newrelic_rpm 3.5.2.17 → 3.5.3.24

data/lib/new_relic/agent/sql_sampler.rb

@@ -1,5 +1,10 @@
+require 'zlib'
+require 'base64'
+require 'digest/md5'
+
 require 'new_relic/agent'
 require 'new_relic/control'
+
 module NewRelic
   module Agent
 
@@ -218,21 +223,35 @@ module NewRelic
         Agent.config[:'slow_sql.explain_enabled']
       end
 
-      def to_json(*a)
-        [@path, @url, @sql_id, @sql, @database_metric_name, @call_count, @total_call_time, @min_call_time, @max_call_time, @params].to_json(*a)
+      def to_collector_array(marshaller)
+        params = if marshaller.respond_to?(:encode_compress)
+          marshaller.encode_compress(@params)
+        else
+          @params
+        end
+
+        [ @path, @url, @sql_id, @sql, @database_metric_name, @call_count,
+          Helper.time_to_millis(@total_call_time),
+          Helper.time_to_millis(@min_call_time),
+          Helper.time_to_millis(@max_call_time),
+          params ]
       end
 
       private
 
-      def consistent_hash(string)
-        if NewRelic::LanguageSupport.using_version?('1.9.2')
-          # String#hash is salted differently on every VM start in 1.9
-          require 'digest/md5'
-          Digest::MD5.hexdigest(string).hex
+      def compress(data)
+        if NewRelic::Agent::NewRelicService::JsonMarshaller.is_supported?
+          require 'json'
+          Base64.encode64(Zlib::Deflate.deflate(JSON.dump(data), Zlib::DEFAULT_COMPRESSION))
         else
-          string.hash
-        end.modulo(2**31-1)
-        # modulo ensures sql_id fits in an INT(11)
+          data
+        end
+      end
+
+      def consistent_hash(string)
+        # need to hash the same way in every process
+        Digest::MD5.hexdigest(string).hex \
+          .modulo(2**31-1)  # ensure sql_id fits in an INT(11)
       end
     end
   end

data/lib/new_relic/agent/stats_engine/metric_stats.rb

@@ -63,6 +63,19 @@ module NewRelic
 
             backtrace = t.backtrace.map { |b| "\t#{b}" }.join("\n")
             "\t#{t}\n#{backtrace}"
+
+          rescue Exception => e
+            # JRuby 1.7.0 has a nasty habit of raising a
+            # java.lang.NullPointerException when we iterate through threads
+            # asking for backtraces.  This line allows us to swallow java
+            # exceptions without referencing their classes (since they don't
+            # exist in MRI).  It also prevents us from swallowing signals or
+            # other nasty things that can happen when you rescue Exception.
+            NewRelic::Control.instance.log.warn(
+              "Error collecting thread backtraces: #{e.class.name}: #{e.message}")
+            NewRelic::Control.instance.log.debug( e.backtrace.join("\n") )
+
+            raise e if e.class.ancestors.include? Exception
           end
         end
 
@@ -106,6 +119,33 @@ module NewRelic
           stats_hash[NewRelic::MetricSpec.new(metric_name, scope_name)]
         end
 
+
+        # Helper method for timing supportability metrics
+        def record_supportability_metrics_timed(metrics)
+          start_time = Time.now
+          yield
+          end_time = Time.now
+          duration = (end_time - start_time).to_f
+        ensure
+          record_supportability_metrics(duration, metrics) do |value, metric|
+            metric.record_data_point(value)
+          end
+        end
+
+        # Helper for recording a straight value into the count
+        def record_supportability_metrics_count(value, *metrics)
+          record_supportability_metrics(value, *metrics) do |value, metric|
+            metric.call_count = value
+          end
+        end
+
+        # Helper method for recording supportability metrics consistently
+        def record_supportability_metrics(value, *metrics)
+          metrics.each do |metric|
+              yield(value, get_stats_no_scope("Supportability/#{metric}"))
+          end
+        end
+
         # This module was extracted from the harvest method and should
         # be refactored
         module Harvest

data/lib/new_relic/agent/stats_engine/samplers.rb

@@ -27,7 +27,7 @@ module Agent
         # start up a thread that will periodically poll for metric samples
         return if periodic_samplers.empty?
 
