ddtrace 1.9.0 → 1.10.0

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

@@ -1,7 +1,5 @@
 # frozen_string_literal: true
 
-# typed: false
-
 require 'set'
 require 'json'
 

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

@@ -1,10 +1,8 @@
-# typed: false
-
 module Datadog
   module Profiling
     module Collectors
-      # Used to trigger the periodic execution of Collectors::CpuAndWallTime, which implements all of the sampling logic
-      # itself; this class only implements the "doing it periodically" part.
+      # Used to trigger the periodic execution of Collectors::ThreadState, which implements all of the sampling logic
+      # itself; this class only implements the "when to do it" part.
       # Almost all of this class is implemented as native code.
       #
       # Methods prefixed with _native_ are implemented in `collectors_cpu_and_wall_time_worker.c`
@@ -20,10 +18,17 @@ module Datadog
           max_frames:,
           tracer:,
           gc_profiling_enabled:,
-          cpu_and_wall_time_collector: CpuAndWallTime.new(recorder: recorder, max_frames: max_frames, tracer: tracer),
+          allocation_counting_enabled:,
+          thread_context_collector: ThreadContext.new(recorder: recorder, max_frames: max_frames, tracer: tracer),
           idle_sampling_helper: IdleSamplingHelper.new
         )
-          self.class._native_initialize(self, cpu_and_wall_time_collector, gc_profiling_enabled, idle_sampling_helper)
+          self.class._native_initialize(
+            self,
+            thread_context_collector,
+            gc_profiling_enabled,
+            idle_sampling_helper,
+            allocation_counting_enabled
+          )
           @worker_thread = nil
           @failure_exception = nil
           @start_stop_mutex = Mutex.new
@@ -40,7 +45,7 @@ module Datadog
 
             @worker_thread = Thread.new do
               begin
-                Thread.current.name = self.class.name unless Gem::Version.new(RUBY_VERSION) < Gem::Version.new('2.3')
+                Thread.current.name = self.class.name
 
                 self.class._native_sampling_loop(self)
 

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

@@ -1,5 +1,3 @@
-# typed: false
-
 module Datadog
   module Profiling
     module Collectors

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

@@ -1,5 +1,3 @@
-# typed: false
-
 module Datadog
   module Profiling
     module Collectors
@@ -31,7 +29,7 @@ module Datadog
 
             @worker_thread = Thread.new do
               begin
-                Thread.current.name = self.class.name unless Gem::Version.new(RUBY_VERSION) < Gem::Version.new('2.3')
+                Thread.current.name = self.class.name
 
                 self.class._native_idle_sampling_loop(self)
 

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

@@ -1,5 +1,3 @@
-# typed: true
-
 require_relative '../../core/utils/only_once'
 require_relative '../../core/utils/time'
 require_relative '../../core/worker'
@@ -75,9 +73,7 @@ module Datadog
           # Cache this buffer, since it's pretty expensive to keep accessing it
           @stack_sample_event_recorder = recorder[Events::StackSample]
           # See below for details on why this is needed
-          @needs_process_waiter_workaround =
-            Gem::Version.new(RUBY_VERSION) >= Gem::Version.new('2.3') &&
-            Gem::Version.new(RUBY_VERSION) < Gem::Version.new('2.7')
+          @needs_process_waiter_workaround = Gem::Version.new(RUBY_VERSION) < Gem::Version.new('2.7')
         end
 
         def start

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

@@ -1,5 +1,3 @@
-# typed: false
-
 module Datadog
   module Profiling
     module Collectors

data/lib/datadog/profiling/collectors/{cpu_and_wall_time.rb → thread_context.rb}

@@ -1,16 +1,19 @@
-# typed: false
+# frozen_string_literal: true
 
 module Datadog
   module Profiling
     module Collectors
-      # Used to periodically sample threads, recording elapsed CPU-time and Wall-time between samples.
-      # Triggering of this component (e.g. deciding when to take a sample) is implemented in
+      # Used to trigger sampling of threads, based on external "events", such as:
+      # * periodic timer for cpu-time and wall-time
+      # * VM garbage collection events
+      # * VM object allocation events
+      # Triggering of this component (e.g. watching for the above "events") is implemented by
       # Collectors::CpuAndWallTimeWorker.
       # The stack collection itself is handled using the Datadog::Profiling::Collectors::Stack.
       # Almost all of this class is implemented as native code.
       #
-      # Methods prefixed with _native_ are implemented in `collectors_cpu_and_wall_time.c`
-      class CpuAndWallTime
+      # Methods prefixed with _native_ are implemented in `collectors_thread_context.c`
+      class ThreadContext
         def initialize(recorder:, max_frames:, tracer:)
           tracer_context_key = safely_extract_context_key_from(tracer)
           self.class._native_initialize(self, recorder, max_frames, tracer_context_key)

