datadog 2.12.1 → 2.19.0

data/lib/datadog/profiling/collectors/thread_context.rb

@@ -21,7 +21,8 @@ module Datadog
           endpoint_collection_enabled:,
           timeline_enabled:,
           waiting_for_gvl_threshold_ns:,
-          otel_context_enabled:
+          otel_context_enabled:,
+          native_filenames_enabled:
         )
           tracer_context_key = safely_extract_context_key_from(tracer)
           self.class._native_initialize(
@@ -33,6 +34,7 @@ module Datadog
             timeline_enabled: timeline_enabled,
             waiting_for_gvl_threshold_ns: waiting_for_gvl_threshold_ns,
             otel_context_enabled: otel_context_enabled,
+            native_filenames_enabled: validate_native_filenames(native_filenames_enabled),
           )
         end
 
@@ -44,6 +46,7 @@ module Datadog
           timeline_enabled: false,
           waiting_for_gvl_threshold_ns: 10_000_000,
           otel_context_enabled: false,
+          native_filenames_enabled: true,
           **options
         )
           new(
@@ -54,13 +57,14 @@ module Datadog
             timeline_enabled: timeline_enabled,
             waiting_for_gvl_threshold_ns: waiting_for_gvl_threshold_ns,
             otel_context_enabled: otel_context_enabled,
+            native_filenames_enabled: native_filenames_enabled,
             **options,
           )
         end
 
         def inspect
           # Compose Ruby's default inspect with our custom inspect for the native parts
-          result = super()
+          result = super
           result[-1] = "#{self.class._native_inspect(self)}>"
           result
         end
@@ -81,6 +85,17 @@ module Datadog
           context = provider.instance_variable_get(:@context)
           context&.instance_variable_get(:@key)
         end
+
+        def validate_native_filenames(native_filenames_enabled)
+          if native_filenames_enabled && !Datadog::Profiling::Collectors::Stack._native_filenames_available?
+            Datadog.logger.debug(
+              "Native filenames are enabled, but the required dladdr API was not available. Disabling native filenames."
+            )
+            false
+          else
+            native_filenames_enabled
+          end
+        end
       end
     end
   end

data/lib/datadog/profiling/component.rb

@@ -67,6 +67,7 @@ module Datadog
           allocation_profiling_enabled: allocation_profiling_enabled,
           allocation_counting_enabled: settings.profiling.advanced.allocation_counting_enabled,
           gvl_profiling_enabled: enable_gvl_profiling?(settings, logger),
+          sighandler_sampling_enabled: settings.profiling.advanced.sighandler_sampling_enabled,
         )
 
         internal_metadata = {
@@ -96,6 +97,7 @@ module Datadog
           timeline_enabled: timeline_enabled,
           waiting_for_gvl_threshold_ns: settings.profiling.advanced.waiting_for_gvl_threshold_ns,
           otel_context_enabled: settings.profiling.advanced.preview_otel_context_enabled,
+          native_filenames_enabled: settings.profiling.advanced.native_filenames_enabled,
         )
       end
 
@@ -221,13 +223,14 @@ module Datadog
         end
 
         unless allocation_profiling_enabled
-          raise ArgumentError, "Heap profiling requires allocation profiling to be enabled"
+          logger.warn(
+            "Heap profiling was requested but allocation profiling is not enabled. " \
+            "Heap profiling has been disabled."
+          )
+          return false
         end
 
-        logger.warn(
-          "Enabled experimental heap profiling: heap_sample_rate=#{heap_sample_rate}. This is experimental, not " \
-          "recommended, and will increase overhead!"
-        )
+        logger.debug("Enabled heap profiling: heap_sample_rate=#{heap_sample_rate}")
 
         true
       end
@@ -237,10 +240,6 @@ module Datadog
 
         return false unless heap_profiling_enabled && heap_size_profiling_enabled
 
-        logger.warn(
-          "Enabled experimental heap size profiling. This is experimental, not recommended, and will increase overhead!"
-        )
-
         true
       end
 

data/lib/datadog/profiling/encoded_profile.rb

@@ -0,0 +1,11 @@
+# frozen_string_literal: true
+
+module Datadog
+  module Profiling
+    # This class exists to wrap a ddog_prof_EncodedProfile into a Ruby object
+    #
+    # This class is not empty; all of this class is implemented as native code.
+    class EncodedProfile # rubocop:disable Lint/EmptyClass
+    end
+  end
+end

