riemann-tools 1.1.1 → 1.4.0

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

@@ -0,0 +1,317 @@
+# frozen_string_literal: true
+
+require 'English'
+require 'riemann/tools'
+
+# Forwards information on a Riak node to Riemann.
+module Riemann
+  module Tools
+    class Riak
+      include Riemann::Tools
+      require 'net/http'
+      require 'net/https'
+      require 'yajl/json_gem'
+
+      opt :riak_host, 'Riak host for stats <IP> or SSL http(s)://<IP>', default: Socket.gethostname
+      opt :data_dir, 'Riak data directory', default: '/var/lib/riak'
+      opt :stats_port, 'Riak HTTP port for stats', default: 8098
+      opt :stats_path, 'Riak HTTP stats path', default: '/stats'
+      opt :node_name, 'Riak erlang node name', default: "riak@#{Socket.gethostname}"
+      opt :cookie, 'Riak cookie to use', default: 'riak'
+
+      opt :get_50_warning, 'FSM 50% get time warning threshold (ms)', default: 1000
+      opt :put_50_warning, 'FSM 50% put time warning threshold (ms)', default: 1000
+      opt :get_95_warning, 'FSM 95% get time warning threshold (ms)', default: 2000
+      opt :put_95_warning, 'FSM 95% put time warning threshold (ms)', default: 2000
+      opt :get_99_warning, 'FSM 99% get time warning threshold (ms)', default: 10_000
+      opt :put_99_warning, 'FSM 99% put time warning threshold (ms)', default: 10_000
+
+      def initialize
+        detect_features
+
+        @httpstatus = true
+
+        begin
+          uri = URI.parse(opts[:riak_host])
+          uri.host = opts[:riak_host] if uri.host.nil?
+          http = ::Net::HTTP.new(uri.host, opts[:stats_port])
+          http.use_ssl = uri.scheme == 'https'
+          http.verify_mode = OpenSSL::SSL::VERIFY_NONE if http.use_ssl?
+          http.start do |h|
+            h.get opts[:stats_path]
+          end
+        rescue StandardError => _e
+          @httpstatus = false
+        end
+
+        # we're going to override the emulator setting to allow users to
+        # dynamically input the cookie
+        # this is done only once - hopefully it doesn't get overridden.
+        ENV['ERL_AFLAGS'] = "-setcookie #{opts[:cookie]}"
+      end
+
+      # Identifies whether escript and riak-admin are installed
+      def detect_features
+        @escript = true # Whether escript is present on this machine
+        @riakadmin = true # Whether riak-admin is present
+
+        @escript = false if `which escript` =~ /^\s*$/
+
+        @riakadmin = false if `which riak-admin` =~ /^\s*$/
+      end
+
+      def check_ring
+        str = if @escript
+                `riemann-riak-ring #{opts[:node_name]}`.chomp
+              elsif @riakadmin
+                `riak-admin ringready`
+              end
+
+        return if str.nil?
+
+        if str =~ /^TRUE/
+          report(
+            host: opts[:riak_host],
+            service: 'riak ring',
+            state: 'ok',
+            description: str,
+          )
+        else
+          report(
+            host: opts[:riak_host],
+            service: 'riak ring',
+            state: 'warning',
+            description: str,
+          )
+        end
+      end
+
+      def check_keys
+        keys = `riemann-riak-keys #{opts[:node_name]}`.chomp
+        if keys =~ /^\d+$/
+          report(
+            host: opts[:riak_host],
+            service: 'riak keys',
+            state: 'ok',
+            metric: keys.to_i,
+            description: keys,
+          )
+        else
+          report(
+            host: opts[:riak_host],
+            service: 'riak keys',
+            state: 'unknown',
+            description: keys,
+          )
+        end
+      end
+
+      def check_transfers
+        str = (`riak-admin transfers` if @riakadmin)
+
+        return if str.nil?
+
+        if str =~ /'#{opts[:node_name]}' waiting to handoff (\d+) partitions/
+          report(
+            host: opts[:riak_host],
+            service: 'riak transfers',
+            state: 'critical',
+            metric: Regexp.last_match(1).to_i,
+            description: "waiting to handoff #{Regexp.last_match(1)} partitions",
+          )
+        else
+          report(
+            host: opts[:riak_host],
+            service: 'riak transfers',
+            state: 'ok',
+            metric: 0,
+            description: 'No pending transfers',
+          )
+        end
+      end
+
+      def check_disk
+        gb = `du -Ls #{opts[:data_dir]}`.split(/\s+/).first.to_i / (1024.0**2)
+        report(
+          host: opts[:riak_host],
+          service: 'riak disk',
+          state: 'ok',
+          metric: gb,
+          description: "#{gb} GB in #{opts[:data_dir]}",
+        )
+      end
+
+      # Returns the riak stat for the given fsm type and percentile.