data/lib/datadog/profiling/component.rb

@@ -0,0 +1,161 @@
+# frozen_string_literal: true
+
+module Datadog
+  module Profiling
+    # Profiling component
+    module Component
+      def build_profiler(settings, agent_settings, tracer)
+        return unless settings.profiling.enabled
+
+        # Workaround for weird dependency direction: the Core::Configuration::Components class currently has a
+        # dependency on individual products, in this case the Profiler.
+        # (Note "currently": in the future we want to change this so core classes don't depend on specific products)
+        #
+        # If the current file included a `require 'datadog/profiler'` at its beginning, we would generate circular
+        # requires when used from profiling:
+        #
+        # datadog/profiling
+        #     └─requires─> datadog/core
+        #                      └─requires─> datadog/core/configuration/components
+        #                                       └─requires─> datadog/profiling       # Loop!
+        #
+        # ...thus in #1998 we removed such a require.
+        #
+        # On the other hand, if datadog/core is loaded by a different product and no general `require 'ddtrace'` is
+        # done, then profiling may not be loaded, and thus to avoid this issue we do a require here (which is a
+        # no-op if profiling is already loaded).
+        require_relative '../profiling'
+        return unless Profiling.supported?
+
+        unless defined?(Profiling::Tasks::Setup)
+          # In #1545 a user reported a NameError due to this constant being uninitialized
+          # I've documented my suspicion on why that happened in
+          # https://github.com/DataDog/dd-trace-rb/issues/1545#issuecomment-856049025
+          #
+          # > Thanks for the info! It seems to feed into my theory: there's two moments in the code where we check if
+          # > profiler is "supported": 1) when loading ddtrace (inside preload) and 2) when starting the profile
+          # > after Datadog.configure gets run.
+          # > The problem is that the code assumes that both checks 1) and 2) will always reach the same conclusion:
+          # > either profiler is supported, or profiler is not supported.
+          # > In the problematic case, it looks like in your case check 1 decides that profiler is not
+          # > supported => doesn't load it, and then check 2 decides that it is => assumes it is loaded and tries to
+          # > start it.
+          #
+          # I was never able to validate if this was the issue or why exactly .supported? would change its mind BUT
+          # just in case it happens again, I've left this check which avoids breaking the user's application AND
+          # would instead direct them to report it to us instead, so that we can investigate what's wrong.
+          #
+          # TODO: As of June 2021, most checks in .supported? are related to the google-protobuf gem; so it's
+          # very likely that it was the origin of the issue we saw. Thus, if, as planned we end up moving away from
+          # protobuf OR enough time has passed and no users saw the issue again, we can remove this check altogether.
+          Datadog.logger.error(
+            'Profiling was marked as supported and enabled, but setup task was not loaded properly. ' \
+            'Please report this at https://github.com/DataDog/dd-trace-rb/blob/master/CONTRIBUTING.md#found-a-bug'
+          )
+
+          return
+        end
+
+        # Load extensions needed to support some of the Profiling features
+        Profiling::Tasks::Setup.new.run
+
+        # NOTE: Please update the Initialization section of ProfilingDevelopment.md with any changes to this method
+
+        if settings.profiling.advanced.force_enable_new_profiler
+          print_new_profiler_warnings
+
+          recorder = Datadog::Profiling::StackRecorder.new(
+            cpu_time_enabled: RUBY_PLATFORM.include?('linux'), # Only supported on Linux currently
