ddtrace 1.2.0 → 1.10.1

data/lib/datadog/profiling/encoding/profile.rb

@@ -1,8 +1,6 @@
-# typed: true
-
 require 'time'
 
-require 'datadog/profiling/pprof/template'
+require_relative '../pprof/template'
 
 module Datadog
   module Profiling

data/lib/datadog/profiling/event.rb

@@ -1,5 +1,3 @@
-# typed: true
-
 module Datadog
   module Profiling
     # Describes a sample of some data obtained from the runtime.

data/lib/datadog/profiling/events/stack.rb

@@ -1,6 +1,4 @@
-# typed: true
-
-require 'datadog/profiling/event'
+require_relative '../event'
 
 module Datadog
   module Profiling

data/lib/datadog/profiling/exporter.rb

@@ -1,8 +1,5 @@
-# typed: true
-
-require 'datadog/profiling/ext'
-require 'datadog/core/utils/compression'
-require 'datadog/profiling/tag_builder'
+require_relative 'ext'
+require_relative 'tag_builder'
 
 module Datadog
   module Profiling
@@ -22,22 +19,30 @@ module Datadog
       attr_reader \
         :pprof_recorder,
         :code_provenance_collector, # The code provenance collector acts both as collector and as a recorder
-        :minimum_duration
+        :minimum_duration_seconds,
+        :time_provider,
+        :last_flush_finish_at,
+        :created_at
 
       public
 
       def initialize(
         pprof_recorder:,
         code_provenance_collector:,
-        minimum_duration: PROFILE_DURATION_THRESHOLD_SECONDS
+        minimum_duration_seconds: PROFILE_DURATION_THRESHOLD_SECONDS,
+        time_provider: Time
       )
         @pprof_recorder = pprof_recorder
         @code_provenance_collector = code_provenance_collector
-        @minimum_duration = minimum_duration
+        @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
 
@@ -52,22 +57,26 @@ module Datadog
           start: start,
           finish: finish,
           pprof_file_name: Datadog::Profiling::Ext::Transport::HTTP::PPROF_DEFAULT_FILENAME,
-          pprof_data: Datadog::Core::Utils::Compression.gzip(uncompressed_pprof),
+          pprof_data: uncompressed_pprof.to_s,
           code_provenance_file_name: Datadog::Profiling::Ext::Transport::HTTP::CODE_PROVENANCE_FILENAME,
-          code_provenance_data:
-            (Datadog::Core::Utils::Compression.gzip(uncompressed_code_provenance) if uncompressed_code_provenance),
+          code_provenance_data: uncompressed_code_provenance,
           tags_as_array: Datadog::Profiling::TagBuilder.call(settings: Datadog.configuration).to_a,
         )
       end
 
-      def empty?
-        pprof_recorder.empty?
+      def can_flush?
+        !duration_below_threshold?(last_flush_finish_at || created_at, time_provider.now.utc)
+      end
+
+      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
+        (finish - start) < minimum_duration_seconds
       end
     end
   end

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

@@ -1,10 +1,13 @@
-# typed: false
-
 module Datadog
   module Profiling
     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)
@@ -13,33 +16,22 @@ 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?
@@ -52,9 +44,6 @@ module Datadog
                             end
                           end
 
-            # Trigger :prepare callback
-            ddtrace_at_fork_blocks[:prepare].each(&:call) if ddtrace_at_fork_blocks.key?(:prepare)
-
             # Start fork
             # If a block is provided, use the wrapped version.
             result = child_block.nil? ? super : super(&child_block)
@@ -62,22 +51,14 @@ 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
-              ddtrace_at_fork_blocks[:child].each(&:call) if ddtrace_at_fork_blocks.key?(:child)
-            else
-              # Trigger :parent callback
-              ddtrace_at_fork_blocks[:parent].each(&:call) if ddtrace_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
 
             ddtrace_at_fork_blocks[stage] = [] unless ddtrace_at_fork_blocks.key?(stage)
             ddtrace_at_fork_blocks[stage] << block
@@ -93,6 +74,22 @@ module Datadog
             # 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

@@ -1,5 +1,3 @@
-# typed: true
-
 module Datadog
   module Profiling
     module Ext
@@ -9,6 +7,7 @@ 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

data/lib/datadog/profiling/flush.rb

@@ -1,8 +1,3 @@
-# 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.

data/lib/datadog/profiling/http_transport.rb

@@ -1,5 +1,3 @@
-# typed: false
-
 module Datadog
   module Profiling
     # Used to report profiling data to Datadog.
@@ -28,7 +26,7 @@ module Datadog
           upload_timeout_milliseconds: @upload_timeout_milliseconds,
 
           # why "timespec"?
-          # libddprof represents time using POSIX's struct timespec, see
+          # 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,
@@ -58,7 +56,7 @@ module Datadog
         end
       end
 
