ddtrace 1.0.0 → 1.9.0

data/lib/datadog/profiling/exporter.rb

@@ -1,24 +1,84 @@
 # typed: true
 
-require 'datadog/profiling/transport/io/client'
+require_relative 'ext'
+require_relative 'tag_builder'
 
 module Datadog
   module Profiling
-    # Writes profiling data to a given transport
+    # Exports profiling data gathered by the multiple recorders in a `Flush`.
+    #
+    # @ivoanjo: Note that the recorder that gathers pprof data is special, since we use its start/finish/empty? to
+    # decide if there's data to flush, as well as the timestamp for that data.
+    # I could've made the whole design more generic, but I'm unsure if we'll ever have more than a handful of
+    # recorders, so I've decided to make it specific until we actually need to support more recorders.
+    #
     class Exporter
+      # Profiles with duration less than this will not be reported
+      PROFILE_DURATION_THRESHOLD_SECONDS = 1
+
+      private
+
       attr_reader \
-        :transport
+        :pprof_recorder,
+        :code_provenance_collector, # The code provenance collector acts both as collector and as a recorder
+        :minimum_duration_seconds,
+        :time_provider,
+        :last_flush_finish_at,
+        :created_at
+
+      public
+
+      def initialize(
+        pprof_recorder:,
+        code_provenance_collector:,
+        minimum_duration_seconds: PROFILE_DURATION_THRESHOLD_SECONDS,
+        time_provider: Time
+      )
+        @pprof_recorder = pprof_recorder
+        @code_provenance_collector = code_provenance_collector
+        @minimum_duration_seconds = minimum_duration_seconds
+        @time_provider = time_provider
+        @last_flush_finish_at = nil
+        @created_at = time_provider.now.utc
+      end
+
+      def flush
+        start, finish, uncompressed_pprof = pprof_recorder.serialize
+        @last_flush_finish_at = finish
+
+        return if uncompressed_pprof.nil? # We don't want to report empty profiles
 
-      def initialize(transport)
-        unless transport.is_a?(Profiling::Transport::Client)
-          raise ArgumentError, 'Unsupported transport for profiling exporter.'
+        if duration_below_threshold?(start, finish)
+          Datadog.logger.debug('Skipped exporting profiling events as profile duration is below minimum')
+          return
         end
 
-        @transport = transport
+        uncompressed_code_provenance = code_provenance_collector.refresh.generate_json if code_provenance_collector
+
+        Flush.new(
+          start: start,
+          finish: finish,
+          pprof_file_name: Datadog::Profiling::Ext::Transport::HTTP::PPROF_DEFAULT_FILENAME,
+          pprof_data: uncompressed_pprof.to_s,
+          code_provenance_file_name: Datadog::Profiling::Ext::Transport::HTTP::CODE_PROVENANCE_FILENAME,
+          code_provenance_data: uncompressed_code_provenance,
+          tags_as_array: Datadog::Profiling::TagBuilder.call(settings: Datadog.configuration).to_a,
+        )
+      end
+
+      def can_flush?
+        !duration_below_threshold?(last_flush_finish_at || created_at, time_provider.now.utc)
       end
 
-      def export(flush)
-        transport.send_profiling_flush(flush)
+      def reset_after_fork
+        @last_flush_finish_at = time_provider.now.utc
+        nil
+      end
+
+      private
+
+      def duration_below_threshold?(start, finish)
+        (finish - start) < minimum_duration_seconds
       end
     end
   end

data/lib/datadog/profiling/ext/forking.rb

@@ -5,6 +5,11 @@ module Datadog
     module Ext
       # Monkey patches `Kernel#fork`, adding a `Kernel#at_fork` callback mechanism which is used to restore
       # profiling abilities after the VM forks.
+      #
+      # Known limitations: Does not handle `BasicObject`s that include `Kernel` directly; e.g.
+      # `Class.new(BasicObject) { include(::Kernel); def call; fork { }; end }.new.call`.
+      #
+      # This will be fixed once we moved to hooking into `Process._fork`
       module Forking
         def self.supported?
           Process.respond_to?(:fork)
@@ -13,48 +18,34 @@ module Datadog
         def self.apply!
           return false unless supported?
 