+      def fsm_stat(type, property, percentile)
+        "node_#{type}_fsm_#{property}_#{percentile == 50 ? 'median' : percentile}"
+      end
+
+      # Returns the alerts state for the given fsm.
+      def fsm_state(type, percentile, val)
+        limit = opts["#{type}_#{percentile}_warning".to_sym]
+        case val
+        when 0..limit
+          'ok'
+        when limit..limit * 2
+          'warning'
+        else
+          'critical'
+        end
+      end
+
+      # Get current stats via HTTP
+      def stats_http
+        begin
+          uri = URI.parse(opts[:riak_host])
+          uri.host = opts[:riak_host] if uri.host.nil?
+          http = ::Net::HTTP.new(uri.host, opts[:stats_port])
+          http.use_ssl = uri.scheme == 'https'
+          http.verify_mode = OpenSSL::SSL::VERIFY_NONE if http.use_ssl?
+          res = http.start do |h|
+            h.get opts[:stats_path]
+          end
+        rescue StandardError => e
+          report(
+            host: opts[:riak_host],
+            service: 'riak',
+            state: 'critical',
+            description: "error fetching #{opts[:riak_host]}:#{opts[:stats_port]} #{e.class}, #{e.message}",
+          )
+          raise
+        end
+
+        if res.code.to_i == 200
+          JSON.parse(res.body)
+        else
+          report(
+            host: opts[:riak_host],
+            service: 'riak',
+            state: 'critical',
+            description: "stats returned HTTP #{res.code}:\n\n#{res.body}",
+          )
+          raise "Can't fetch stats via HTTP: #{res.core}:\n\n#{res.body}"
+        end
+      end
+
+      # Get current stats via riak-admin
+      def stats_riak_admin
+        str = `riak-admin status`
+        raise 'riak-admin failed' unless $CHILD_STATUS == 0
+
+        Hash[str.split(/\n/).map { |i| i.split(/ : /) }]
+      end
+
+      # Get current stats as a hash
+      def stats
+        if @httpstatus
+          stats_http
+        elsif @riakadmin
+          stats_riak_admin
+        else
+          report(
+            host: opts[:riak_host],
+            service: 'riak',
+            state: 'critical',
+            description: 'No mechanism for fetching Riak stats: neither HTTP nor riak-admin available.',
+          )
+          raise 'No mechanism for fetching Riak stats: neither HTTP nor riak-admin available.'
+        end
+      end
+
+      def core_services
+        %w[vnode_gets
+           vnode_puts
+           node_gets
+           node_puts
+           node_gets_set
+           node_puts_set
+           read_repairs]
+      end
+
+      def fsm_types
+        [{ 'get' => 'time' }, { 'put' => 'time' },
+         { 'get' => 'set_objsize' },]
+      end
+
+      def fsm_percentiles
+        [50, 95, 99]
+      end
+
+      # Reports current stats to Riemann
+      def check_stats
+        begin
+          stats = self.stats
+        rescue StandardError => e
+          event = {
+            state: 'critical',
+            description: e.message,
+            host: opts[:riak_host],
+          }
+          # Report errors
+          report(event.merge(service: 'riak'))
+          core_services.each do |s|
+            report(event.merge(service: "riak #{s}"))
+          end
+          fsm_types.each do |typespec|
+            typespec.each do |type, prop|
+              fsm_percentiles.each do |percentile|
+                report(event.merge(service: "riak #{type} #{prop} #{percentile}"))
+              end
+            end
+          end
+          return
+        end
+
+        # Riak itself
+        report(
+          host: opts[:riak_host],
+          service: 'riak',
+          state: 'ok',
+        )
+
+        # Gets/puts/rr
+        core_services.each do |s|
+          report(
+            host: opts[:riak_host],
+            service: "riak #{s}",
