flapjack 0.6.29 → 0.6.30

data/README.md

@@ -62,7 +62,7 @@ There's some rspec unit tests as well, run these like so:
 
     $ rspec spec
 
-NB, if the cucumber tests fail with a spurious lexing error on line 2 of events.feature, then try this:
+NB, if the cucumber tests fail with a [spurious lexing error](https://github.com/cucumber/gherkin/issues/182) on line 2 of events.feature, then try this:
 
     $ cucumber -f fuubar features
 

data/bin/flapjack-populator

@@ -1,11 +1,18 @@
 #!/usr/bin/env ruby
 
-require 'yajl'
+require 'yajl/json_gem'
 require 'redis'
-require 'yaml'
 require 'optparse'
 require 'ostruct'
-require 'flapjack/data/entity_check'
+
+# add lib to the default include path
+unless $:.include?(File.dirname(__FILE__) + '/../lib/')
+  $: << File.dirname(__FILE__) + '/../lib'
+end
+
+require 'flapjack/configuration'
+require 'flapjack/data/contact'
+require 'flapjack/data/entity'
 
 options = OpenStruct.new
 options.config    = File.join('etc', 'flapjack_config.yaml')
@@ -35,109 +42,58 @@ end.parse!(ARGV)
 
 FLAPJACK_ENV = ENV['FLAPJACK_ENV'] || 'development'
 
-# load the config hash for the current environment
-
-if File.file?(options.config)
-  config = YAML::load(File.open(options.config))
-else
-  puts "Could not find config file at '#{options.config}', you may want to specify one with the --config option"
-  exit(false)
-end
-
-config_env = config[FLAPJACK_ENV]
-
-if config_env.nil? || config_env.empty?
-  puts "No config data for environment '#{FLAPJACK_ENV}' found in '#{options.config}'"
-  exit(false)
-end
+config_env = Flapjack::Configuration.new.load(options.config)
 
 if options.from
   filename = options.from
   file     = File.new(filename)
 end
 
-parser   = Yajl::Parser.new
-
-@redis_host = config_env['redis']['host'] || 'localhost'
-@redis_port = config_env['redis']['port'] || '6379'
-@redis_path = config_env['redis']['path'] || nil
-@redis_db   = config_env['redis']['db']   || 0
-
-# add lib to the default include path
-unless $:.include?(File.dirname(__FILE__) + '/../lib/')
-  $: << File.dirname(__FILE__) + '/../lib'
-end
-
-def get_redis_connection
-  if @redis_path
-    redis = Redis.new(:db => @redis_db, :path => @redis_path)
-  else
-    redis = Redis.new(:db => @redis_db, :host => @redis_host, :port => @redis_port)
-  end
-  redis
+def get_redis_connection(cfg)
+  opts = cfg['path'] ? {:path => cfg['path']} :
+                       {:host => cfg['host'], :port => cfg['port']}
+  Redis.new(opts.merge(:db => cfg['db']))
 end
 
 case command
 when "import-contacts"
-  @persistence = get_redis_connection
-  contacts = parser.parse(file)
-  contacts.each {|contact|
-    if not contact['id']
-      puts "Contact not imported as it has no id: " + contact.inspect
-      next
-    end
-    @persistence.multi
-    @persistence.del("contact:#{contact['id']}")
-    @persistence.del("contact_media:#{contact['id']}")
-    @persistence.del("contact_pagerduty:#{contact['id']}")
-    @persistence.hset("contact:#{contact['id']}", 'first_name', contact['first_name'])
-    @persistence.hset("contact:#{contact['id']}", 'last_name',  contact['last_name'])
-    @persistence.hset("contact:#{contact['id']}", 'email',      contact['email'])
-    contact['media'].each_pair {|medium, address|
-      case medium
-      when 'pagerduty'
-        @persistence.hset("contact_media:#{contact['id']}", medium, address['service_key'])
-        @persistence.hset("contact_pagerduty:#{contact['id']}", 'subdomain', address['subdomain'])
-        @persistence.hset("contact_pagerduty:#{contact['id']}", 'username',  address['username'])
-        @persistence.hset("contact_pagerduty:#{contact['id']}", 'password',  address['password'])
-      else
-        @persistence.hset("contact_media:#{contact['id']}", medium, address)
+  contacts = JSON.parse(file)
+
+  if contacts && contacts.is_a?(Enumerable) && contacts.any? {|e| !e['id'].nil?}
+    @persistence = get_redis_connection(config_env['redis'])
+    @persistence.multi do
+      contacts.each do |contact|
+        unless contact['id']
+          puts "Contact not imported as it has no id: " + contact.inspect
+          next
+        end
+        Flapjack::Data::Contact.add(contact, :redis => @persistence)
       end
-    }
-    @persistence.exec
-  }
+    end
+    @persistence.quit
+  end
 