data/lib/datadog/profiling/exporter.rb

@@ -26,7 +26,8 @@ module Datadog
         :last_flush_finish_at,
         :created_at,
         :internal_metadata,
-        :info_json
+        :info_json,
+        :sequence_tracker
 
       public
 
@@ -37,7 +38,8 @@ module Datadog
         code_provenance_collector:,
         internal_metadata:,
         minimum_duration_seconds: PROFILE_DURATION_THRESHOLD_SECONDS,
-        time_provider: Time
+        time_provider: Time,
+        sequence_tracker: Datadog::Profiling::SequenceTracker
       )
         @pprof_recorder = pprof_recorder
         @worker = worker
@@ -49,7 +51,8 @@ module Datadog
         @internal_metadata = internal_metadata
         # NOTE: At the time of this comment collected info does not change over time so we'll hardcode
         #       it on startup to prevent serializing the same info on every flush.
-        @info_json = JSON.fast_generate(info_collector.info).freeze
+        @info_json = JSON.generate(info_collector.info).freeze
+        @sequence_tracker = sequence_tracker
       end
 
       def flush
@@ -57,7 +60,7 @@ module Datadog
         serialization_result = pprof_recorder.serialize
         return if serialization_result.nil?
 
-        start, finish, compressed_pprof, profile_stats = serialization_result
+        start, finish, encoded_profile, profile_stats = serialization_result
         @last_flush_finish_at = finish
 
         if duration_below_threshold?(start, finish)
@@ -70,11 +73,13 @@ module Datadog
         Flush.new(
           start: start,
           finish: finish,
-          pprof_file_name: Datadog::Profiling::Ext::Transport::HTTP::PPROF_DEFAULT_FILENAME,
-          pprof_data: compressed_pprof.to_s,
+          encoded_profile: encoded_profile,
           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,
+          tags_as_array: Datadog::Profiling::TagBuilder.call(
+            settings: Datadog.configuration,
+            profile_seq: sequence_tracker.get_next,
+          ).to_a,
           internal_metadata: internal_metadata.merge(
             {
               worker_stats: worker_stats,

data/lib/datadog/profiling/ext.rb

@@ -11,22 +11,8 @@ module Datadog
 
       module Transport
         module HTTP
-          FORM_FIELD_TAG_ENV = "env"
-          FORM_FIELD_TAG_HOST = "host"
-          FORM_FIELD_TAG_LANGUAGE = "language"
-          FORM_FIELD_TAG_PID = "process_id"
           FORM_FIELD_TAG_PROFILER_VERSION = "profiler_version"
-          FORM_FIELD_TAG_RUNTIME = "runtime"
-          FORM_FIELD_TAG_RUNTIME_ENGINE = "runtime_engine"
-          FORM_FIELD_TAG_RUNTIME_ID = "runtime-id"
-          FORM_FIELD_TAG_RUNTIME_PLATFORM = "runtime_platform"
-          FORM_FIELD_TAG_RUNTIME_VERSION = "runtime_version"
-          FORM_FIELD_TAG_SERVICE = "service"
-          FORM_FIELD_TAG_VERSION = "version"
-          TAG_GIT_REPOSITORY_URL = "git.repository_url"
-          TAG_GIT_COMMIT_SHA = "git.commit.sha"
 
-          PPROF_DEFAULT_FILENAME = "rubyprofile.pprof"
           CODE_PROVENANCE_FILENAME = "code-provenance.json"
         end
       end

data/lib/datadog/profiling/flush.rb

@@ -9,10 +9,9 @@ module Datadog
       attr_reader \
         :start,
         :finish,
-        :pprof_file_name,
-        :pprof_data, # gzipped pprof bytes
+        :encoded_profile,
         :code_provenance_file_name,
-        :code_provenance_data, # gzipped json bytes
+        :code_provenance_data,
         :tags_as_array,
         :internal_metadata_json,
         :info_json
@@ -20,8 +19,7 @@ module Datadog
       def initialize(
         start:,
         finish:,
-        pprof_file_name:,
-        pprof_data:,
+        encoded_profile:,
         code_provenance_file_name:,
         code_provenance_data:,
         tags_as_array:,
@@ -30,12 +28,11 @@ module Datadog
       )
         @start = start
         @finish = finish
-        @pprof_file_name = pprof_file_name
-        @pprof_data = pprof_data
+        @encoded_profile = encoded_profile
         @code_provenance_file_name = code_provenance_file_name
         @code_provenance_data = code_provenance_data
         @tags_as_array = tags_as_array
-        @internal_metadata_json = JSON.fast_generate(internal_metadata)
+        @internal_metadata_json = JSON.generate(internal_metadata)
         @info_json = info_json
       end
     end

data/lib/datadog/profiling/http_transport.rb

@@ -20,34 +20,16 @@ module Datadog
             [:agent, agent_settings.url].freeze
           end
 