-          modules = [::Process, ::Kernel]
-          # TODO: Ruby < 2.3 doesn't support Binding#receiver.
-          #       Remove "else #eval" clause when Ruby < 2.3 support is dropped.
-          # NOTE: Modifying the "main" object as we do here is (as far as I know) irreversible. During tests, this change
-          #       will stick around even if we otherwise stub `Process` and `Kernel`.
-          modules << (TOPLEVEL_BINDING.respond_to?(:receiver) ? TOPLEVEL_BINDING.receiver : TOPLEVEL_BINDING.eval('self'))
-
-          # Patch top-level binding, Kernel, Process.
-          # NOTE: We could instead do Kernel.module_eval { def fork; ... end }
-          #       however, this method rewrite is more invasive and irreversible.
-          #       It could also have collisions with other libraries that patch.
-          #       Opt to modify the inheritance of each relevant target instead.
-          modules.each do |mod|
-            clazz = if mod.class <= Module
-                      mod.singleton_class
-                    else
-                      mod.class
-                    end
-
-            clazz.prepend(Kernel)
-          end
+          [
+            ::Process.singleton_class, # Process.fork
+            ::Kernel.singleton_class,  # Kernel.fork
+            ::Object,                  # fork without explicit receiver (it's defined as a method in ::Kernel)
+            # Note: Modifying Object as we do here is irreversible. During tests, this
+            # change will stick around even if we otherwise stub `Process` and `Kernel`
+          ].each { |target| target.prepend(Kernel) }
+
+          ::Process.singleton_class.prepend(ProcessDaemonPatch)
         end
 
         # Extensions for kernel
+        #
+        # TODO: Consider hooking into `Process._fork` on Ruby 3.1+ instead, see
+        #       https://github.com/ruby/ruby/pull/5017 and https://bugs.ruby-lang.org/issues/17795
         module Kernel
-          FORK_STAGES = [:prepare, :parent, :child].freeze
-
           def fork
             # If a block is provided, it must be wrapped to trigger callbacks.
             child_block = if block_given?
                             proc do
                               # Trigger :child callback
-                              at_fork_blocks[:child].each(&:call) if at_fork_blocks.key?(:child)
+                              ddtrace_at_fork_blocks[:child].each(&:call) if ddtrace_at_fork_blocks.key?(:child)
 
                               # Invoke original block
                               yield
                             end
                           end
 
-            # Trigger :prepare callback
-            at_fork_blocks[:prepare].each(&:call) if at_fork_blocks.key?(:prepare)
-
             # Start fork
             # If a block is provided, use the wrapped version.
             result = child_block.nil? ? super : super(&child_block)
@@ -62,37 +53,45 @@ module Datadog
             # Trigger correct callbacks depending on whether we're in the parent or child.
             # If we're in the fork, result = nil: trigger child callbacks.
             # If we're in the parent, result = fork PID: trigger parent callbacks.
-            # rubocop:disable Style/IfInsideElse
-            if result.nil?
-              # Trigger :child callback
-              at_fork_blocks[:child].each(&:call) if at_fork_blocks.key?(:child)
-            else
-              # Trigger :parent callback
-              at_fork_blocks[:parent].each(&:call) if at_fork_blocks.key?(:parent)
-            end
-            # rubocop:enable Style/IfInsideElse
+            ddtrace_at_fork_blocks[:child].each(&:call) if result.nil? && ddtrace_at_fork_blocks.key?(:child)
 
             # Return PID from #fork
             result
           end
 
-          def at_fork(stage = :prepare, &block)
-            raise ArgumentError, 'Bad \'stage\' for ::at_fork' unless FORK_STAGES.include?(stage)
+          def at_fork(stage, &block)
+            raise ArgumentError, 'Bad \'stage\' for ::at_fork' unless stage == :child
 