 when "import-entities"
-  @persistence = get_redis_connection
-  entities = parser.parse(file)
-  entities.each {|entity|
-    if not entity['id']
-      puts "Entity not imported as it has no id: " + entity.inspect
-      next
-    end
+  entities = JSON.parse(file)
+
+  if entities && entities.is_a?(Enumerable) && entities.any? {|e| !e['id'].nil?}
+    @persistence = get_redis_connection(config_env['redis'])
     @persistence.multi
-    existing_name = @persistence.hget("entity:#{entity['id']}", 'name')
-    @persistence.del("entity_id:#{existing_name}") unless existing_name == entity['name']
-    @persistence.set("entity_id:#{entity['name']}", entity['id'])
-    @persistence.hset("entity:#{entity['id']}", 'name', entity['name'])
-
-    @persistence.del("contacts_for:#{entity['id']}")
-    entity['contacts'].each {|contact|
-      @persistence.sadd("contacts_for:#{entity['id']}", contact)
-    }
+    entities.each do |entity|
+      unless entity['id']
+        puts "Entity not imported as it has no id: " + entity.inspect
+        next
+      end
+      Flapjack::Data::Entity.add(entity, :redis => @persistence)
+    end
     @persistence.exec
-
-  }
+    @persistence.quit
+  end
 
 when "purge-events"
-  @persistence = get_redis_connection
-  events_size = @persistence.llen('events')
-  puts "purging #{events_size} events..."
-  timestamp = Time.now.to_i
-  puts "renaming events to events.#{timestamp}"
-  @persistence.rename('events', "events.#{timestamp}")
-  puts "setting expiry of events.#{timestamp} to 8 hours"
-  @persistence.expire("events.#{timestamp}", (60 * 60 * 8))
+  @persistence = get_redis_connection(config_env['redis'])
+  Flapjack::Data.Event.purge_all(:redis => @persistence)
+  @persistence.quit
 
 else
   puts "you need to give me something to do, eg a command like 'import-entities' or 'import-clients' etc"

data/lib/flapjack/api.rb

@@ -25,7 +25,9 @@ module Flapjack
     # doesn't work with Rack::Test for some reason
     unless 'test'.eql?(FLAPJACK_ENV)
       rescue_exception = Proc.new { |env, exception|
-        [503, {}, exception.message]
+        logger.error exception.message
+        logger.error exception.backtrace.join("\n")
+        [503, {}, {:status => 503, :reason => exception.message}.to_json]
       }
 
       use Rack::FiberPool, :size => 25, :rescue_exception => rescue_exception
@@ -38,17 +40,6 @@ module Flapjack
       Flapjack::API.bootstrap
     end
 
-    helpers do
-      def json_status(code, reason)
-        status code
-        {:status => code, :reason => reason}.to_json
-      end
-
-      def logger
-        Flapjack::API.logger
-      end
-    end
-
     get '/entities' do
       content_type :json
       ret = Flapjack::Data::Entity.all(:redis => @@redis).sort_by(&:name).collect {|e|
@@ -211,7 +202,7 @@ module Flapjack
         params[:check], :redis => @@redis)
       entity_check.create_scheduled_maintenance(:start_time => params[:start_time],
         :duration => params[:duration], :summary => params[:summary])
-      status 201
+      status 204
     end
 