+            alloc_samples_enabled: false, # Always disabled for now -- work in progress
+          )
+          collector = Datadog::Profiling::Collectors::CpuAndWallTimeWorker.new(
+            recorder: recorder,
+            max_frames: settings.profiling.advanced.max_frames,
+            tracer: tracer,
+            gc_profiling_enabled: should_enable_gc_profiling?(settings),
+            allocation_counting_enabled: settings.profiling.advanced.allocation_counting_enabled,
+          )
+        else
+          trace_identifiers_helper = Profiling::TraceIdentifiers::Helper.new(
+            tracer: tracer,
+            endpoint_collection_enabled: settings.profiling.advanced.endpoint.collection.enabled
+          )
+
+          recorder = build_profiler_old_recorder(settings)
+          collector = build_profiler_oldstack_collector(settings, recorder, trace_identifiers_helper)
+        end
+
+        exporter = build_profiler_exporter(settings, recorder)
+        transport = build_profiler_transport(settings, agent_settings)
+        scheduler = Profiling::Scheduler.new(exporter: exporter, transport: transport)
+
+        Profiling::Profiler.new([collector], scheduler)
+      end
+
+      private
+
+      def build_profiler_old_recorder(settings)
+        Profiling::OldRecorder.new([Profiling::Events::StackSample], settings.profiling.advanced.max_events)
+      end
+
+      def build_profiler_exporter(settings, recorder)
+        code_provenance_collector =
+          (Profiling::Collectors::CodeProvenance.new if settings.profiling.advanced.code_provenance_enabled)
+
+        Profiling::Exporter.new(pprof_recorder: recorder, code_provenance_collector: code_provenance_collector)
+      end
+
+      def build_profiler_oldstack_collector(settings, old_recorder, trace_identifiers_helper)
+        Profiling::Collectors::OldStack.new(
+          old_recorder,
+          trace_identifiers_helper: trace_identifiers_helper,
+          max_frames: settings.profiling.advanced.max_frames
+        )
+      end
+
+      def build_profiler_transport(settings, agent_settings)
+        settings.profiling.exporter.transport ||
+          Profiling::HttpTransport.new(
+            agent_settings: agent_settings,
+            site: settings.site,
+            api_key: settings.api_key,
+            upload_timeout_seconds: settings.profiling.upload.timeout_seconds,
+          )
+      end
+
+      def should_enable_gc_profiling?(settings)
+        # See comments on the setting definition for more context on why it exists.
+        if settings.profiling.advanced.force_enable_gc_profiling
+          if Gem::Version.new(RUBY_VERSION) >= Gem::Version.new('3')
+            Datadog.logger.debug(
+              'Profiling time/resources spent in Garbage Collection force enabled. Do not use Ractors in combination ' \
+              'with this option as profiles will be incomplete.'
+            )
+          end
+
+          true
+        else
+          false
+        end
+      end
+
+      def print_new_profiler_warnings
+        if Gem::Version.new(RUBY_VERSION) >= Gem::Version.new('2.6')
+          Datadog.logger.warn(
+            'New Ruby profiler has been force-enabled. This is a beta feature. Please report any issues ' \
+            'you run into to Datadog support or via <https://github.com/datadog/dd-trace-rb/issues/new>!'
+          )
+        else
+          # For more details on the issue, see the "BIG Issue" comment on `gvl_owner` function in
+          # `private_vm_api_access.c`.
+          Datadog.logger.warn(
+            'New Ruby profiler has been force-enabled on a legacy Ruby version (< 2.6). This is not recommended in ' \
+            'production environments, as due to limitations in Ruby APIs, we suspect it may lead to crashes in very ' \
+            'rare situations. Please report any issues you run into to Datadog support or ' \
+            'via <https://github.com/datadog/dd-trace-rb/issues/new>!'
+          )
+        end
+      end
+    end
+  end
+end