-      # Used to log soft failures in `ddprof_ffi_Vec_tag_push` (e.g. we still report the profile in these cases)
+      # 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}")
@@ -80,8 +78,9 @@ module Datadog
       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'
+          raise ArgumentError,
+            "Unsupported transport configuration for profiling: Adapter #{agent_settings.adapter} " \
+                        ' is not supported'
         end
 
         if agent_settings.deprecated_for_removal_transport_configuration_proc

data/lib/datadog/profiling/load_native_extension.rb

@@ -1,5 +1,3 @@
-# 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

data/lib/datadog/profiling/native_extension.rb

@@ -1,5 +1,3 @@
-# typed: true
-
 module Datadog
   module Profiling
     # This module contains classes and methods which are implemented using native code in the

data/lib/datadog/profiling/old_recorder.rb

@@ -1,7 +1,5 @@
-# typed: true
-
-require 'datadog/profiling/buffer'
-require 'datadog/profiling/encoding/profile'
+require_relative 'buffer'
+require_relative 'encoding/profile'
 
 module Datadog
   module Profiling
@@ -73,9 +71,12 @@ module Datadog
         [start, finish, encoded_pprof]
       end
 
-      # NOTE: Remember that if the recorder is being accessed by multiple threads, this is an inherently racy operation.
-      def empty?
-        @buffers.values.all?(&:empty?)
+      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

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

@@ -1,11 +1,9 @@
 # frozen_string_literal: true
 
-# 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 +1,4 @@
-# typed: true
-
-require 'datadog/profiling/ext'
+require_relative '../ext'
 
 module Datadog
   module Profiling

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

@@ -1,6 +1,4 @@
-# typed: true
-
-require 'datadog/core/utils/object_set'
+require_relative '../../core/utils/object_set'
 
 module Datadog
   module Profiling

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

@@ -1,5 +1,3 @@
-# typed: true
-
 module Datadog
   module Profiling
     module Pprof

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

@@ -1,5 +1,3 @@
-# typed: false
-
 # Generated by the protocol buffer compiler.  DO NOT EDIT!
 # source: lib/datadog/profiling/pprof/pprof.proto
 

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

@@ -1,9 +1,7 @@
-# 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

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

@@ -1,6 +1,4 @@
-# typed: strict
-
-require 'datadog/core/utils/string_table'
+require_relative '../../core/utils/string_table'
 
 module Datadog
   module Profiling

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

@@ -1,11 +1,9 @@
-# typed: false
+require_relative 'payload'
+require_relative 'message_set'
+require_relative 'builder'
 
-require 'datadog/profiling/pprof/payload'
-require 'datadog/profiling/pprof/message_set'
-require 'datadog/profiling/pprof/builder'
-
-require 'datadog/profiling/events/stack'
-require 'datadog/profiling/pprof/stack_sample'
+require_relative '../events/stack'
+require_relative 'stack_sample'
 
 module Datadog
   module Profiling

data/lib/datadog/profiling/preload.rb

@@ -1,5 +1,3 @@
-# typed: strict
-
-require 'ddtrace'
+require_relative '../../ddtrace'
 
 Datadog::Profiling.start_if_enabled

data/lib/datadog/profiling/profiler.rb

@@ -1,9 +1,9 @@
-# typed: true
-
 module Datadog
   module Profiling
     # Profiling entry point, which coordinates collectors and a scheduler
     class Profiler
+      include Datadog::Core::Utils::Forking
+
       attr_reader \
         :collectors,
         :scheduler
@@ -14,6 +14,11 @@ module Datadog
       end
 
       def start
+        after_fork! do
+          collectors.each(&:reset_after_fork)
+          scheduler.reset_after_fork
+        end
+
         collectors.each(&:start)
         scheduler.start
       end

data/lib/datadog/profiling/scheduler.rb

@@ -1,9 +1,7 @@
-# typed: true
+require_relative '../core/utils/time'
 
-require 'datadog/core/utils/time'
-
-require 'datadog/core/worker'
-require 'datadog/core/workers/polling'
+require_relative '../core/worker'
+require_relative '../core/workers/polling'
 
 module Datadog
   module Profiling
@@ -66,12 +64,6 @@ module Datadog
         end
       end
 