-            at_fork_blocks[stage] = [] unless at_fork_blocks.key?(stage)
-            at_fork_blocks[stage] << block
+            ddtrace_at_fork_blocks[stage] = [] unless ddtrace_at_fork_blocks.key?(stage)
+            ddtrace_at_fork_blocks[stage] << block
           end
 
           module_function
 
-          def at_fork_blocks
+          def ddtrace_at_fork_blocks
             # Blocks should be shared across all users of this module,
             # e.g. Process#fork, Kernel#fork, etc. should all invoke the same callbacks.
             # rubocop:disable Style/ClassVars
-            @@at_fork_blocks ||= {}
+            @@ddtrace_at_fork_blocks ||= {}
             # rubocop:enable Style/ClassVars
           end
         end
+
+        # A call to Process.daemon ( https://rubyapi.org/3.1/o/process#method-c-daemon ) forks the current process and
+        # keeps executing code in the child process, killing off the parent, thus effectively replacing it.
+        #
+        # This monkey patch makes the `Kernel#at_fork` mechanism defined above also work in this situation.
+        module ProcessDaemonPatch
+          def daemon(*args)
+            ddtrace_at_fork_blocks = Datadog::Profiling::Ext::Forking::Kernel.ddtrace_at_fork_blocks
+
+            result = super
+
+            ddtrace_at_fork_blocks[:child].each(&:call) if ddtrace_at_fork_blocks.key?(:child)
+
+            result
+          end
+        end
       end
     end
   end

data/lib/datadog/profiling/ext.rb

@@ -9,6 +9,7 @@ module Datadog
       ENV_AGENTLESS = 'DD_PROFILING_AGENTLESS'.freeze
       ENV_ENDPOINT_COLLECTION_ENABLED = 'DD_PROFILING_ENDPOINT_COLLECTION_ENABLED'.freeze
 
+      # TODO: Consider removing this once the Ruby-based pprof encoding is removed and replaced by libdatadog
       module Pprof
         LABEL_KEY_LOCAL_ROOT_SPAN_ID = 'local root span id'.freeze
         LABEL_KEY_SPAN_ID = 'span id'.freeze
@@ -22,15 +23,10 @@ module Datadog
 
       module Transport
         module HTTP
-          URI_TEMPLATE_DD_API = 'https://intake.profile.%s/'.freeze
-
-          FORM_FIELD_RECORDING_START = 'start'.freeze
-          FORM_FIELD_RECORDING_END = 'end'.freeze
-          FORM_FIELD_FAMILY = 'family'.freeze
           FORM_FIELD_TAG_ENV = 'env'.freeze
           FORM_FIELD_TAG_HOST = 'host'.freeze
           FORM_FIELD_TAG_LANGUAGE = 'language'.freeze
-          FORM_FIELD_TAG_PID = 'pid'.freeze
+          FORM_FIELD_TAG_PID = 'process_id'.freeze
           FORM_FIELD_TAG_PROFILER_VERSION = 'profiler_version'.freeze
           FORM_FIELD_TAG_RUNTIME = 'runtime'.freeze
           FORM_FIELD_TAG_RUNTIME_ENGINE = 'runtime_engine'.freeze
@@ -39,17 +35,9 @@ module Datadog
           FORM_FIELD_TAG_RUNTIME_VERSION = 'runtime_version'.freeze
           FORM_FIELD_TAG_SERVICE = 'service'.freeze
           FORM_FIELD_TAG_VERSION = 'version'.freeze
-          FORM_FIELD_TAGS = 'tags'.freeze
-          FORM_FIELD_INTAKE_VERSION = 'version'.freeze
-
-          HEADER_CONTENT_TYPE = 'Content-Type'.freeze
-          HEADER_CONTENT_TYPE_OCTET_STREAM = 'application/octet-stream'.freeze
-
-          FORM_FIELD_PPROF_DATA = 'data[rubyprofile.pprof]'.freeze
-          PPROF_DEFAULT_FILENAME = 'rubyprofile.pprof.gz'.freeze
 
-          FORM_FIELD_CODE_PROVENANCE_DATA = 'data[code-provenance.json]'.freeze
-          CODE_PROVENANCE_FILENAME = 'code-provenance.json.gz'.freeze
+          PPROF_DEFAULT_FILENAME = 'rubyprofile.pprof'.freeze
+          CODE_PROVENANCE_FILENAME = 'code-provenance.json'.freeze
         end
       end
     end