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

@@ -1,5 +1,3 @@
-# typed: true
-
 require 'time'
 
 require_relative '../pprof/template'

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

@@ -1,5 +1,3 @@
-# typed: true
-
 require_relative '../event'
 
 module Datadog

data/lib/datadog/profiling/exporter.rb

@@ -1,5 +1,3 @@
-# typed: true
-
 require_relative 'ext'
 require_relative 'tag_builder'
 

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

@@ -1,5 +1,3 @@
-# typed: false
-
 module Datadog
   module Profiling
     module Ext

data/lib/datadog/profiling/ext.rb

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

data/lib/datadog/profiling/flush.rb

@@ -1,5 +1,3 @@
-# typed: false
-
 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.

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

@@ -1,5 +1,3 @@
-# typed: true
-
 require_relative 'buffer'
 require_relative 'encoding/profile'
 

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

@@ -1,7 +1,5 @@
 # frozen_string_literal: true
 
-# typed: true
-
 require_relative '../flush'
 require_relative 'message_set'
 require_relative 'string_table'

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

@@ -1,5 +1,3 @@
-# typed: true
-
 require_relative '../ext'
 
 module Datadog

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

@@ -1,5 +1,3 @@
-# typed: true
-
 require_relative '../../core/utils/object_set'
 
 module Datadog

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

@@ -1,5 +1,3 @@
-# typed: true
-
 require_relative '../ext'
 require_relative '../events/stack'
 require_relative 'builder'

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

@@ -1,5 +1,3 @@
-# typed: strict
-
 require_relative '../../core/utils/string_table'
 
 module Datadog

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

@@ -1,5 +1,3 @@
-# typed: false
-
 require_relative 'payload'
 require_relative 'message_set'
 require_relative 'builder'

data/lib/datadog/profiling/preload.rb

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

data/lib/datadog/profiling/profiler.rb

@@ -1,5 +1,3 @@
-# typed: true
-
 module Datadog
   module Profiling
     # Profiling entry point, which coordinates collectors and a scheduler

data/lib/datadog/profiling/scheduler.rb

@@ -1,5 +1,3 @@
-# typed: true
-
 require_relative '../core/utils/time'
 
 require_relative '../core/worker'

data/lib/datadog/profiling/stack_recorder.rb

@@ -1,12 +1,10 @@
-# typed: false
-
 module Datadog
   module Profiling
     # 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
+      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 } }`.
@@ -14,6 +12,8 @@ module Datadog
         # 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
@@ -66,11 +66,6 @@ module Datadog
         end
       end
 
-      # Used only for Ruby 2.2 which doesn'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)
-      end
-
       def reset_after_fork
         self.class._native_reset_after_fork(self)
       end

data/lib/datadog/profiling/tag_builder.rb

@@ -1,5 +1,3 @@
-# typed: true
-
 require_relative '../core/utils'
 
 module Datadog

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

@@ -1,5 +1,3 @@
-# typed: false
-
 require_relative '../../core/utils/only_once'
 require_relative '../ext/forking'
 

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

@@ -1,7 +1,5 @@
 # frozen_string_literal: true
 
-# typed: true
-
 require_relative '../../tracing'
 require_relative '../../tracing/metadata/ext'
 

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

@@ -1,7 +1,5 @@
 # frozen_string_literal: true
 
-# typed: true
-
 require_relative 'ddtrace'
 
 module Datadog

data/lib/datadog/profiling.rb

@@ -1,11 +1,9 @@
-# typed: true
-
 require_relative 'core'
 require_relative 'core/environment/variable_helpers'
 require_relative 'core/utils/only_once'
 
 module Datadog
-  # Contains profiler for generating stack profiles, etc.
+  # Datadog Continuous Profiler implementation: https://docs.datadoghq.com/profiler/
   module Profiling
     GOOGLE_PROTOBUF_MINIMUM_VERSION = Gem::Version.new('3.0')
     private_constant :GOOGLE_PROTOBUF_MINIMUM_VERSION
@@ -41,6 +39,39 @@ module Datadog
       !!profiler
     end
 
+    # Returns an ever-increasing counter of the number of allocations observed by the profiler in this thread.
+    #
+    # Note 1: This counter may not start from zero on new threads. It should only be used to measure how many
+    # allocations have happened between two calls to this API:
+    # ```
+    # allocations_before = Datadog::Profiling.allocation_count
+    # do_some_work()
+    # allocations_after = Datadog::Profiling.allocation_count
+    # puts "Allocations during do_some_work: #{allocations_after - allocations_before}"
+    # ```
+    # (This is similar to some OS-based time representations.)
+    #
+    # Note 2: All fibers in the same thread will share the same counter values.
+    #
+    # Only available when the profiler is running, the new CPU Profiling 2.0 profiler is in use, and allocation-related
+    # features are not disabled via configuration.
+    # For instructions on enabling CPU Profiling 2.0 see the ddtrace release notes.
+    #
+    # @return [Integer] number of allocations observed in the current thread.
+    # @return [nil] when not available.
+    # @public_api
+    def self.allocation_count
+      # This no-op implementation is used when profiling failed to load.
+      # It gets replaced inside #replace_noop_allocation_count.
+      nil
+    end
+
+    private_class_method def self.replace_noop_allocation_count
+      def self.allocation_count # rubocop:disable Lint/DuplicateMethods, Lint/NestedMethodDefinition (On purpose!)
+        Datadog::Profiling::Collectors::CpuAndWallTimeWorker._native_allocation_count
+      end
+    end
+
     private_class_method def self.native_library_compilation_skipped?
       skipped_reason = try_reading_skipped_reason_file
 