-        status, result = validate_exporter(exporter_configuration)
+        status, result = self.class._native_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,
-          internal_metadata_json: flush.internal_metadata_json,
-
-          info_json: flush.info_json
+        status, result = self.class._native_do_export(
+          exporter_configuration,
+          @upload_timeout_milliseconds,
+          flush
         )
 
         if status == :ok
@@ -55,7 +37,7 @@ module Datadog
             Datadog.logger.debug("Successfully reported profiling data")
             true
           else
-            Datadog.logger.error(
+            Datadog.logger.warn(
               "Failed to report profiling data (#{config_without_api_key}): " \
               "server returned unexpected HTTP #{result} status code"
             )
@@ -65,7 +47,7 @@ module Datadog
             false
           end
         else
-          Datadog.logger.error("Failed to report profiling data (#{config_without_api_key}): #{result}")
+          Datadog.logger.warn("Failed to report profiling data (#{config_without_api_key}): #{result}")
           Datadog::Core::Telemetry::Logger.error("Failed to report profiling data")
           false
         end
@@ -77,42 +59,6 @@ module Datadog
         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:,
-        internal_metadata_json:,
-        info_json:
-      )
-        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,
-          internal_metadata_json,
-          info_json,
-        )
-      end
-
       def config_without_api_key
         "#{exporter_configuration[0]}: #{exporter_configuration[1]}"
       end

data/lib/datadog/profiling/profiler.rb

@@ -50,6 +50,8 @@ module Datadog
           "Detected issue with profiler (#{failed_component} component), stopping profiling. " \
           "See previous log messages for details."
         )
+        Datadog::Core::Telemetry::Logger
+          .error("Detected issue with profiler (#{failed_component} component), stopping profiling")
 
         # We explicitly not stop the crash tracker in this situation, under the assumption that, if a component failed,
         # we're operating in a degraded state and crash tracking may still be helpful.

data/lib/datadog/profiling/scheduler.rb

@@ -37,6 +37,7 @@ module Datadog
         @exporter = exporter
         @transport = transport
         @profiler_failed = false
+        @stop_requested = false
 
         # Workers::Async::Thread settings
         self.fork_policy = fork_policy
@@ -67,6 +68,7 @@ module Datadog
           "Cause: #{e.class.name} #{e.message} Location: #{Array(e.backtrace).first}"
         )
         on_failure_proc&.call
+        Datadog::Core::Telemetry::Logger.report(e, description: "Profiling::Scheduler thread error")
         raise
       ensure
         Datadog.logger.debug("#flush was interrupted or failed before it could complete") if interrupted
@@ -88,7 +90,7 @@ module Datadog
       end
 
       def work_pending?
-        !profiler_failed && exporter.can_flush?
+        !profiler_failed && exporter.can_flush? && (run_loop? || !stop_requested?)
       end
 
       def reset_after_fork
@@ -132,14 +134,20 @@ module Datadog
         begin
           transport.export(flush)
         rescue => e
-          Datadog.logger.error(
+          Datadog.logger.warn(
             "Unable to report profile. Cause: #{e.class.name} #{e.message} Location: #{Array(e.backtrace).first}"
           )
           Datadog::Core::Telemetry::Logger.report(e, description: "Unable to report profile")
         end
 
+        @stop_requested = !run_loop?
+
         true
       end
+
+      def stop_requested?
+        @stop_requested
+      end
     end
   end
 end

data/lib/datadog/profiling/sequence_tracker.rb

@@ -0,0 +1,44 @@
+# frozen_string_literal: true
+
+require_relative '../core/utils/forking'
+
+module Datadog
+  module Profiling
+    # Used to generate the `profile_seq` tag, which effectively counts how many profiles we've attempted to report
+    # from a given runtime-id.
+    #
+    # Note that the above implies a few things:
+    # 1. The sequence number only gets incremented when we decide to report a profile and create a `Flush` for it
+    # 2. The `SequenceTracker` must live across profiler reconfigurations and resets, since no matter how many
+    #    profiler instances get created due to reconfiguration, the runtime-id is still the same, so the sequence number
+    #    should be kept and not restarted from 0
+    # 3. The `SequenceTracker` must be reset after a fork, since the runtime-id will change, and we want to start
+    #    counting from 0 again
+    #
+    # This is why this module is implemented as a singleton that we reuse, not as an instance that we recreate.
+    #
+    # Note that this module is not thread-safe, so it's up to the callers to make sure
+    # it's only used by a single thread at a time (which is what the `Profiling::Exporter`)
+    # is doing.
+    module SequenceTracker
+      class << self
+        include Core::Utils::Forking
+
+        def get_next
+          reset! unless defined?(@sequence_number)
+          after_fork! { reset! }
+
+          next_seq = @sequence_number
+          @sequence_number += 1
+          next_seq
+        end
+
+        private
+
+        def reset!
+          @sequence_number = 0
+        end
+      end
+    end
+  end
+end