data/lib/datadog/profiling/flush.rb

@@ -1,69 +1,38 @@
 # typed: false
 
-require 'datadog/core/environment/identity'
-require 'datadog/core/environment/socket'
-
 module Datadog
   module Profiling
+    # Represents a collection of events of a specific type being flushed.
+    EventGroup = Struct.new(:event_class, :events)
+
     # Entity class used to represent metadata for a given profile
-    Flush = Struct.new(
-      :start,
-      :finish,
-      :event_groups,
-      :event_count,
-      :code_provenance,
-      :runtime_id,
-      :service,
-      :env,
-      :version,
-      :host,
-      :language,
-      :runtime_engine,
-      :runtime_platform,
-      :runtime_version,
-      :profiler_version,
-      :tags
-    ) do
+    class Flush
+      attr_reader \
+        :start,
+        :finish,
+        :pprof_file_name,
+        :pprof_data, # gzipped pprof bytes
+        :code_provenance_file_name,
+        :code_provenance_data, # gzipped json bytes
+        :tags_as_array
+
       def initialize(
         start:,
         finish:,
-        event_groups:,
-        event_count:,
-        code_provenance:,
-        runtime_id: Core::Environment::Identity.id,
-        service: Datadog.configuration.service,
-        env: Datadog.configuration.env,
-        version: Datadog.configuration.version,
-        host: Core::Environment::Socket.hostname,
-        language: Core::Environment::Identity.lang,
-        runtime_engine: Core::Environment::Identity.lang_engine,
-        runtime_platform: Core::Environment::Identity.lang_platform,
-        runtime_version: Core::Environment::Identity.lang_version,
-        profiler_version: Core::Environment::Identity.tracer_version,
-        tags: Datadog.configuration.tags
+        pprof_file_name:,
+        pprof_data:,
+        code_provenance_file_name:,
+        code_provenance_data:,
+        tags_as_array:
       )
-        super(
-          start,
-          finish,
-          event_groups,
-          event_count,
-          code_provenance,
-          runtime_id,
-          service,
-          env,
-          version,
-          host,
-          language,
-          runtime_engine,
-          runtime_platform,
-          runtime_version,
-          profiler_version,
-          tags,
-        )
+        @start = start
+        @finish = finish
+        @pprof_file_name = pprof_file_name
+        @pprof_data = pprof_data
+        @code_provenance_file_name = code_provenance_file_name
+        @code_provenance_data = code_provenance_data
+        @tags_as_array = tags_as_array
       end
     end
-
-    # Represents a collection of events of a specific type being flushed.
-    EventGroup = Struct.new(:event_class, :events).freeze
   end
 end