+            state: 'ok',
+            metric: stats[s].to_i / 60.0,
+            description: "#{stats[s].to_i / 60.0}/sec",
+          )
+        end
+
+        # FSMs
+        fsm_types.each do |typespec|
+          typespec.each do |type, prop|
+            fsm_percentiles.each do |percentile|
+              val = stats[fsm_stat(type, prop, percentile)].to_i || 0
+              val = 0 if val == 'undefined'
+              val /= 1000.0 if prop == 'time' # Convert us to ms
+              state = if prop == 'time'
+                        fsm_state(type, percentile, val)
+                      else
+                        'ok'
+                      end
+              report(
+                host: opts[:riak_host],
+                service: "riak #{type} #{prop} #{percentile}",
+                state: state,
+                metric: val,
+                description: "#{val} ms",
+              )
+            end
+          end
+        end
+      end
+
+      def tick
+        # This can utterly destroy a cluster, so we disable
+        # check_keys
+        check_stats
+        check_ring
+        check_disk
+        check_transfers
+      end
+    end
+  end
+end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: riemann-tools
 version: !ruby/object:Gem::Version
-  version: 1.1.1
+  version: 1.4.0
 platform: ruby
 authors:
 - Kyle Kingsbury
-autorequire:
+autorequire: 
 bindir: bin
 cert_chain: []
-date: 2022-07-02 00:00:00.000000000 Z
+date: 2022-08-30 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: json
@@ -72,6 +72,20 @@ dependencies:
     - - ">="
       - !ruby/object:Gem::Version
         version: '0'
+- !ruby/object:Gem::Dependency
+  name: racc
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
 - !ruby/object:Gem::Dependency
   name: rake
   requirement: !ruby/object:Gem::Requirement
@@ -158,6 +172,7 @@ executables:
 - riemann-haproxy
 - riemann-health
 - riemann-kvminstance
+- riemann-md
 - riemann-memcached
 - riemann-net
 - riemann-nginx-status
@@ -165,7 +180,9 @@ executables:
 - riemann-portcheck
 - riemann-proc
 - riemann-varnish
+- riemann-wrapper
 - riemann-zookeeper
+- riemann-zpool
 extensions: []
 extra_rdoc_files: []
 files:
@@ -177,7 +194,7 @@ files:
 - ".gitignore"
 - ".rspec"
 - ".rubocop.yml"
-- ".travis.yml"
+- ".ruby-version"
 - CHANGELOG.md
 - Gemfile
 - ISSUE_TEMPLATE.md
@@ -197,6 +214,7 @@ files:
 - bin/riemann-haproxy
 - bin/riemann-health
 - bin/riemann-kvminstance
+- bin/riemann-md
 - bin/riemann-memcached
 - bin/riemann-net
 - bin/riemann-nginx-status
@@ -204,13 +222,37 @@ files:
 - bin/riemann-portcheck
 - bin/riemann-proc
 - bin/riemann-varnish
+- bin/riemann-wrapper
 - bin/riemann-zookeeper
+- bin/riemann-zpool
 - lib/riemann/tools.rb
+- lib/riemann/tools/apache_status.rb
+- lib/riemann/tools/bench.rb
+- lib/riemann/tools/cloudant.rb
+- lib/riemann/tools/consul_health.rb
+- lib/riemann/tools/dir_files_count.rb
+- lib/riemann/tools/dir_space.rb
+- lib/riemann/tools/diskstats.rb
+- lib/riemann/tools/fd.rb
+- lib/riemann/tools/freeswitch.rb
+- lib/riemann/tools/haproxy.rb
 - lib/riemann/tools/health.rb
+- lib/riemann/tools/kvm.rb
+- lib/riemann/tools/md.rb
+- lib/riemann/tools/mdstat_parser.tab.rb
+- lib/riemann/tools/memcached.rb
 - lib/riemann/tools/net.rb
+- lib/riemann/tools/nginx_status.rb
 - lib/riemann/tools/ntp.rb
+- lib/riemann/tools/portcheck.rb
+- lib/riemann/tools/proc.rb
+- lib/riemann/tools/riemann_client_wrapper.rb
+- lib/riemann/tools/uptime_parser.tab.rb
 - lib/riemann/tools/utils.rb
+- lib/riemann/tools/varnish.rb
 - lib/riemann/tools/version.rb
+- lib/riemann/tools/zookeeper.rb
+- lib/riemann/tools/zpool.rb
 - riemann-tools.gemspec
 - tools/riemann-aws/LICENSE
 - tools/riemann-aws/README.md
@@ -222,43 +264,55 @@ files:
 - tools/riemann-aws/bin/riemann-elb-metrics
 - tools/riemann-aws/bin/riemann-s3-list
