riemann-tools 1.1.1 → 1.2.0

data/tools/riemann-chronos/bin/riemann-chronos

@@ -3,159 +3,6 @@
 
 Process.setproctitle($PROGRAM_NAME)
 
-require 'riemann/tools'
+require 'riemann/tools/chronos'
 
-module Riemann
-  module Tools
-    class Chronos
-      include Riemann::Tools
-
-      require 'faraday'
-      require 'json'
-      require 'uri'
-
-      opt :read_timeout, 'Faraday read timeout', type: :int, default: 2
-      opt :open_timeout, 'Faraday open timeout', type: :int, default: 1
-      opt :path_prefix,
-          'Chronos path prefix for proxied installations e.g. "chronos" for target http://localhost/chronos/metrics', default: '/'
-      opt :chronos_host, 'Chronos host', default: 'localhost'
-      opt :chronos_port, 'Chronos port', type: :int, default: 4400
-
-      def initialize
-        options[:interval] = 60
-        options[:ttl] = 120
-      end
-
-      # 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 StandardError => e
-          report(
-            host: uri.host,
-            service: 'chronos health',
-            state: 'critical',
-            description: "HTTP connection error: #{e.class} - #{e.message}",
-          )
-        end
-        response
-      end
-
-      def health_url
-        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[:chronos_host]}:#{options[:chronos_port]}#{path_prefix.length.positive? ? '/' : ''}#{path_prefix}/metrics"
-      end
-
-      def jobs_url
-        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[:chronos_host]}:#{options[:chronos_port]}#{path_prefix.length.positive? ? '/' : ''}#{path_prefix}/scheduler/jobs"
-      end
-
-      def tick
-        tick_health
-        tick_jobs
-      end
-
-      def tick_health
-        uri = URI(health_url)
-        response = safe_get(uri)
-
-        return if response.nil?
-
-        if response.status != 200
-          report(
-            host: uri.host,
-            service: 'chronos health',
-            state: 'critical',
-            description: "HTTP connection error: #{response.status} - #{response.body}",
-          )
-        else
-          # Assuming that a 200 will give json
-          json = JSON.parse(response.body)
-
-          report(
-            host: uri.host,
-            service: 'chronos health',
-            state: 'ok',
-          )
-
-          json.each_pair do |t, d|
-            next unless d.respond_to? :each_pair
-
-            d.each_pair do |service, counters|
-              report(
-                host: uri.host,
-                service: "chronos_metric #{t} #{service}",
-                metric: 1,
-                tags: ['metric_name'],
-                ttl: 600,
-              )
-              next unless counters.respond_to? :each_pair
-
-              counters.each_pair do |k, v|
-                next unless v.is_a? Numeric
-
-                report(
-                  host: uri.host,
-                  service: "chronos #{service} #{k}",
-                  metric: v,
-                  tags: ['metric', t.to_s],
-                  ttl: 600,
-                )
-              end
-            end
-          end
-        end
-      end
-
-      def tick_jobs
-        uri = URI(jobs_url)
-        response = safe_get(uri)
-
-        return if response.nil?
-
-        if response.status != 200
-          report(
-            host: uri.host,
-            service: 'chronos health',
-            state: 'critical',
-            description: "HTTP connection error: #{response.status} - #{response.body}",
-          )
-        else
-          # Assuming that a 200 will give json
-          json = JSON.parse(response.body)
-
-          report(
-            host: uri.host,
-            service: 'chronos health',
-            state: 'ok',
-          )
-
-          json.each do |job|
-            job.each_pair do |k, v|
-              next unless v.is_a? Numeric
-
-              report(
-                host: uri.host,
-                service: "chronos job #{job['name']} #{k}",
-                metric: v,
-                ttl: 120,
-              )
-            end
-          end
-        end
-      end
-    end
-  end
-end
 Riemann::Tools::Chronos.run

