flapjack 0.7.14 → 0.7.15

data/lib/flapjack/gateways/api/entity_check_presenter.rb

@@ -18,6 +18,19 @@ module Flapjack
           @entity_check = entity_check
         end
 
+        def status
+          {'name'                              => @entity_check.check,
+           'state'                             => @entity_check.state,
+           'summary'                           => @entity_check.summary,
+           'details'                           => @entity_check.details,
+           'in_unscheduled_maintenance'        => @entity_check.in_unscheduled_maintenance?,
+           'in_scheduled_maintenance'          => @entity_check.in_scheduled_maintenance?,
+           'last_update'                       => @entity_check.last_update,
+           'last_problem_notification'         => entity_check.last_notification_for_state(:problem)[:timestamp],
+           'last_recovery_notification'        => entity_check.last_notification_for_state(:recovery)[:timestamp],
+           'last_acknowledgement_notification' => entity_check.last_notification_for_state(:acknowledgement)[:timestamp]}
+        end
+
         def outages(start_time, end_time, options = {})
           # hist_states is an array of hashes, with [state, timestamp, summary] keys
           hist_states = @entity_check.historical_states(start_time, end_time)
@@ -26,32 +39,50 @@ module Flapjack
           initial = @entity_check.historical_state_before(hist_states.first[:timestamp])
           hist_states.unshift(initial) if initial
 
+          # TODO the following works, but isn't the neatest
           num_states = hist_states.size
 
-          hist_states.each_with_index do |obj, index|
-            ts = obj.delete(:timestamp)
-            if index == (num_states - 1)
-              # last (even if the only one)
-              obj[:start_time] = start_time ? [ts, start_time].max : ts
-              obj[:end_time]   = end_time
-            elsif (index == 0)
-              # initial
-              obj[:start_time] = start_time ? [ts, start_time].max : ts
-              obj[:end_time]   = hist_states[index + 1][:timestamp]
-            else
-              # except for first and last
-              obj[:start_time] = ts
-              obj[:end_time]   = hist_states[index + 1][:timestamp]
+          index = 0
+          result = []
+          obj = nil
+
+          while index < num_states do
+            last_obj = obj
+            obj = hist_states[index]
+            index += 1
+
+            if last_obj && (last_obj[:state] == obj[:state])
+              # TODO maybe build up arrays of these instead, and leave calling
+              # classes to join them together if needed?
+              result.last[:summary] << " / #{obj[:summary]}"
+              result.last[:details] << " / #{obj[:details]}"
+              next
             end
-            obj[:duration] = obj[:end_time] ? (obj[:end_time] - obj[:start_time]) : nil
-          end
 
-          # p hist_states
+            next if obj[:state] == 'ok'
+
+            ts = obj[:timestamp]
+
+            obj_st  = (last_obj || !start_time) ? ts : [ts, start_time].max
+
+            next_ts_obj = hist_states[index..-1].detect {|hs| hs[:state] != obj[:state] }
+            obj_et  = next_ts_obj ? next_ts_obj[:timestamp] : end_time
+
+            obj_dur = obj_et ? obj_et - obj_st : nil
+
+            result << {:state      => obj[:state],
+                       :start_time => obj_st,
+                       :end_time   => obj_et,
+                       :duration   => obj_dur,
+                       :summary    => obj[:summary] || '',
+                       :details    => obj[:details] || ''
+                      }
+          end
 
-          hist_states.reject {|obj| obj[:state] == 'ok'}
+          result
         end
 
-        def unscheduled_maintenance(start_time, end_time)
+        def unscheduled_maintenances(start_time, end_time)
           # unsched_maintenance is an array of hashes, with [duration, timestamp, summary] keys
           unsched_maintenance = @entity_check.maintenances(start_time, end_time,
             :scheduled => false)
@@ -66,7 +97,7 @@ module Flapjack
           start_in_unsched + unsched_maintenance
         end
 
-        def scheduled_maintenance(start_time, end_time)
+        def scheduled_maintenances(start_time, end_time)
           # sched_maintenance is an array of hashes, with [duration, timestamp, summary] keys
           sched_maintenance = @entity_check.maintenances(start_time, end_time,
             :scheduled => true)
@@ -87,7 +118,7 @@ module Flapjack
         #
         # TODO test performance with larger data sets
         def downtime(start_time, end_time)
-          sched_maintenances = scheduled_maintenance(start_time, end_time)
+          sched_maintenances = scheduled_maintenances(start_time, end_time)
 
           outs = outages(start_time, end_time)
 

data/lib/flapjack/gateways/api/entity_presenter.rb

@@ -22,29 +22,34 @@ module Flapjack
           @redis = options[:redis]
         end
 