     # create an acknowledgement for a service on an entity
@@ -232,15 +223,11 @@ module Flapjack
         params[:check], :redis => @@redis)
       entity_check.create_acknowledgement('summary' => params[:summary],
         'duration' => duration)
-      status 201
+      status 204
     end
 
     not_found do
-      json_status 404, "Not found"
-    end
-
-    error do
-      json_status 500, env['sinatra.error'].message
+      [404, {}, {:status => 404, :reason => "Not found"}.to_json]
     end
 
     private
@@ -265,7 +252,7 @@ module Flapjack
         return unless value
         Time.iso8601(value).getutc.to_i
       rescue ArgumentError => e
-        # FIXME log error
+        logger.error "Couldn't parse time from '#{value}'"
         nil
       end
 

data/lib/flapjack/api/entity_check_presenter.rb

@@ -16,56 +16,34 @@ module Flapjack
         @entity_check = entity_check
       end
 
-      # if options[:chop] is true, overlapping outages at start and end
-      # times will be sliced to fit.
       def outages(start_time, end_time, options = {})
-        # states is an array of hashes, with [state, timestamp, summary] keys
-        states = @entity_check.historical_states(start_time, end_time)
-        return states if states.empty?
-
-        # if it started failed, prepend the earlier event
-        initial = @entity_check.historical_state_before(states.first[:timestamp])
-        if (initial && (initial[:state] == Flapjack::Data::EntityCheck::STATE_CRITICAL))
-          states.unshift(initial)
-        end
-
-        # if it ended failed, append the event when it recovered
-        if states.last[:state] == Flapjack::Data::EntityCheck::STATE_CRITICAL
-          # TODO ensure this event is not CRITICAL, get first non-CRITICAL if so
-          last = @entity_check.historical_state_after(states.last[:timestamp])
-          states.push(last) if last
-        end
-
-        last_state = nil
-
-        # returns an array of hashes, with [:start_time, :end_time, :summary]
-        time_periods = states.inject([]) do |ret, obj|
-          if (obj[:state] == Flapjack::Data::EntityCheck::STATE_CRITICAL) &&
-            (last_state.nil? || (last_state != Flapjack::Data::EntityCheck::STATE_CRITICAL))
-
-            # flipped to failed, mark next outage
-            last_state = obj[:state]
-            ret << {:start_time => obj[:timestamp], :end_time => nil, :summary => obj[:summary]}
-          elsif (obj[:state] != Flapjack::Data::EntityCheck::STATE_CRITICAL) &&
-            (last_state == Flapjack::Data::EntityCheck::STATE_CRITICAL)
-
-            # flipped to not failed, mark end time for the current outage
-            last_state = obj[:state]
-            ret.last[:end_time] = obj[:timestamp]
+        # hist_states is an array of hashes, with [state, timestamp, summary] keys
+        hist_states = @entity_check.historical_states(start_time, end_time)
+        return hist_states if hist_states.empty?
+
+        initial = @entity_check.historical_state_before(hist_states.first[:timestamp])
+        hist_states.unshift(initial) if initial
+
+        num_states = hist_states.size
+
+        hist_states.each_with_index do |obj, index|
+          if (index == 0)
+            # initial
+            obj[:start_time] = [obj.delete(:timestamp), start_time].max
+            obj[:end_time]   = hist_states[index + 1][:timestamp]
+          elsif index == (num_states - 1)
+            # last
+            obj[:start_time] = obj.delete(:timestamp)
+            obj[:end_time]   = end_time
+          else
+            # except for first and last
+            obj[:start_time] = obj.delete(:timestamp)
+            obj[:end_time]   = hist_states[index + 1][:timestamp]
           end
-          ret
+          obj[:duration] = obj[:end_time] - obj[:start_time]
         end
 
-        if options[:chop]
-          if start_time && (time_periods.first[:start_time] < start_time)
-            time_periods.first[:start_time] = start_time
-          end
-          if time_periods.last[:end_time].nil? || (end_time && (time_periods.last[:end_time] > end_time))
-            time_periods.last[:end_time] = (end_time || Time.now.to_i)
-          end
-        end
-
-        time_periods
+        hist_states.reject {|obj| obj[:state] == 'ok'}
       end
 
       def unscheduled_maintenance(start_time, end_time)
@@ -106,10 +84,15 @@ module Flapjack
       def downtime(start_time, end_time)
         sched_maintenances = scheduled_maintenance(start_time, end_time)
 