-      def after_fork
-        # Clear any existing profiling state.
-        # We don't want the child process to report profiling data from its parent.
-        exporter.flush
-      end
-
       # Configure Workers::IntervalLoop to not report immediately when scheduler starts
       #
       # When a scheduler gets created (or reset), we don't want it to immediately try to flush; we want it to wait for
@@ -82,7 +74,11 @@ module Datadog
       end
 
       def work_pending?
-        !exporter.empty?
+        exporter.can_flush?
+      end
+
+      def reset_after_fork
+        exporter.reset_after_fork
       end
 
       private

data/lib/datadog/profiling/stack_recorder.rb

@@ -1,12 +1,23 @@
-# typed: false
-
 module Datadog
   module Profiling
-    # Used to wrap a ddprof_ffi_Profile in a Ruby object and expose Ruby-level serialization APIs
+    # Stores stack samples in a native libdatadog data structure and expose Ruby-level serialization APIs
+    # Note that `record_sample` is only accessible from native code.
     # Methods prefixed with _native_ are implemented in `stack_recorder.c`
     class StackRecorder
+      def initialize(cpu_time_enabled:, alloc_samples_enabled:)
+        # This mutex works in addition to the fancy C-level mutexes we have in the native side (see the docs there).
+        # It prevents multiple Ruby threads calling serialize at the same time -- something like
+        # `10.times { Thread.new { stack_recorder.serialize } }`.
+        # 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
+
+        self.class._native_initialize(self, cpu_time_enabled, alloc_samples_enabled)
+      end
+
       def serialize
-        status, result = self.class._native_serialize(self)
+        status, result = @no_concurrent_synchronize_mutex.synchronize { self.class._native_serialize(self) }
 
         if status == :ok
           start, finish, encoded_pprof = result
@@ -23,10 +34,40 @@ module Datadog
         end
       end
 
-      # Used only for Ruby 2.2 and below which don't have the native `rb_time_timespec_new` API
-      # Called from native code
-      def self.ruby_time_from(timespec_seconds, timespec_nanoseconds)
-        Time.at(0).utc + timespec_seconds + (timespec_nanoseconds.to_r / 1_000_000_000)
+      def serialize!
+        status, result = @no_concurrent_synchronize_mutex.synchronize { self.class._native_serialize(self) }
+
+        if status == :ok
+          _start, _finish, encoded_pprof = result
+
+          encoded_pprof
+        else
+          error_message = result
+
+          raise("Failed to serialize profiling data: #{error_message}")
+        end
+      end
+
+      def clear
+        status, result = @no_concurrent_synchronize_mutex.synchronize { self.class._native_clear(self) }
+
+        if status == :ok
+          finish_timestamp = result
+
+          Datadog.logger.debug { "Cleared profile at #{finish_timestamp}" }
+
+          finish_timestamp
+        else
+          error_message = result
+
+          Datadog.logger.error("Failed to clear profiling data: #{error_message}")
+
+          nil
+        end
+      end
+
+      def reset_after_fork
+        self.class._native_reset_after_fork(self)
       end
     end
   end

data/lib/datadog/profiling/tag_builder.rb

@@ -1,4 +1,4 @@
-# typed: true
+require_relative '../core/utils'
 
 module Datadog
   module Profiling
@@ -41,7 +41,10 @@ module Datadog
         tags[FORM_FIELD_TAG_SERVICE] = service if service
         tags[FORM_FIELD_TAG_VERSION] = version if version
 
-        user_tags.merge(tags)
+        # 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
       end
     end
   end

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

@@ -1,5 +1,3 @@
-# typed: true
-
 module Datadog
   module Profiling
     module Tasks

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

@@ -1,5 +1,3 @@
-# typed: true
-
 module Datadog
   module Profiling
     module Tasks

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

@@ -1,7 +1,5 @@
-# typed: false
-
-require 'datadog/core/utils/only_once'
-require 'datadog/profiling/ext/forking'
+require_relative '../../core/utils/only_once'
+require_relative '../ext/forking'
 
 module Datadog
   module Profiling
@@ -55,13 +53,6 @@ module Datadog
           if Process.respond_to?(:at_fork)
             Process.at_fork(:child) do
               begin
-                # When Ruby forks, clock IDs for each of the threads
-                # will change. We can only update these IDs from the
-                # execution context of the thread that owns it.
-                # This hook will update the IDs for the main thread
-                # after a fork occurs.
-                Thread.current.send(:update_native_ids) if Thread.current.respond_to?(:update_native_ids, true)
-
                 # Restart profiler, if enabled
                 Profiling.start_if_enabled
               rescue StandardError => e

data/lib/datadog/profiling/trace_identifiers/ddtrace.rb

@@ -1,9 +1,7 @@
 # frozen_string_literal: true
 
-# typed: true
-
-require 'datadog/tracing'
-require 'datadog/tracing/metadata/ext'
+require_relative '../../tracing'
+require_relative '../../tracing/metadata/ext'
 
 module Datadog
   module Profiling

data/lib/datadog/profiling/trace_identifiers/helper.rb

@@ -1,8 +1,6 @@
 # frozen_string_literal: true
 
-# typed: true
-
-require 'datadog/profiling/trace_identifiers/ddtrace'
+require_relative 'ddtrace'
 
 module Datadog
   module Profiling