+        def status
+          checks.collect {|c| {:entity => @entity, :check => c,
+                               :status => check_presenter(c).status } }
+        end
+
         def outages(start_time, end_time)
           checks.collect {|c|
-            {:check => c, :outages => check_presenter(c).outages(start_time, end_time)}
+            {:entity => @entity, :check => c, :outages => check_presenter(c).outages(start_time, end_time)}
           }
         end
 
-        def unscheduled_maintenance(start_time, end_time)
+        def unscheduled_maintenances(start_time, end_time)
           checks.collect {|c|
-            {:check => c, :unscheduled_maintenance =>
-              check_presenter(c).unscheduled_maintenance(start_time, end_time)}
+            {:entity => @entity, :check => c, :unscheduled_maintenances =>
+              check_presenter(c).unscheduled_maintenances(start_time, end_time)}
           }
         end
 
-        def scheduled_maintenance(start_time, end_time)
+        def scheduled_maintenances(start_time, end_time)
           checks.collect {|c|
-            {:check => c, :scheduled_maintenance =>
-              check_presenter(c).scheduled_maintenance(start_time, end_time)}
+            {:entity => @entity, :check => c, :scheduled_maintenances =>
+              check_presenter(c).scheduled_maintenances(start_time, end_time)}
           }
         end
 
         def downtime(start_time, end_time)
           checks.collect {|c|
-            {:check => c, :downtime =>
+            {:entity => @entity, :check => c, :downtime =>
               check_presenter(c).downtime(start_time, end_time)}
           }
         end
@@ -52,7 +57,7 @@ module Flapjack
       private
 
         def checks
-          @check_list ||= @entity.check_list
+          @check_list ||= @entity.check_list.sort
         end
 
         def check_presenter(check)

data/lib/flapjack/gateways/email.rb

@@ -8,6 +8,8 @@ require 'socket'
 require 'em-synchrony'
 require 'em/protocols/smtpclient'
 
+require 'flapjack/utility'
+
 require 'flapjack/data/entity_check'
 
 module Flapjack
@@ -17,6 +19,8 @@ module Flapjack
 
       class << self
 
+        include Flapjack::Utility
+
         def start
           @logger.info("starting")
           @logger.debug("new email gateway pikelet with the following options: #{@config.inspect}")
@@ -32,8 +36,11 @@ module Flapjack
           @contact_last_name          = notification['contact_last_name']
           @state                      = notification['state']
           @summary                    = notification['summary']
+          @last_state                 = notification['last_state']
+          @last_summary               = notification['last_summary']
           @details                    = notification['details']
           @time                       = notification['time']
+          @relative                   = relative_time_ago(Time.at(@time))
           @entity_name, @check        = notification['event_id'].split(':', 2)
 
           entity_check = Flapjack::Data::EntityCheck.for_event_id(notification['event_id'],

data/lib/flapjack/gateways/email/alert.html.haml

@@ -41,7 +41,19 @@
     %tr
       %td
         %strong Time
-      %td= Time.at(@time.to_i).to_s
+      %td
+        = Time.at(@time.to_i).to_s
+        ( #{@relative} ago)
+
+    - if @last_state
+      %tr
+        %td Previous state
+        %td= @last_state
+
+    - if @last_summary
+      %tr
+        %td Previous summary
+        %td= @last_summary
 
 %p Cheers,
 %p Flapjack

data/lib/flapjack/gateways/email/alert.text.erb

@@ -2,11 +2,12 @@ Hi <%= @contact_first_name %>,
 
 Monitoring has detected the following:
 
-Entity: <%= @entity_name %>
-Check: <%= @check %>
-State: <%= @state %>
+Entity:  <%= @entity_name %>
+Check:   <%= @check %>
+State:   <%= @state %>
 Summary: <%= @summary %>
-Time: <%= Time.at(@time.to_i).to_s %>
+Time:    <%= Time.at(@time.to_i).to_s %> (<%= @relative %> ago)
+<%= @last_state ? "\nPrevious state: #{@last_state}" : '' %><%= @last_summary ? "\nPrevious summary: #{@last_summary}" : '' %>
 
 Cheers,
 Flapjack

data/lib/flapjack/gateways/jabber.rb

@@ -36,6 +36,8 @@ module Flapjack
       def initialize(opts = {})
         @config = opts[:config]
         @redis_config = opts[:redis_config]
+        @boot_time = opts[:boot_time]
+
         @redis = Flapjack::RedisPool.new(:config => @redis_config, :size => 2) # first will block
 
         @logger = opts[:logger]
@@ -146,56 +148,110 @@ module Flapjack
             error = "unknown entity" if entity_check.nil?
           end
 