data/lib/datadog/profiling/http_transport.rb

@@ -0,0 +1,132 @@
+# typed: false
+
+module Datadog
+  module Profiling
+    # Used to report profiling data to Datadog.
+    # Methods prefixed with _native_ are implemented in `http_transport.c`
+    class HttpTransport
+      def initialize(agent_settings:, site:, api_key:, upload_timeout_seconds:)
+        @upload_timeout_milliseconds = (upload_timeout_seconds * 1_000).to_i
+
+        validate_agent_settings(agent_settings)
+
+        @exporter_configuration =
+          if agentless?(site, api_key)
+            [:agentless, site, api_key]
+          else
+            [:agent, base_url_from(agent_settings)]
+          end
+
+        status, result = validate_exporter(@exporter_configuration)
+
+        raise(ArgumentError, "Failed to initialize transport: #{result}") if status == :error
+      end
+
+      def export(flush)
+        status, result = do_export(
+          exporter_configuration: @exporter_configuration,
+          upload_timeout_milliseconds: @upload_timeout_milliseconds,
+
+          # why "timespec"?
+          # libdatadog represents time using POSIX's struct timespec, see
+          # https://www.gnu.org/software/libc/manual/html_node/Time-Types.html
+          # aka it represents the seconds part separate from the nanoseconds part
+          start_timespec_seconds: flush.start.tv_sec,
+          start_timespec_nanoseconds: flush.start.tv_nsec,
+          finish_timespec_seconds: flush.finish.tv_sec,
+          finish_timespec_nanoseconds: flush.finish.tv_nsec,
+
+          pprof_file_name: flush.pprof_file_name,
+          pprof_data: flush.pprof_data,
+          code_provenance_file_name: flush.code_provenance_file_name,
+          code_provenance_data: flush.code_provenance_data,
+
+          tags_as_array: flush.tags_as_array,
+        )
+
+        if status == :ok
+          if (200..299).cover?(result)
+            Datadog.logger.debug('Successfully reported profiling data')
+            true
+          else
+            Datadog.logger.error("Failed to report profiling data: server returned unexpected HTTP #{result} status code")
+            false
+          end
+        else
+          Datadog.logger.error("Failed to report profiling data: #{result}")
+          false
+        end
+      end
+
+      # Used to log soft failures in `ddog_Vec_tag_push` (e.g. we still report the profile in these cases)
+      # Called from native code
+      def self.log_failure_to_process_tag(failure_details)
+        Datadog.logger.warn("Failed to add tag to profiling request: #{failure_details}")
+      end
+
+      private
+
+      def base_url_from(agent_settings)
+        case agent_settings.adapter
+        when Datadog::Transport::Ext::HTTP::ADAPTER
+          "#{agent_settings.ssl ? 'https' : 'http'}://#{agent_settings.hostname}:#{agent_settings.port}/"
+        when Datadog::Transport::Ext::UnixSocket::ADAPTER
+          "unix://#{agent_settings.uds_path}"
+        else
+          raise ArgumentError, "Unexpected adapter: #{agent_settings.adapter}"
+        end
+      end
+
+      def validate_agent_settings(agent_settings)
+        supported_adapters = [Datadog::Transport::Ext::HTTP::ADAPTER, Datadog::Transport::Ext::UnixSocket::ADAPTER]
+        unless supported_adapters.include?(agent_settings.adapter)
+          raise ArgumentError,
+            "Unsupported transport configuration for profiling: Adapter #{agent_settings.adapter} " \
+                        ' is not supported'
+        end
+
+        if agent_settings.deprecated_for_removal_transport_configuration_proc
+          Datadog.logger.warn(
+            'Ignoring custom c.tracing.transport_options setting as the profiler does not support it.'
+          )
+        end
+      end
+
+      def agentless?(site, api_key)
+        site && api_key && Core::Environment::VariableHelpers.env_to_bool(Profiling::Ext::ENV_AGENTLESS, false)
+      end
+
+      def validate_exporter(exporter_configuration)
+        self.class._native_validate_exporter(exporter_configuration)
+      end
+
+      def do_export(
+        exporter_configuration:,
+        upload_timeout_milliseconds:,
+        start_timespec_seconds:,
+        start_timespec_nanoseconds:,
+        finish_timespec_seconds:,
+        finish_timespec_nanoseconds:,
+        pprof_file_name:,
+        pprof_data:,
+        code_provenance_file_name:,
+        code_provenance_data:,
+        tags_as_array:
+      )
+        self.class._native_do_export(
+          exporter_configuration,
+          upload_timeout_milliseconds,
+          start_timespec_seconds,
+          start_timespec_nanoseconds,
+          finish_timespec_seconds,
+          finish_timespec_nanoseconds,
+          pprof_file_name,
+          pprof_data,
+          code_provenance_file_name,
+          code_provenance_data,
+          tags_as_array,
+        )
+      end
+    end
+  end
+end

data/lib/datadog/profiling/load_native_extension.rb

@@ -0,0 +1,22 @@
+# typed: ignore
+
+# This file is used to load the profiling native extension. It works in two steps:
+#
+# 1. Load the ddtrace_profiling_loader extension. This extension will be used to load the actual extension, but in
+#    a special way that avoids exposing native-level code symbols. See `ddtrace_profiling_loader.c` for more details.
+#
+# 2. Use the Datadog::Profiling::Loader exposed by the ddtrace_profiling_loader extension to load the actual
+#    profiling native extension.
+#
+# All code on this file is on-purpose at the top-level; this makes it so this file is executed only once,
+# the first time it gets required, to avoid any issues with the native extension being initialized more than once.
+
+require "ddtrace_profiling_loader.#{RUBY_VERSION}_#{RUBY_PLATFORM}"
+
+extension_name = "ddtrace_profiling_native_extension.#{RUBY_VERSION}_#{RUBY_PLATFORM}"
+full_file_path = "#{__dir__}/../../#{extension_name}.#{RbConfig::CONFIG['DLEXT']}"
+init_function_name = "Init_#{extension_name.split('.').first}"
+
+status, result = Datadog::Profiling::Loader._native_load(full_file_path, init_function_name)
+
+raise "Failure to load #{extension_name} due to #{result}" if status == :error

