end
-
-  def indices_url
-    make_es_url("_stats/store")
-  end
-
-  def search_url
-    es_search_index = options[:es_search_index]
-    make_es_url("#{es_search_index}/_stats/search")
-  end
-
-  def is_bad?(response, uri)
-    if response.success?
-      false
-    else
-      report(:host => uri.host,
-             :service => "elasticsearch health",
-             :state => "critical",
-             :description => response.nil? ? "HTTP response is empty!" : "HTTP connection error: #{response.status} - #{response.body}"
-      )
-    end
-  end
-
-  def tick_indices
-    uri = URI(indices_url)
-    response = safe_get(uri)
-
-    return if is_bad?(response, uri)
-
-    # Assuming that a 200 will give json
-    json = JSON.parse(response.body)
-
-    json["indices"].each_pair do |k,v|
-      report(:host => uri.host,
-             :service => "elasticsearch index/#{k}/primaries/size_in_bytes",
-             :metric => v["primaries"]["store"]["size_in_bytes"]
-      )
-      report(:host => uri.host,
-             :service => "elasticsearch index/#{k}/total/size_in_bytes",
-             :metric => v["total"]["store"]["size_in_bytes"]
-      )
-    end
-  end
-
-  def tick_search
-    uri = URI(search_url)
-    response = safe_get(uri)
-
-    return if is_bad?(response, uri)
-
-    es_search_index = options[:es_search_index]
-    # Assuming that a 200 will give json
-    json = JSON.parse(response.body)
-
-    json["_all"].each_pair do |type, data|
-      query = data["search"]["query_time_in_millis"].to_f / data["search"]["query_total"].to_f
-      fetch = data["search"]["fetch_time_in_millis"].to_f / data["search"]["fetch_total"].to_f
-
-      report(:host => uri.host,
-             :service => "elasticsearch search/#{es_search_index}/query",
-             :metric => query
-      )
-      report(:host => uri.host,
-             :service => "elasticsearch search/#{es_search_index}/fetch",
-             :metric => fetch
-      )
-    end
-  end
-
-  def tick
-    begin
-      tick_indices
-      tick_search
-    rescue Exception => e
-      report(:host => options[:es_host],
-             :service => "elasticsearch error",
-             :state => "critical",
-             :description => "Elasticsearch cluster error: #{e.message}")
-    end
-    uri = URI(health_url)
-    response = safe_get(uri)
-
-    return if is_bad?(response, uri)
-
-    # Assuming that a 200 will give json
-    json = JSON.parse(response.body)
-    cluster_name = json.delete("cluster_name")
-    cluster_status = json.delete("status")
-    state = case cluster_status
-              when "green"
-                "ok"
-              when "yellow"
-                "warning"
-              when "red"
-                "critical"
-            end
-
-    report(:host => uri.host,
-           :service => "elasticsearch health",
-           :state => state,
-           :description => "Elasticsearch cluster: #{cluster_name} - #{cluster_status}")
-
-    json.each_pair do |k,v|
-      report(:host => uri.host,
-             :service => "elasticsearch #{k}",
-             :metric => v,
-             :description => "Elasticsearch cluster #{k}"
-      )
-
-    end
-
-  end
-
-
-
-end
 Riemann::Tools::Elasticsearch.run
-