-          if entity_check && entity_check.in_unscheduled_maintenance?
-            error = "#{event_id} is already acknowledged"
-          end
-
           if error
             msg = "ERROR - couldn't ACK #{ackid} - #{error}"
           else
-            msg = "ACKing #{entity_check.check} on #{entity_check.entity_name} (#{ackid})"
+            entity_name, check = event_id.split(':', 2)
+
+            if entity_check.in_unscheduled_maintenance?
+              # ack = entity_check.current_maintenance(:unscheduled => true)
+              # FIXME details from current?
+              msg = "Changing ACK for #{check} on #{entity_name} (#{ackid})"
+            else
+              msg = "ACKing #{check} on #{entity_name} (#{ackid})"
+            end
             action = Proc.new {
-              entity_check.create_acknowledgement('summary' => (comment || ''),
-                'acknowledgement_id' => ackid, 'duration' => duration)
+              Flapjack::Data::Event.create_acknowledgement(
+                entity_name, check,
+                :summary => (comment || ''),
+                :acknowledgement_id => ackid,
+                :duration => duration,
+                :redis => @redis
+              )
             }
           end
 
         when command =~ /^help$/
-          msg  = "commands: \n"
-          msg += "  ACKID <id> <comment> [duration: <time spec>] \n"
-          msg += "  find entities matching /pattern/ \n"
-          msg += "  test notifications for <entity>[:<check>] \n"
-          msg += "  identify \n"
-          msg += "  help \n"
+          msg = "commands: \n" +
+                "  ACKID <id> <comment> [duration: <time spec>] \n" +
+                "  find entities matching /pattern/ \n" +
+                "  test notifications for <entity>[:<check>] \n" +
+                "  tell me about <entity>[:<check>]" +
+                "  identify \n" +
+                "  help \n"
 
         when command =~ /^identify$/
-          t = Process.times
-          fqdn         = `/bin/hostname -f`.chomp
-          pid          = Process.pid
-          instance_id  = "#{@fqdn}:#{@pid}"
-          boot_time = Time.at(@redis.hget("executive_instance:#{instance_id}", 'boot_time').to_i)
-          msg  = "Flapjack #{Flapjack::VERSION} process #{pid} on #{fqdn} \n"
-          msg += "Boot time: #{boot_time}\n"
-          msg += "User CPU Time: #{t.utime}\n"
-          msg += "System CPU Time: #{t.stime}\n"
-          msg += `uname -a`.chomp + "\n"
-
-        when command =~ /^test notifications for\s+([a-z0-9\-\.]+)(:(.+))?$/i
+          t    = Process.times
+          fqdn = `/bin/hostname -f`.chomp
+          pid  = Process.pid
+          msg  = "Flapjack #{Flapjack::VERSION} process #{pid} on #{fqdn} \n" +
+                 "Boot time: #{@boot_time}\n" +
+                 "User CPU Time: #{t.utime}\n" +
+                 "System CPU Time: #{t.stime}\n" +
+                 `uname -a`.chomp + "\n"
+
+        when command =~ /^test notifications for\s+([a-z0-9\-\.]+)(?::(.+))?$/i
+          entity_name = $1
+          check_name  = $2 || 'test'
+
+          if entity = Flapjack::Data::Entity.find_by_name(entity_name, :redis => @redis)
+            msg = "so you want me to test notifications for entity: #{entity_name}, check: #{check_name} eh? ... well OK!"
+
+            summary = "Testing notifications to all contacts interested in entity: #{entity_name}, check: #{check_name}"
+            Flapjack::Data::Event.test_notifications(entity_name, check_name, :summary => summary, :redis => @redis)
+          else
+            msg = "yeah, no I can't see #{entity_name} in my systems"
+          end
+
+        when command =~ /^tell me about\s+([a-z0-9\-\.]+)(?::(.+))?$+/
           entity_name = $1
-          check_name  = $3 ? $3 : 'test'
+          check_name  = $2
 
-          msg = "so you want me to test notifications for entity: #{entity_name}, check: #{check_name} eh? ... well OK!"
+          if entity = Flapjack::Data::Entity.find_by_name(entity_name, :redis => @redis)
+            check_str = check_name.nil? ? '' : ", check: #{check_name}"
+            msg = "so you'd like details on entity: #{entity_name}#{check_str} hmm? ... OK!\n"
 
-          entity = Flapjack::Data::Entity.find_by_name(entity_name, :redis => @redis)
-          if entity
-            summary = "Testing notifications to all contacts interested in entity: #{entity.name}, check: #{check_name}"
+            current_time = Time.now
 
-            entity_check = Flapjack::Data::EntityCheck.for_entity(entity, check_name, :redis => @redis)
-            puts entity_check.inspect
-            entity_check.test_notifications('summary' => summary)
+            get_details = proc {|entity_check|
+              sched   = entity_check.current_maintenance(:scheduled => true)
+              unsched = entity_check.current_maintenance(:unscheduled => true)
 