data/lib/datadog/profiling/stack_recorder.rb

@@ -23,7 +23,7 @@ module Datadog
         # This isn't something we expect to happen normally, but because it would break the assumptions of the
         # C-level mutexes (that there is a single serializer thread), we add it here as an extra safeguard against it
         # accidentally happening.
-        @no_concurrent_synchronize_mutex = Mutex.new
+        @no_concurrent_serialize_mutex = Mutex.new
 
         self.class._native_initialize(
           self_instance: self,
@@ -60,31 +60,31 @@ module Datadog
       end
 
       def serialize
-        status, result = @no_concurrent_synchronize_mutex.synchronize { self.class._native_serialize(self) }
+        status, result = @no_concurrent_serialize_mutex.synchronize { self.class._native_serialize(self) }
 
         if status == :ok
-          start, finish, encoded_pprof, profile_stats = result
+          start, finish, encoded_profile, profile_stats = result
 
           Datadog.logger.debug { "Encoded profile covering #{start.iso8601} to #{finish.iso8601}" }
 
-          [start, finish, encoded_pprof, profile_stats]
+          [start, finish, encoded_profile, profile_stats]
         else
           error_message = result
 
-          Datadog.logger.error("Failed to serialize profiling data: #{error_message}")
-          Datadog::Core::Telemetry::Logger.error("Failed to serialize profiling data")
+          Datadog.logger.warn("Failed to serialize profiling data: #{error_message}")
+          Datadog::Core::Telemetry::Logger.error("Failed to serialize profiling data (#{error_message})")
 
           nil
         end
       end
 
       def serialize!
-        status, result = @no_concurrent_synchronize_mutex.synchronize { self.class._native_serialize(self) }
+        status, result = @no_concurrent_serialize_mutex.synchronize { self.class._native_serialize(self) }
 
         if status == :ok
-          _start, _finish, encoded_pprof = result
+          _start, _finish, encoded_profile = result
 
-          encoded_pprof
+          encoded_profile
         else
           error_message = result
 

data/lib/datadog/profiling/tag_builder.rb

@@ -1,7 +1,7 @@
 # frozen_string_literal: true
 
+require_relative "../core/tag_builder"
 require_relative "../core/utils"
-require_relative "../core/environment/git"
 