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

@@ -0,0 +1,157 @@
+# frozen_string_literal: true
+
+require 'riemann/tools'
+
+module Riemann
+  module Tools
+    class Chronos
+      include Riemann::Tools
+
+      require 'faraday'
+      require 'json'
+      require 'uri'
+
+      opt :read_timeout, 'Faraday read timeout', type: :int, default: 2
+      opt :open_timeout, 'Faraday open timeout', type: :int, default: 1
+      opt :path_prefix,
+          'Chronos path prefix for proxied installations e.g. "chronos" for target http://localhost/chronos/metrics', default: '/'.dup
+      opt :chronos_host, 'Chronos host', default: 'localhost'
+      opt :chronos_port, 'Chronos port', type: :int, default: 4400
+
+      def initialize
+        options[:interval] = 60
+        options[:ttl] = 120
+      end
+
+      # 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 StandardError => e
+          report(
+            host: uri.host,
+            service: 'chronos health',
+            state: 'critical',
+            description: "HTTP connection error: #{e.class} - #{e.message}",
+          )
+        end
+        response
+      end
+
+      def health_url
+        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[:chronos_host]}:#{options[:chronos_port]}#{path_prefix.length.positive? ? '/' : ''}#{path_prefix}/metrics"
+      end
+
+      def jobs_url
+        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[:chronos_host]}:#{options[:chronos_port]}#{path_prefix.length.positive? ? '/' : ''}#{path_prefix}/scheduler/jobs"
+      end
+
+      def tick
+        tick_health
+        tick_jobs
+      end
+
+      def tick_health
+        uri = URI(health_url)
+        response = safe_get(uri)
+
+        return if response.nil?
+
+        if response.status != 200
+          report(
+            host: uri.host,
+            service: 'chronos health',
+            state: 'critical',
+            description: "HTTP connection error: #{response.status} - #{response.body}",
+          )
+        else
+          # Assuming that a 200 will give json
+          json = JSON.parse(response.body)
+
+          report(
+            host: uri.host,
+            service: 'chronos health',
+            state: 'ok',
+          )
+
+          json.each_pair do |t, d|
+            next unless d.respond_to? :each_pair
+
+            d.each_pair do |service, counters|
+              report(
+                host: uri.host,
+                service: "chronos_metric #{t} #{service}",
+                metric: 1,
+                tags: ['metric_name'],
+                ttl: 600,
+              )
+              next unless counters.respond_to? :each_pair
+
+              counters.each_pair do |k, v|
+                next unless v.is_a? Numeric
+
+                report(
+                  host: uri.host,
+                  service: "chronos #{service} #{k}",
+                  metric: v,
+                  tags: ['metric', t.to_s],
+                  ttl: 600,
+                )
+              end
+            end
+          end
+        end
+      end
+
+      def tick_jobs
+        uri = URI(jobs_url)
+        response = safe_get(uri)
+
+        return if response.nil?
+
+        if response.status != 200
+          report(
+            host: uri.host,
+            service: 'chronos health',
+            state: 'critical',
+            description: "HTTP connection error: #{response.status} - #{response.body}",
+          )
+        else
+          # Assuming that a 200 will give json
+          json = JSON.parse(response.body)
+
+          report(
+            host: uri.host,
+            service: 'chronos health',
+            state: 'ok',
+          )
+
+          json.each do |job|
+            job.each_pair do |k, v|
+              next unless v.is_a? Numeric
+
+              report(
+                host: uri.host,
+                service: "chronos job #{job['name']} #{k}",
+                metric: v,
+                ttl: 120,
+              )
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/tools/riemann-docker/Rakefile

@@ -11,10 +11,8 @@ ENV['COPYFILE_DISABLE'] = 'true'
 
 # Gemspec
 gemspec = Gem::Specification.new do |s|
-  s.rubyforge_project = 'riemann-docker'
-
   s.name = 'riemann-docker'
-  s.version = '0.1.3'
+  s.version = '0.1.5'
   s.author = 'Shani Elharrar'
   s.email = ''
   s.homepage = 'https://github.com/riemann/riemann-tools'
@@ -22,14 +20,13 @@ gemspec = Gem::Specification.new do |s|
   s.summary = 'Submits Docker container stats to riemann.'
   s.license = 'MIT'
 
-  s.add_dependency 'riemann-tools', '>= 0.2.13'
-  s.add_dependency 'docker-api', '>= 1.22.0'
+  s.add_runtime_dependency 'riemann-tools', '~> 1.0', '>= 1.1.1'
+  s.add_runtime_dependency 'docker-api', '~> 1.22', '>= 1.22.0'
 
-  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-docker/bin/riemann-docker

@@ -3,204 +3,6 @@
 
 Process.setproctitle($PROGRAM_NAME)
 
-# Reports current CPU, disk, load average, and memory use to riemann.
+require 'riemann/tools/docker'
 
-require 'riemann/tools'
-
-module Riemann
-  module Tools
-    class 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: %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
-
-Riemann::Tools::DockerHealth.run
+Riemann::Tools::Docker.run