 - tools/riemann-aws/bin/riemann-s3-status
+- tools/riemann-aws/lib/riemann/tools/aws/billing.rb
+- tools/riemann-aws/lib/riemann/tools/aws/elb_metrics.rb
+- tools/riemann-aws/lib/riemann/tools/aws/rds_status.rb
+- tools/riemann-aws/lib/riemann/tools/aws/s3_list.rb
+- tools/riemann-aws/lib/riemann/tools/aws/s3_status.rb
+- tools/riemann-aws/lib/riemann/tools/aws/sqs_status.rb
+- tools/riemann-aws/lib/riemann/tools/aws/status.rb
 - tools/riemann-chronos/LICENSE
 - tools/riemann-chronos/README.md
 - tools/riemann-chronos/Rakefile
 - tools/riemann-chronos/bin/riemann-chronos
+- tools/riemann-chronos/lib/riemann/tools/chronos.rb
 - tools/riemann-docker/LICENSE
 - tools/riemann-docker/README.md
 - tools/riemann-docker/Rakefile
 - tools/riemann-docker/bin/riemann-docker
+- tools/riemann-docker/lib/riemann/tools/docker.rb
 - tools/riemann-elasticsearch/LICENSE
 - tools/riemann-elasticsearch/README.md
 - tools/riemann-elasticsearch/Rakefile
 - tools/riemann-elasticsearch/bin/riemann-elasticsearch
+- tools/riemann-elasticsearch/lib/riemann/tools/elasticsearch.rb
 - tools/riemann-marathon/LICENSE
 - tools/riemann-marathon/README.md
 - tools/riemann-marathon/Rakefile
 - tools/riemann-marathon/bin/riemann-marathon
+- tools/riemann-marathon/lib/riemann/tools/marathon.rb
 - tools/riemann-mesos/LICENSE
 - tools/riemann-mesos/README.md
 - tools/riemann-mesos/Rakefile
 - tools/riemann-mesos/bin/riemann-mesos
+- tools/riemann-mesos/lib/riemann/tools/mesos.rb
 - tools/riemann-munin/LICENSE
 - tools/riemann-munin/README.md
 - tools/riemann-munin/Rakefile
 - tools/riemann-munin/bin/riemann-munin
+- tools/riemann-munin/lib/riemann/tools/munin.rb
 - tools/riemann-rabbitmq/LICENSE
 - tools/riemann-rabbitmq/README.md
 - tools/riemann-rabbitmq/Rakefile
 - tools/riemann-rabbitmq/bin/riemann-rabbitmq
+- tools/riemann-rabbitmq/lib/riemann/tools/rabbitmq.rb
 - tools/riemann-riak/LICENSE
 - tools/riemann-riak/README.md
 - tools/riemann-riak/Rakefile
 - tools/riemann-riak/bin/riemann-riak
 - tools/riemann-riak/bin/riemann-riak-keys
 - tools/riemann-riak/bin/riemann-riak-ring
-- tools/riemann-riak/riak_status/key_count.erl
-- tools/riemann-riak/riak_status/riak_status.rb
-- tools/riemann-riak/riak_status/ringready.erl
+- tools/riemann-riak/lib/riemann/tools/riak.rb
 homepage: https://github.com/aphyr/riemann-tools
 licenses:
 - MIT
@@ -267,7 +321,7 @@ metadata:
   homepage_uri: https://github.com/aphyr/riemann-tools
   source_code_uri: https://github.com/aphyr/riemann-tools
   changelog_uri: https://github.com/aphyr/riemann-tools
-post_install_message:
+post_install_message: 
 rdoc_options: []
 require_paths:
 - lib
@@ -282,8 +336,8 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.3.16
-signing_key:
+rubygems_version: 3.2.5
+signing_key: 
 specification_version: 4
 summary: Utilities which submit events to Riemann.
 test_files: []

data/.travis.yml

@@ -1,31 +0,0 @@
-sudo: required
-
-services:
-  - docker
-
-addons:
-  apt:
-    packages:
-      - docker-ce
-
-env:
-  - EXTRA_TOOL=
-  - EXTRA_TOOL=riemann-aws
-  - EXTRA_TOOL=riemann-chronos
-  - EXTRA_TOOL=riemann-docker
-  - EXTRA_TOOL=riemann-elasticsearch
-  - EXTRA_TOOL=riemann-marathon
-  - EXTRA_TOOL=riemann-mesos
-  - EXTRA_TOOL=riemann-munin
-  - EXTRA_TOOL=riemann-rabbitmq
-  - EXTRA_TOOL=riemann-riak
-
-before_script:
-  - docker version
-
-script:
-  - .docker/publish.sh
-
-branches:
-  only:
-    - master

data/tools/riemann-riak/riak_status/key_count.erl

@@ -1,13 +0,0 @@
-#!/usr/bin/env escript
-Process.setproctitle($0)
-%%! -name riakstatuscheck -setcookie riak -hidden
-
-main([]) -> main(["riak@127.0.0.1"]);
-main([Node]) ->
-  io:format("~w\n", [
-    lists:foldl(
-      fun({_VNode, Count}, Sum) -> Sum + Count end,
-      0,
-      rpc:call(list_to_atom(Node), riak_kv_bitcask_backend, key_counts, [])
-    )
-  ]).