-        outs = outages(start_time, end_time, :chop => true)
+        outs = outages(start_time, end_time)
 
-        total_secs = 0
-        percentage = 0
+        total_secs  = {}
+        percentages = {}
+
+        outs.collect {|obj| obj[:state]}.uniq.each do |st|
+          total_secs[st]  = 0
+          percentages[st] = (start_time.nil? || end_time.nil?) ? nil : 0
+        end
 
         unless outs.empty?
 
@@ -126,10 +109,12 @@ module Flapjack
               next unless o[:start_time] < sm[:start_time] &&
                 o[:end_time] > sm[:end_time]
               o[:delete] = true
-              split_outs += [{:start_time => o[:start_time],
+              split_outs += [{:state => o[:state],
+                              :start_time => o[:start_time],
                               :end_time => sm[:start_time],
                               :summary => "#{o[:summary]} [split start]"},
-                             {:start_time => sm[:end_time],
+                             {:state => o[:state],
+                              :start_time => sm[:end_time],
                               :end_time => o[:end_time],
                               :summary => "#{o[:summary]} [split finish]"}]
             }
@@ -165,16 +150,21 @@ module Flapjack
             outs.reject! {|o| o[:delete]}
           end
 
-          # sum outage times
-          total_secs = outs.inject(0) {|sum, o|
-            sum += (o[:end_time] - o[:start_time])
+          total_secs = outs.inject(total_secs) {|ret, o|
+            ret[o[:state]] += (o[:end_time] - o[:start_time])
+            ret
           }
 
-          percentage = (start_time.nil? || end_time.nil?) ? nil :
-                         (total_secs * 100) / (end_time - start_time)
+          unless (start_time.nil? || end_time.nil?)
+            total_secs.each_pair do |st, ts|
+              percentages[st] = (total_secs[st] * 100.0) / (end_time.to_f - start_time.to_f)
+            end
+          end
         end
 
-        {:total_seconds => total_secs, :percentage => percentage, :downtime => outs}
+        total_secs['ok'] = (end_time - start_time) - total_secs.values.reduce(:+)
+        percentages['ok'] = 100 - percentages.values.reduce(:+)
+        {:total_seconds => total_secs, :percentages => percentages, :downtime => outs}
       end
 
     end

data/lib/flapjack/configuration.rb

@@ -1,10 +1,62 @@
 #!/usr/bin/env ruby
 
+require 'yaml'
+
 module Flapjack
 
   class Configuration
 
-    # TODO consolidate configuration code
+    def initialize(opts = {})
+      @logger = opts[:logger]
+      unless @logger
+        @logger       = Logger.new(STDOUT)
+        @logger.level = Logger::ERROR
+      end
+    end
+
+    def logger
+      @logger
+    end
+
+    def load(filename)
+      unless File.file?(filename)
+        logger.error "Could not find file '#{filename}'"
+        return
+      end
+
+      unless defined?(FLAPJACK_ENV)
+        logger.error "Environment variable 'FLAPJACK_ENV' is not set"
+        return
+      end
+
+      config = YAML::load_file(filename)
+
+      if config.nil?
+        logger.error "Could not load config file '#{filename}'"
+        return
+      end
+
+      config_env = config[FLAPJACK_ENV]
+
+      if config_env.nil?
+        logger.error "No config data for environment '#{FLAPJACK_ENV}' found in '#{filename}'"
+        return
+      end
+
+      redis_defaults = {'host' => 'localhost',
+                        'port' => 6379,
+                        'path' => nil,
+                        'db'   => 0}
+
+      config_env['redis'] = {} unless config_env.has_key?('redis')
+      redis_defaults.each_pair do |k,v|
+        next if config_env['redis'].has_key?(k) && (config_env['redis'][k] &&
+          !(config_env['redis'][k].is_a?(String) && config_env['redis'][k].empty?))
+        config_env['redis'][k] = v
+      end
+
+      config_env
+    end
 
   end
 

data/lib/flapjack/data/contact.rb

@@ -9,9 +9,8 @@ module Flapjack
       # takes a check, looks up contacts that are interested in this check (or in the check's entity)
       # and returns an array of contact ids
       def self.find_all_for_entity_check(entity_check, options = {})
-        logger = options[:logger]
-        logger = nil
         raise "Redis connection not set" unless redis = options[:redis]
+        logger = options[:logger]
 
         entity = entity_check.entity
         check  = entity_check.check
@@ -26,16 +25,38 @@ module Flapjack
         union
       end
 