data/lib/datadog/profiling/old_recorder.rb

@@ -0,0 +1,109 @@
+# typed: true
+
+require_relative 'buffer'
+require_relative 'encoding/profile'
+
+module Datadog
+  module Profiling
+    # Stores profiling events gathered by the `Stack` collector
+    class OldRecorder
+      attr_reader :max_size
+
+      def initialize(
+        event_classes,
+        max_size,
+        last_flush_time: Time.now.utc
+      )
+        @buffers = {}
+        @last_flush_time = last_flush_time
+        @max_size = max_size
+
+        # Add a buffer for each class
+        event_classes.each do |event_class|
+          @buffers[event_class] = Profiling::Buffer.new(max_size)
+        end
+
+        # Event classes can only be added ahead of time
+        @buffers.freeze
+      end
+
+      def [](event_class)
+        @buffers[event_class]
+      end
+
+      def push(events)
+        if events.is_a?(Array)
+          # Push multiple events
+          event_class = events.first.class
+          raise UnknownEventError, event_class unless @buffers.key?(event_class)
+
+          @buffers[event_class].concat(events)
+        else
+          # Push single event
+          event_class = events.class
+          raise UnknownEventError, event_class unless @buffers.key?(event_class)
+
+          @buffers[event_class].push(events)
+        end
+      end
+
+      def serialize
+        event_count = 0
+
+        event_groups, start, finish = update_time do
+          @buffers.collect do |event_class, buffer|
+            events = buffer.pop
+            next if events.empty?
+
+            event_count += events.length
+            EventGroup.new(event_class, events)
+          end.compact
+        end
+
+        return if event_count.zero? # We don't want to report empty profiles
+
+        encoded_pprof =
+          Datadog::Profiling::Encoding::Profile::Protobuf.encode(
+            event_count: event_count,
+            event_groups: event_groups,
+            start: start,
+            finish: finish,
+          )
+
+        [start, finish, encoded_pprof]
+      end
+
+      def reset_after_fork
+        Datadog.logger.debug('Resetting OldRecorder in child process after fork')
+
+        # NOTE: A bit of a heavy-handed approach, but it doesn't happen often and this class will be removed soon anyway
+        serialize
+        nil
+      end
+
+      # Error when event of an unknown type is used with the OldRecorder
+      class UnknownEventError < StandardError
+        attr_reader :event_class
+
+        def initialize(event_class)
+          @event_class = event_class
+        end
+
+        def message
+          @message ||= "Unknown event class '#{event_class}' for profiling recorder."
+        end
+      end
+
+      private
+
+      def update_time
+        start = @last_flush_time
+        result = yield
+        @last_flush_time = Time.now.utc
+
+        # Return event groups, start time, finish time
+        [result, start, @last_flush_time]
+      end
+    end
+  end
+end

data/lib/datadog/profiling/pprof/builder.rb

@@ -2,10 +2,10 @@
 
 # typed: true
 
-require 'datadog/profiling/flush'
-require 'datadog/profiling/pprof/message_set'
-require 'datadog/profiling/pprof/string_table'
-require 'datadog/core/utils/time'
+require_relative '../flush'
+require_relative 'message_set'
+require_relative 'string_table'
+require_relative '../../core/utils/time'
 
 module Datadog
   module Profiling

data/lib/datadog/profiling/pprof/converter.rb

@@ -1,6 +1,6 @@
 # typed: true
 
-require 'datadog/profiling/ext'
+require_relative '../ext'
 
 module Datadog
   module Profiling

data/lib/datadog/profiling/pprof/message_set.rb

@@ -1,6 +1,6 @@
 # typed: true
 
-require 'datadog/core/utils/object_set'
+require_relative '../../core/utils/object_set'
 
 module Datadog
   module Profiling

data/lib/datadog/profiling/pprof/stack_sample.rb

@@ -1,9 +1,9 @@
 # typed: true
 
-require 'datadog/profiling/ext'
-require 'datadog/profiling/events/stack'
-require 'datadog/profiling/pprof/builder'
-require 'datadog/profiling/pprof/converter'
+require_relative '../ext'
+require_relative '../events/stack'
+require_relative 'builder'
+require_relative 'converter'
 
 module Datadog
   module Profiling