 module Datadog
   module Profiling
@@ -11,49 +11,15 @@ module Datadog
 
       def self.call(
         settings:,
-        # Unified service tagging
-        env: settings.env,
-        service: settings.service,
-        version: settings.version,
         # Other metadata
-        host: Core::Environment::Socket.hostname,
-        language: Core::Environment::Identity.lang,
-        pid: Process.pid.to_s,
-        # TODO: If profiling is extracted and its version diverges from the datadog gem, this is inaccurate.
-        #       Update if this ever occurs.
-        profiler_version: Core::Environment::Identity.gem_datadog_version,
-        runtime_engine: Core::Environment::Identity.lang_engine,
-        runtime_id: Core::Environment::Identity.id,
-        runtime_platform: Core::Environment::Identity.lang_platform,
-        runtime_version: Core::Environment::Identity.lang_version,
-        git_repository_url: Core::Environment::Git.git_repository_url,
-        git_commit_sha: Core::Environment::Git.git_commit_sha,
-        # User-provided tags
-        user_tags: settings.tags
+        profile_seq:,
+        profiler_version: Core::Environment::Identity.gem_datadog_version
       )
-        tags = {
-          # When changing or adding these, make sure they are kept in sync with
-          # https://docs.google.com/spreadsheets/d/1LOGMf4c4Avbtn36uZ2SWvhIGKRPLM1BoWkUP4JYj7hA/ (Datadog internal link)
-          FORM_FIELD_TAG_HOST => host,
-          FORM_FIELD_TAG_LANGUAGE => language,
-          FORM_FIELD_TAG_PID => pid,
+        hash = Core::TagBuilder.tags(settings).merge(
           FORM_FIELD_TAG_PROFILER_VERSION => profiler_version,
-          FORM_FIELD_TAG_RUNTIME => language, # This is known to be repeated from language, above
-          FORM_FIELD_TAG_RUNTIME_ENGINE => runtime_engine,
-          FORM_FIELD_TAG_RUNTIME_ID => runtime_id,
-          FORM_FIELD_TAG_RUNTIME_PLATFORM => runtime_platform,
-          FORM_FIELD_TAG_RUNTIME_VERSION => runtime_version,
-        }
-        tags[FORM_FIELD_TAG_ENV] = env if env
-        tags[FORM_FIELD_TAG_SERVICE] = service if service
-        tags[FORM_FIELD_TAG_VERSION] = version if version
-        tags[TAG_GIT_REPOSITORY_URL] = git_repository_url if git_repository_url
-        tags[TAG_GIT_COMMIT_SHA] = git_commit_sha if git_commit_sha
-
-        # Make sure everything is an utf-8 string, to avoid encoding issues in native code/libddprof/further downstream
-        user_tags.merge(tags).map do |key, value|
-          [Datadog::Core::Utils.utf8_encode(key), Datadog::Core::Utils.utf8_encode(value)]
-        end.to_h
+          'profile_seq' => profile_seq.to_s,
+        )
+        Core::Utils.encode_tags(hash)
       end
     end
   end

data/lib/datadog/profiling/tasks/setup.rb

@@ -19,6 +19,7 @@ module Datadog
               "Profiler extensions unavailable. Cause: #{e.class.name} #{e.message} " \
               "Location: #{Array(e.backtrace).first}"
             end
+            Datadog::Core::Telemetry::Logger.report(e, description: "Profiler extensions unavailable")
           end
         end
 
@@ -33,6 +34,7 @@ module Datadog
               "Error during post-fork hooks. Cause: #{e.class.name} #{e.message} " \
               "Location: #{Array(e.backtrace).first}"
             end
+            Datadog::Core::Telemetry::Logger.report(e, description: "Error during post-fork hooks")
           end
         end
       end

data/lib/datadog/profiling.rb

@@ -78,8 +78,11 @@ module Datadog
     end
 
     private_class_method def self.replace_noop_allocation_count
-      def self.allocation_count # rubocop:disable Lint/NestedMethodDefinition (On purpose!)
-        Datadog::Profiling::Collectors::CpuAndWallTimeWorker._native_allocation_count
+      class << self
+        remove_method :allocation_count
+        def allocation_count
+          Datadog::Profiling::Collectors::CpuAndWallTimeWorker._native_allocation_count
+        end
       end
     end
 
@@ -146,6 +149,7 @@ module Datadog
       require_relative 'profiling/collectors/thread_context'
       require_relative 'profiling/stack_recorder'
       require_relative 'profiling/exporter'
+      require_relative 'profiling/encoded_profile'
       require_relative 'profiling/flush'
       require_relative 'profiling/scheduler'
       require_relative 'profiling/tasks/setup'
@@ -153,6 +157,7 @@ module Datadog
       require_relative 'profiling/native_extension'
       require_relative 'profiling/tag_builder'
       require_relative 'profiling/http_transport'
+      require_relative 'profiling/sequence_tracker'
 
       replace_noop_allocation_count
 

data/lib/datadog/single_step_instrument.rb

@@ -5,8 +5,17 @@
 #
 # This file's path is private. Do not reference this file.
 #
+
+module Datadog
+  # This module handles conditional loading of single step auto-instrumentation,
+  # which enables Datadog tracing and profiling features when available.
+  module SingleStepInstrument
+  end
+end
+
 begin
   require_relative 'auto_instrument'
+  Datadog::SingleStepInstrument::LOADED = true
 rescue StandardError, LoadError => e
   warn "Single step instrumentation failed: #{e.class}:#{e.message}\n\tSource:\n\t#{Array(e.backtrace).join("\n\t")}"
 end

data/lib/datadog/tracing/analytics.rb

@@ -16,7 +16,7 @@ module Datadog
         def set_measured(span_op, value = true)
           return if span_op.nil?
 
-          value = value == true || value == 1 ? 1 : 0 # rubocop:disable Style/MultipleComparison
+          value = value == true || value == 1 ? 1 : 0
           span_op.set_metric(Metadata::Ext::Analytics::TAG_MEASURED, value)
         end
       end