-      def self.pagerduty_credentials_for_contact(contact, options = {})
-        logger = options[:logger]
+      # NB: should probably be called in the context of a Redis multi block; not doing so
+      # here as calling classes may well be adding/updating multiple records in the one
+      # operation
+      def self.add(contact, options = {})
         raise "Redis connection not set" unless redis = options[:redis]
 
-        service_key = redis.hget("contact_media:#{contact}", 'pagerduty')
-        return nil unless service_key
+        redis.del("contact:#{contact['id']}")
+        redis.del("contact_media:#{contact['id']}")
+        redis.del("contact_pagerduty:#{contact['id']}")
+        redis.hset("contact:#{contact['id']}", 'first_name', contact['first_name'])
+        redis.hset("contact:#{contact['id']}", 'last_name',  contact['last_name'])
+        redis.hset("contact:#{contact['id']}", 'email',      contact['email'])
+        contact['media'].each_pair {|medium, address|
+          case medium
+          when 'pagerduty'
+            redis.hset("contact_media:#{contact['id']}", medium, address['service_key'])
+            redis.hset("contact_pagerduty:#{contact['id']}", 'subdomain', address['subdomain'])
+            redis.hset("contact_pagerduty:#{contact['id']}", 'username',  address['username'])
+            redis.hset("contact_pagerduty:#{contact['id']}", 'password',  address['password'])
+          else
+            redis.hset("contact_media:#{contact['id']}", medium, address)
+          end
+        }
+      end
+
+      def self.pagerduty_credentials_for_contact(contact, options = {})
+        raise "Redis connection not set" unless redis = options[:redis]
 
-        deets = redis.hgetall("contact_pagerduty:#{contact}")
-        return deets.merge('service_key' => service_key)
+        return unless service_key = redis.hget("contact_media:#{contact}", 'pagerduty')
 
+        redis.hgetall("contact_pagerduty:#{contact}").
+          merge('service_key' => service_key)
       end
 
     end

data/lib/flapjack/data/entity.rb

@@ -16,12 +16,14 @@ module Flapjack
         }
       end
 
+      # NB: should probably be called in the context of a Redis multi block; not doing so
+      # here as calling classes may well be adding/updating multiple records in the one
+      # operation
       def self.add(entity, options = {})
         raise "Redis connection not set" unless redis = options[:redis]
         raise "Entity name not provided" unless entity['name'] && !entity['name'].empty?
 
         if entity['id']
-          redis.multi
           existing_name = redis.hget("entity:#{entity['id']}", 'name')
           redis.del("entity_id:#{existing_name}") unless existing_name == entity['name']
           redis.set("entity_id:#{entity['name']}", entity['id'])
@@ -33,7 +35,6 @@ module Flapjack
               redis.sadd("contacts_for:#{entity['id']}", contact)
             }
           end
-          redis.exec
         else
           # empty string is the redis equivalent of a Ruby nil, i.e. key with
           # no value

data/lib/flapjack/data/event.rb

@@ -45,6 +45,17 @@ module Flapjack
         opts[:persistence].llen('events')
       end
 