@@ -65,7 +96,7 @@ module Datadog
       # NOTE: On environments where protobuf is already loaded, we skip the check. This allows us to support environments
       # where no Gem.loaded_version is NOT available but customers are able to load protobuf; see for instance
       # https://github.com/teamcapybara/capybara/commit/caf3bcd7664f4f2691d0ca9ef3be9a2a954fecfb
-      if !defined?(::Google::Protobuf) && Gem.loaded_specs['google-protobuf'].nil?
+      if !protobuf_already_loaded? && Gem.loaded_specs['google-protobuf'].nil?
         "Missing google-protobuf dependency; please add `gem 'google-protobuf', '~> 3.0'` to your Gemfile or gems.rb file"
       end
     end
@@ -74,12 +105,16 @@ module Datadog
       # See above for why we skip the check when protobuf is already loaded; note that when protobuf was already loaded
       # we skip the version check to avoid the call to Gem.loaded_specs. Unfortunately, protobuf does not seem to
       # expose the gem version constant elsewhere, so in that setup we are not able to check the version.
-      if !defined?(::Google::Protobuf) && Gem.loaded_specs['google-protobuf'].version < GOOGLE_PROTOBUF_MINIMUM_VERSION
+      if !protobuf_already_loaded? && Gem.loaded_specs['google-protobuf'].version < GOOGLE_PROTOBUF_MINIMUM_VERSION
         'Your google-protobuf is too old; ensure that you have google-protobuf >= 3.0 by ' \
         "adding `gem 'google-protobuf', '~> 3.0'` to your Gemfile or gems.rb file"
       end
     end
 
+    private_class_method def self.protobuf_already_loaded?
+      defined?(::Google::Protobuf) && !defined?(::Protobuf)
+    end
+
     private_class_method def self.protobuf_failed_to_load?
       unless protobuf_loaded_successfully?
         'There was an error loading the google-protobuf library; see previous warning message for details'
@@ -105,7 +140,7 @@ module Datadog
         # NOTE: We use Kernel#warn here because this code gets run BEFORE Datadog.logger is actually set up.
         # In the future it'd be nice to shuffle the logger startup to happen first to avoid this special case.
         Kernel.warn(
-          '[DDTRACE] Error while loading google-protobuf gem. ' \
+          '[ddtrace] Error while loading google-protobuf gem. ' \
           "Cause: '#{e.class.name} #{e.message}' Location: '#{Array(e.backtrace).first}'. " \
           'This can happen when google-protobuf is missing its native components. ' \
           'To fix this, try removing and reinstalling the gem, forcing it to recompile the components: ' \
@@ -147,12 +182,12 @@ module Datadog
 
       require_relative 'profiling/ext/forking'
       require_relative 'profiling/collectors/code_provenance'
-      require_relative 'profiling/collectors/cpu_and_wall_time'
       require_relative 'profiling/collectors/cpu_and_wall_time_worker'
       require_relative 'profiling/collectors/dynamic_sampling_rate'
       require_relative 'profiling/collectors/idle_sampling_helper'
       require_relative 'profiling/collectors/old_stack'
       require_relative 'profiling/collectors/stack'
+      require_relative 'profiling/collectors/thread_context'
       require_relative 'profiling/stack_recorder'
       require_relative 'profiling/old_recorder'
       require_relative 'profiling/exporter'
@@ -165,6 +200,8 @@ module Datadog
       require_relative 'profiling/tag_builder'
       require_relative 'profiling/http_transport'
 
+      replace_noop_allocation_count
+
       true
     end
 

data/lib/datadog/tracing/analytics.rb

@@ -1,5 +1,3 @@
-# typed: true
-
 require_relative 'metadata/ext'
 
 module Datadog