data/tools/riemann-riak/riak_status/riak_status.rb

@@ -1,152 +0,0 @@
-#!/usr/bin/env ruby
-# frozen_string_literal: true
-
-Process.setproctitle($PROGRAM_NAME)
-
-$LOAD_PATH.unshift File.expand_path("#{File.dirname(__FILE__)}/../vodpod-common/lib")
-require 'rubygems'
-require 'vodpod-common'
-require 'vodpod/alerts'
-require 'vodpod/starling'
-require 'net/http'
-require 'yajl/json_gem'
-
-class RiakStatus
-  PORT = 8098
-  PATH = '/stats'
-  INTERVAL = 10
-
-  FSM_LIMITS = {
-    get: {
-      50 => 1000,
-      95 => 2000,
-      99 => 10_000,
-    },
-    put: {
-      50 => 1000,
-      95 => 2000,
-      99 => 10_000,
-    },
-  }.freeze
-
-  def initialize(opts = {})
-    @host = opts[:host] || `hostname`.chomp
-    @port = opts[:port] || PORT
-    @path = opts[:path] || PATH
-  end
-
-  def alert(subservice, state, metric, description)
-    Vodpod.alert(
-      service: "riak #{subservice}",
-      state: state,
-      metric: metric,
-      description: description,
-    )
-  end
-
-  def check_ring
-    str = %x(#{__dir__}/ringready.erl riak@#{`hostname`}).chomp
-    if str =~ /^TRUE/
-      alert 'ring', :ok, nil, str
-    else
-      alert 'ring', :warning, nil, str
-    end
-  end
-
-  def check_keys
-    keys = %x(#{__dir__}/key_count.erl riak@#{`hostname`}).chomp
-    if keys =~ /^\d+$/
-      alert 'keys', :ok, keys.to_i, keys
-    else
-      alert 'keys', :error, nil, keys
-    end
-  end
-
-  def check_disk
-    gb = `du -s /var/lib/riak/bitcask/`.split(/\s+/).first.to_i / (1024.0**2)
-    alert 'disk', :ok, gb, "#{gb} GB in bitcask"
-  end
-
-  # Returns the riak stat for the given fsm type and percentile.
-  def fsm_stat(type, percentile)
-    "node_#{type}_fsm_time_#{percentile == 50 ? 'median' : percentile}"
-  end
-
-  # Returns the alerts state for the given fsm.
-  def fsm_state(type, percentile, val)
-    limit = FSM_LIMITS[type][percentile]
-    case val
-    when 0..limit
-      :ok
-    when limit..limit * 2
-      :warning
-    else
-      :critical
-    end
-  end
-
-  def check_stats
-    begin
-      res = Net::HTTP.start(@host, @port) do |http|
-        http.get('/stats')
-      end
-    rescue StandardError => e
-      Vodpod.alert(
-        service: 'riak',
-        state: :critical,
-        description: "error fetching /stats: #{e.class}, #{e.message}",
-      )
-      return
-    end
-
-    if res.code.to_i == 200
-      stats = JSON.parse(res.body)
-    else
-      Vodpod.alert(
-        service: 'riak',
-        state: :critical,
-        description: "stats returned HTTP #{res.code}:\n\n#{res.body}",
-      )
-      return
-    end
-
-    Vodpod.alert(
-      service: 'riak',
-      state: :ok,
-    )
-
-    # Gets/puts/rr
-    %w[
-      vnode_gets
-      vnode_puts
-      node_gets
-      node_puts
-      read_repairs
-    ].each do |s|
-      alert s, :ok, stats[s] / 60.0, "#{stats[s] / 60.0}/sec"
-    end
-
-    # FSMs
-    %i[get put].each do |type|
-      [50, 95, 99].each do |percentile|
-        val = stats[fsm_stat(type, percentile)] || 0
-        val = 0 if val == 'undefined'
-        val /= 1000.0 # Convert us to ms
-        state = fsm_state(type, percentile, val)
-        alert "#{type} #{percentile}", state, val, "#{val} ms"
-      end
-    end
-  end
-
-  def run
-    loop do
-      #      check_keys
-      check_stats
-      check_ring
-      check_disk
-      sleep INTERVAL
-    end
-  end
-end
-
-RiakStatus.new.run if $PROGRAM_NAME == __FILE__