+      # FIXME make this use a logger taken from the opts
+      def self.purge_all(opts = {})
+        events_size = opts[:redis].llen('events')
+        puts "purging #{events_size} events..."
+        timestamp = Time.now.to_i
+        puts "renaming events to events.#{timestamp}"
+        opts[:redis].rename('events', "events.#{timestamp}")
+        puts "setting expiry of events.#{timestamp} to 8 hours"
+        opts[:redis].expire("events.#{timestamp}", (60 * 60 * 8))
+      end
+
       def initialize(attrs={})
         @attrs = attrs
         @attrs['time'] = Time.now.to_i unless @attrs.has_key?('time')

data/lib/flapjack/jabber.rb

@@ -283,9 +283,6 @@ module Flapjack
                 close # Blather::Client.close
               end
             end
-            # FIXME: should we also set something so should_quit? returns true
-            # to prevent retrieving more notifications from the queue while closing?
-            # or does close only return once the connection is really and truely closed?
           else
             entity, check = event['event_id'].split(':')
             state         = event['state']

data/lib/flapjack/version.rb

@@ -1,5 +1,5 @@
 #!/usr/bin/env ruby
 
 module Flapjack
-  VERSION = "0.6.29"
+  VERSION = "0.6.30"
 end

data/lib/flapjack/web.rb

@@ -18,8 +18,8 @@ module Flapjack
     # doesn't work with Rack::Test for some reason
     unless 'test'.eql?(FLAPJACK_ENV)
       rescue_exception = Proc.new { |env, exception|
-        p exception.message
-        puts exception.backtrace.join("\n")
+        logger.error exception.message
+        logger.error exception.backtrace.join("\n")
         [503, {}, exception.message]
       }
 

data/spec/lib/flapjack/api/entity_check_presenter_spec.rb

@@ -104,10 +104,13 @@ describe 'Flapjack::API::EntityCheck::Presenter' do
     ecp = Flapjack::API::EntityCheckPresenter.new(entity_check)
     downtimes = ecp.downtime(time - (12 * 60 * 60), time)
 
+
     # 22 minutes, 3 + 8 + 11
     downtimes.should be_a(Hash)
-    downtimes[:total_seconds].should == (22 * 60)
-    downtimes[:percentage].should == (((22 * 60) * 100) / (12 * 60 * 60))
+    downtimes[:total_seconds].should == {'critical' => (22 * 60),
+      'ok' => ((12 * 60 * 60) - (22 * 60))}
+    downtimes[:percentages].should == {'critical' => (((22 * 60) * 100.0) / (12 * 60 * 60)),
+      'ok' => ((((12 * 60 * 60) - (22 * 60)) * 100.0) / (12 * 60 *60))}
     downtimes[:downtime].should be_an(Array)
     # the last outage gets split by the intervening maintenance period,
     # but the fully covered one gets removed.

data/spec/lib/flapjack/executive_spec.rb

@@ -34,10 +34,9 @@ describe Flapjack::Executive, :redis => true do
     shutdown_evt.should_receive(:time).and_return(Time.now)
     Flapjack::Data::Event.should_receive(:next).and_return(shutdown_evt)
 
-    EventMachine::Synchrony::ConnectionPool.should_receive(:new).and_return(@redis)
-
     executive = Flapjack::Executive.new
     executive.bootstrap(:config => {})
+    executive.should_receive(:build_redis_connection_pool).and_return(@redis)
 
     # hacky, but the behaviour it's mimicking (shutdown from another thread) isn't
     # conducive to nice tests

data/spec/lib/flapjack/jabber_spec.rb

@@ -139,7 +139,7 @@ describe Flapjack::Jabber do
 
     fj.should_receive(:connect)
     fj.should_receive(:connected?).exactly(3).times.and_return(true)
-    fj.should_receive(:should_quit?).exactly(3).times.and_return(false, false, true)
+    fj.should_receive(:should_quit?).exactly(4).times.and_return(false, false, true)
     redis.should_receive(:blpop).twice.and_return(
       ["jabber_notifications", %q{{"notification_type":"problem","event_id":"main-example.com:ping","state":"critical","summary":"!!!"}}],
       ["jabber_notifications", %q{{"notification_type":"shutdown"}}]

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: flapjack
 version: !ruby/object:Gem::Version
-  version: 0.6.29
+  version: 0.6.30
   prerelease: 
 platform: ruby
 authors:
@@ -11,7 +11,7 @@ authors:
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2012-09-21 00:00:00.000000000 Z
+date: 2012-09-24 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: daemons