-        @sampler_thread = Thread.new do
+        @sampler_thread = NewRelic::Agent::Thread.new('Sampler Tasks') do
           loop do
             now = Time.now
             begin
@@ -40,7 +40,6 @@ module Agent
             end
           end
         end
-        @sampler_thread['newrelic_label'] = 'Sampler Tasks'
       end
 
       private

data/lib/new_relic/agent/thread.rb

@@ -0,0 +1,27 @@
+module NewRelic
+  module Agent
+
+    class Thread < ::Thread
+      def initialize(label)
+        NewRelic::Agent.logger.debug("Creating New Relic thread: #{label}")
+        self[:newrelic_label] = label
+        super
+      end
+
+      def self.bucket_thread(thread, profile_agent_code)
+        if thread.key?(:newrelic_label)
+          return profile_agent_code ? :agent : :ignore
+        elsif !thread[:newrelic_metric_frame].nil?
+          thread[:newrelic_metric_frame].request.nil? ? :background : :request
+        else
+          :other
+        end
+      end
+
+      def self.scrub_backtrace(thread, profile_agent_code)
+        return thread.backtrace if profile_agent_code
+        thread.backtrace.select {|t| t !~ /\/newrelic_rpm-\d/ }
+      end
+    end
+  end
+end

data/lib/new_relic/agent/thread_profiler.rb

@@ -0,0 +1,295 @@
+require 'new_relic/agent/thread'
+require 'new_relic/agent/worker_loop'
+
+module NewRelic
+  module Agent
+
+    class ThreadProfiler
+
+      attr_reader :profile
+
+      def self.is_supported?
+        RUBY_VERSION >= "1.9.2"
+      end
+
+      def start(profile_id, duration, interval, profile_agent_code)
+        if !ThreadProfiler.is_supported?
+          log.debug("Not starting thread profile as it isn't supported on this environment")
+          @profile = nil
+        else
+          log.debug("Starting thread profile. profile_id=#{profile_id}, duration=#{duration}")
+          @profile = ThreadProfile.new(profile_id, duration, interval, profile_agent_code)
+          @profile.run
+        end
+      end
+
+      def stop(report_data)
+        @profile.stop unless @profile.nil?
+        @profile = nil if !report_data
+      end
+
+      def harvest
+        profile = @profile
+        @profile = nil
+        profile
+      end
+
+      def respond_to_commands(commands, &notify_results)
+        return if commands.empty? || commands.first.size < 2 
+
+        # Doesn't deal with multiple commands in the return set  as 
+        # we currently only have start/stop of thread profiling
+        command_id = commands.first[0]
+        command = commands.first[1]
+
+        name = command["name"]
+        arguments = command["arguments"]
+
+        case name
+          when "start_profiler"
+            start_unless_running_and_notify(command_id, arguments, &notify_results)
+
+          when "stop_profiler"
+            stop_and_notify(command_id, arguments, &notify_results)
+        end
+      end
+
+      def running?
+        !@profile.nil?
+      end
+
+      def finished?
+        @profile && @profile.finished?
+      end
+
+      private
+
+      def start_unless_running_and_notify(command_id, arguments)
+        profile_id = arguments.fetch("profile_id", -1)
+        duration =   arguments.fetch("duration", 120)
+        interval =   arguments.fetch("sample_period", 0.1)
+        profile_agent_code = arguments.fetch("profile_agent_code", true)
+
+        if running?
+          msg = "Profile already in progress. Ignoring agent command to start another."
+          log.debug(msg)
+          yield(command_id, msg) if block_given?
+        else
+          start(profile_id, duration, interval, profile_agent_code)
+          yield(command_id) if block_given?
+        end
+      end
+
+      def stop_and_notify(command_id, arguments)
+        report_data = arguments.fetch("report_data", true)
+        stop(report_data)
+        yield(command_id) if block_given?
+      end
+
+      def log
+        NewRelic::Agent.logger
+      end
+    end
+
+    class ThreadProfile
+
+      attr_reader :profile_id, 
+        :traces, 
+        :profile_agent_code, :interval, 
+        :poll_count, :sample_count, 
+        :start_time, :stop_time
+
+      def initialize(profile_id, duration, interval, profile_agent_code)
+        @profile_id = profile_id
+        @profile_agent_code = profile_agent_code
+
+        @worker_loop = NewRelic::Agent::WorkerLoop.new(:duration => duration)
+        @interval = interval
+        @finished = false
+
+        @traces = {
+          :agent => [],
+          :background => [],
+          :other => [],
+          :request => []
+        }
+        @flattened_nodes = []
+
+        @poll_count = 0
+        @sample_count = 0
+      end
+
+      def run
