riemann-tools 1.0.0 → 1.2.0

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: 1.0.0
+  version: 1.2.0
 platform: ruby
 authors:
 - Kyle Kingsbury
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2022-06-22 00:00:00.000000000 Z
+date: 2022-08-17 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
@@ -100,6 +114,48 @@ dependencies:
     - - ">="
       - !ruby/object:Gem::Version
         version: '0'
+- !ruby/object:Gem::Dependency
+  name: rubocop
+  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: rubocop-rake
+  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: rubocop-rspec
+  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'
 description: Collection of utilities which submit events to Riemann,
 email:
 - aphyr@aphyr.com
@@ -123,22 +179,27 @@ executables:
 - riemann-portcheck
 - riemann-proc
 - riemann-varnish
+- riemann-wrapper
 - riemann-zookeeper
 extensions: []
 extra_rdoc_files: []
 files:
 - ".docker/Dockerfile"
 - ".docker/publish.sh"
+- ".github/dependabot.yml"
 - ".github/workflows/ci.yml"
+- ".github/workflows/codeql-analysis.yml"
 - ".gitignore"
 - ".rspec"
-- ".travis.yml"
+- ".rubocop.yml"
+- ".ruby-version"
 - CHANGELOG.md
 - Gemfile
 - ISSUE_TEMPLATE.md
 - LICENSE
 - README.markdown
 - Rakefile
+- SECURITY.md
 - bin/riemann-apache-status
 - bin/riemann-bench
 - bin/riemann-cloudant
@@ -158,14 +219,37 @@ files:
 - bin/riemann-portcheck
 - bin/riemann-proc
 - bin/riemann-varnish
+- bin/riemann-wrapper
 - bin/riemann-zookeeper
 - 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/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
 - riemann-tools.gemspec
 - tools/riemann-aws/LICENSE
 - tools/riemann-aws/README.md
-- tools/riemann-aws/Rakefile.rb
+- tools/riemann-aws/Rakefile
 - tools/riemann-aws/bin/riemann-aws-billing
 - tools/riemann-aws/bin/riemann-aws-rds-status
 - tools/riemann-aws/bin/riemann-aws-sqs-status
@@ -173,43 +257,55 @@ 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.rb
+- 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.rb
+- 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.rb
+- 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.rb
+- 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.rb
+- 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.rb
+- 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.rb
+- 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.rb
+- 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
@@ -226,7 +322,7 @@ required_ruby_version: !ruby/object:Gem::Requirement
   requirements:
   - - ">="
     - !ruby/object:Gem::Version
-      version: 2.5.0
+      version: 2.6.0
 required_rubygems_version: !ruby/object:Gem::Requirement
   requirements:
   - - ">="

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, [])
-    )
-  ]).