+              if (sched || unsched) && check_name.nil?
+                check = entity_check.check
+                msg += "---\n#{entity_name}:#{check}\n"
+              end
+
+              unless sched.nil?
+                start  = Time.at(sched[:start_time])
+                finish = Time.at(sched[:start_time] + sched[:duration])
+                remain = time_period_in_words( (finish - current_time).ceil )
+                # TODO a simpler time format?
+                msg += "Currently in scheduled maintenance: #{start} -> #{finish} (#{remain} remaining)\n"
+              end
+
+              unless unsched.nil?
+                start  = Time.at(unsched[:start_time])
+                finish = Time.at(unsched[:start_time] + unsched[:duration])
+                remain = time_period_in_words( (finish - current_time).ceil )
+                # TODO a simpler time format?
+                msg += "Currently in unscheduled maintenance: #{start} -> #{finish} (#{remain} remaining)\n"
+              end
+            }
+
+            check_names = check_name.nil? ? entity.check_list.sort : [check_name]
+
+            if check_names.empty?
+              msg += "I couldn't find any checks for entity: #{entity_name}"
+            else
+              check_names.each do |check|
+                entity_check = Flapjack::Data::EntityCheck.for_entity(entity, check, :redis => @redis)
+                next if entity_check.nil?
+                get_details.call(entity_check)
+              end
+            end
           else
-            msg = "yeah, no i can't see #{entity_name} in my systems"
+            msg = "hmmm, I can't see #{entity_name} in my systems"
           end
 
         when command =~ /^(find )?entities matching\s+\/(.*)\/.*$/i

data/lib/flapjack/gateways/pagerduty.rb

@@ -177,12 +177,16 @@ module Flapjack
           end
 
           pg_acknowledged_by = acknowledged[:pg_acknowledged_by]
-          @logger.info "#{entity_check.entity_name}:#{check} is acknowledged in pagerduty, creating flapjack acknowledgement... "
+          entity_name = entity_check.entity_name
+          @logger.info "#{entity_name}:#{check} is acknowledged in pagerduty, creating flapjack acknowledgement... "
           who_text = ""
           if !pg_acknowledged_by.nil? && !pg_acknowledged_by['name'].nil?
             who_text = " by #{pg_acknowledged_by['name']}"
           end
-          entity_check.create_acknowledgement('summary' => "Acknowledged on PagerDuty" + who_text)
+          Flapjack::Data::Event.create_acknowledgement(
+            entity_name, check,
+            :summary => "Acknowledged on PagerDuty" + who_text,
+            :redis => @redis)
         end
 
       end

data/lib/flapjack/gateways/web.rb

@@ -53,7 +53,6 @@ module Flapjack
             use Flapjack::CommonLogger, access_logger
           end
 
-
         end
       end
 
@@ -191,6 +190,9 @@ module Flapjack
 
         @contacts                   = entity_check.contacts
 
+        @state_changes = entity_check.historical_states(nil, Time.now.to_i,
+                           :order => 'desc', :limit => 20)
+
         haml :check
       end
 
@@ -203,11 +205,14 @@ module Flapjack
         dur = ChronicDuration.parse(params[:duration] || '')
         @duration = (dur.nil? || (dur <= 0)) ? (4 * 60 * 60) : dur
 
-        entity_check = get_entity_check(@entity, @check)
-        return 404 if entity_check.nil?
+        return 404 if get_entity_check(@entity, @check).nil?
 
-        ack = entity_check.create_acknowledgement('summary' => (@summary || ''),
-          'acknowledgement_id' => @acknowledgement_id, 'duration' => @duration)
+        ack = Flapjack::Data::Event.create_acknowledgement(
+          @entity, @check,
+          :summary => (@summary || ''),
+          :acknowledgement_id => @acknowledgement_id,
+          :duration => @duration,
+          :redis => redis)
 
         redirect back
       end
@@ -321,9 +326,9 @@ module Flapjack
         entity_check = Flapjack::Data::EntityCheck.for_entity(entity,
           check, :redis => redis)
         latest_notif =
-          {:problem         => entity_check.last_problem_notification,
-           :recovery        => entity_check.last_recovery_notification,
-           :acknowledgement => entity_check.last_acknowledgement_notification
+          {:problem         => entity_check.last_notification_for_state(:problem)[:timestamp],
+           :recovery        => entity_check.last_notification_for_state(:recovery)[:timestamp],
+           :acknowledgement => entity_check.last_notification_for_state(:acknowledgement)[:timestamp]
           }.max_by {|n| n[1] || 0}
         [(entity_check.state       || '-'),
          (entity_check.last_change || '-'),