+        Thread.new('Thread Profiler') do
+          @start_time = now_in_millis
+
+          @worker_loop.run(@interval) do
+            NewRelic::Agent.instance.stats_engine.
+              record_supportability_metrics_timed("ThreadProfiler/PollingTime") do
+
+              @poll_count += 1
+              Thread.list.each do |t|
+                @sample_count += 1
+
+                bucket = Thread.bucket_thread(t, @profile_agent_code)
+                backtrace = Thread.scrub_backtrace(t, @profile_agent_code)
+                aggregate(backtrace, @traces[bucket]) unless bucket == :ignore
+              end
+            end
+          end
+
+          mark_done
+          log.debug("Finished thread profile. Will send with next harvest.")
+        end
+      end
+
+      def stop
+        @worker_loop.stop
+        mark_done
+        log.debug("Stopping thread profile.")
+      end
+
+      def aggregate(trace, trees=@traces[:request], parent=nil)
+        return nil if trace.nil? || trace.empty?
+        node = Node.new(trace.last)
+        existing = trees.find {|n| n == node}
+
+        if existing.nil?
+          existing = node
+          @flattened_nodes << node
+        end
+
+        if parent
+          parent.add_child(node)
+        else
+          trees << node unless trees.include? node
+        end
+
+        existing.runnable_count += 1
+        aggregate(trace[0..-2], existing.children, existing)
+
+        existing
+      end
+
+      def prune!(count_to_keep)
+        @flattened_nodes.sort!(&:order_for_pruning)
+
+        NewRelic::Agent.instance.stats_engine.
+          record_supportability_metrics_count(@flattened_nodes.size, "ThreadProfiler/NodeCount")
+
+        mark_for_pruning(@flattened_nodes, count_to_keep)
+
+        traces.each { |_, nodes| Node.prune!(nodes) }
+      end
+
+      THREAD_PROFILER_NODES = 20_000
+
+      def to_compressed_array
+        prune!(THREAD_PROFILER_NODES)
+
+        traces = {
+          "OTHER" => @traces[:other].map{|t| t.to_array },
+          "REQUEST" => @traces[:request].map{|t| t.to_array },
+          "AGENT" => @traces[:agent].map{|t| t.to_array },
+          "BACKGROUND" => @traces[:background].map{|t| t.to_array }
+        }
+
+        [[@profile_id,
+          @start_time.to_f, @stop_time.to_f,
+          @poll_count, 
+          ThreadProfile.compress(JSON.dump(traces)),
+          @sample_count, 0]]
+      end
+
+      def now_in_millis
+        Time.now.to_f * 1_000
+      end
+
+      def finished?
+        @finished
+      end
+
+      def mark_done
+        @finished = true
+        @stop_time = now_in_millis
+      end
+
+      def mark_for_pruning(nodes, count_to_keep)
+        to_prune = nodes[count_to_keep..-1] || []
+        to_prune.each { |n| n.to_prune = true }
+      end
+
+      def self.flattened_nodes(nodes)
+        nodes.map { |n| [n, flattened_nodes(n.children)] }.flatten
+      end
+
+      def self.compress(json)
+        compressed = Base64.encode64(Zlib::Deflate.deflate(json, Zlib::DEFAULT_COMPRESSION))
+      end
+
+      def self.parse_backtrace(trace)
+        trace.map do |line|
+          line =~ /(.*)\:(\d+)\:in `(.*)'/
+          { :method => $3, :line_no => $2.to_i, :file => $1 }
+        end
+      end
+
+      class Node
+        attr_reader :file, :method, :line_no, :children
+        attr_accessor :runnable_count, :to_prune, :depth
+
+        def initialize(line, parent=nil)
+          line =~ /(.*)\:(\d+)\:in `(.*)'/
+          @file = $1
+          @method = $3
+          @line_no = $2.to_i
+          @children = []
+          @runnable_count = 0
+          @to_prune = false
+          @depth = 0
+
+          parent.add_child(self) if parent
+        end
+
+        def ==(other)
+          @file == other.file &&
+            @method == other.method &&
+            @line_no == other.line_no
+        end
+
+        def total_count
+          @runnable_count
+        end
+
+        # Descending order on count, ascending on depth of nodes
+        def order_for_pruning(y)
+          [-runnable_count, depth] <=> [-y.runnable_count, y.depth]
+        end
+
+        def to_array
+          [[@file, @method, @line_no],
+            @runnable_count, 0,
+            @children.map {|c| c.to_array}]
+        end
+
+        def add_child(child)
+          child.depth = @depth + 1
+          @children << child unless @children.include? child
+        end
+
+        def prune!
+          Node.prune!(@children)
+        end
+
+        def self.prune!(kids)
+          kids.delete_if { |child| child.to_prune }
+          kids.each { |child| child.prune! }
+        end
+      end
+
+      def log
+        NewRelic::Agent.logger
+      end
+    end
+  end
+end