datadog 2.0.0.beta1

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

@@ -0,0 +1,60 @@
+# frozen_string_literal: true
+
+require_relative '../../core/utils/only_once'
+require_relative '../ext/forking'
+
+module Datadog
+  module Profiling
+    module Tasks
+      # Takes care of loading our extensions/monkey patches to handle fork() and validating if CPU-time profiling is usable
+      class Setup
+        ACTIVATE_EXTENSIONS_ONLY_ONCE = Core::Utils::OnlyOnce.new
+
+        def run
+          ACTIVATE_EXTENSIONS_ONLY_ONCE.run do
+            begin
+              activate_forking_extensions
+              setup_at_fork_hooks
+            rescue StandardError, ScriptError => e
+              Datadog.logger.warn do
+                "Profiler extensions unavailable. Cause: #{e.class.name} #{e.message} " \
+                "Location: #{Array(e.backtrace).first}"
+              end
+            end
+          end
+        end
+
+        private
+
+        def activate_forking_extensions
+          if Ext::Forking.supported?
+            Ext::Forking.apply!
+          elsif Datadog.configuration.profiling.enabled
+            Datadog.logger.debug('Profiler forking extensions skipped; forking not supported.')
+          end
+        rescue StandardError, ScriptError => e
+          Datadog.logger.warn do
+            "Profiler forking extensions unavailable. Cause: #{e.class.name} #{e.message} " \
+            "Location: #{Array(e.backtrace).first}"
+          end
+        end
+
+        def setup_at_fork_hooks
+          if Process.respond_to?(:at_fork)
+            Process.at_fork(:child) do
+              begin
+                # Restart profiler, if enabled
+                Profiling.start_if_enabled
+              rescue StandardError => e
+                Datadog.logger.warn do
+                  "Error during post-fork hooks. Cause: #{e.class.name} #{e.message} " \
+                  "Location: #{Array(e.backtrace).first}"
+                end
+              end
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/lib/datadog/profiling.rb

@@ -0,0 +1,152 @@
+# frozen_string_literal: true
+
+require_relative 'core'
+require_relative 'core/environment/variable_helpers'
+require_relative 'core/utils/only_once'
+
+module Datadog
+  # Datadog Continuous Profiler implementation: https://docs.datadoghq.com/profiler/
+  module Profiling
+    def self.supported?
+      unsupported_reason.nil?
+    end
+
+    def self.unsupported_reason
+      # NOTE: Only the first matching reason is returned, so try to keep a nice order on reasons -- e.g. tell users
+      # first that they can't use this on JRuby before telling them that something else failed
+
+      native_library_compilation_skipped? || native_library_failed_to_load?
+    end
+
+    # Starts the profiler, if the profiler is supported by in
+    # this runtime environment and if the profiler has been enabled
+    # in configuration.
+    #
+    # @return [Boolean] `true` if the profiler has successfully started, otherwise `false`.
+    # @public_api
+    def self.start_if_enabled
+      # If the profiler was not previously touched, getting the profiler instance triggers start as a side-effect
+      # otherwise we get nil
+      profiler = Datadog.send(:components).profiler
+      # ...but we still try to start it BECAUSE if the process forks, the profiler will exist but may
+      # not yet have been started in the fork
+      profiler.start if profiler
+      !!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, and allocation-related features are not disabled via configuration.
+    #
+    # @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
+
+    def self.enabled?
+      profiler = Datadog.send(:components).profiler
+      # Use .send(...) to avoid exposing the attr_reader as an API to the outside
+      !!(profiler.send(:scheduler).running? if profiler)
+    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
+      end
+    end
+
+    private_class_method def self.native_library_compilation_skipped?
+      skipped_reason = try_reading_skipped_reason_file
+
+      "Your datadog installation is missing support for the Continuous Profiler because #{skipped_reason}" if skipped_reason
+    end
+
+    private_class_method def self.try_reading_skipped_reason_file(file_api = File)
+      # This file, if it exists, is recorded by extconf.rb during compilation of the native extension
+      skipped_reason_file = "#{__dir__}/../../ext/datadog_profiling_native_extension/skipped_reason.txt"
+
+      begin
+        return unless file_api.exist?(skipped_reason_file)
+
+        contents = file_api.read(skipped_reason_file).strip
+        contents unless contents.empty?
+      rescue StandardError
+        # Do nothing
+      end
+    end
+
+    private_class_method def self.native_library_failed_to_load?
+      success, exception = try_loading_native_library
+
+      unless success
+        if exception
+          'There was an error loading the profiling native extension due to ' \
+          "'#{exception.class.name} #{exception.message}' at '#{Array(exception.backtrace).first}'"
+        else
+          'The profiling native extension did not load correctly. ' \
+          'For help solving this issue, please contact Datadog support at <https://docs.datadoghq.com/help/>.' \
+        end
+      end
+    end
+
+    private_class_method def self.try_loading_native_library
+      begin
+        require_relative 'profiling/load_native_extension'
+
+        success =
+          defined?(Profiling::NativeExtension) && Profiling::NativeExtension.send(:native_working?)
+        [success, nil]
+      rescue StandardError, LoadError => e
+        [false, e]
+      end
+    end
+
+    # All requires for the profiler should be directly added here; and everything should be loaded eagerly.
+    # (Currently there's a leftovers that need to be cleaned up, but we should avoid other exceptions.)
+    #
+    # All of the profiler should be loaded and ready to go when this method returns `true`.
+    private_class_method def self.load_profiling
+      return false unless supported?
+
+      require_relative 'profiling/ext/forking'
+      require_relative 'profiling/collectors/info'
+      require_relative 'profiling/collectors/code_provenance'
+      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/stack'
+      require_relative 'profiling/collectors/thread_context'
+      require_relative 'profiling/stack_recorder'
+      require_relative 'profiling/exporter'
+      require_relative 'profiling/flush'
+      require_relative 'profiling/scheduler'
+      require_relative 'profiling/tasks/setup'
+      require_relative 'profiling/profiler'
+      require_relative 'profiling/native_extension'
+      require_relative 'profiling/tag_builder'
+      require_relative 'profiling/http_transport'
+
+      replace_noop_allocation_count
+
+      true
+    end
+
+    load_profiling
+  end
+end

data/lib/datadog/tracing/analytics.rb

@@ -0,0 +1,25 @@
+# frozen_string_literal: true
+
+require_relative 'metadata/ext'
+
+module Datadog
+  module Tracing
+    # Defines analytics behavior
+    module Analytics
+      class << self
+        def set_sample_rate(span_op, sample_rate)
+          return if span_op.nil? || !sample_rate.is_a?(Numeric)
+
+          span_op.set_metric(Metadata::Ext::Analytics::TAG_SAMPLE_RATE, sample_rate)
+        end
+
+        def set_measured(span_op, value = true)
+          return if span_op.nil?
+
+          value = value == true || value == 1 ? 1 : 0 # rubocop:disable Style/MultipleComparison
+          span_op.set_metric(Metadata::Ext::Analytics::TAG_MEASURED, value)
+        end
+      end
+    end
+  end
+end

data/lib/datadog/tracing/buffer.rb

@@ -0,0 +1,129 @@
+# frozen_string_literal: true
+
+require_relative '../core/environment/ext'
+require_relative '../core/buffer/thread_safe'
+require_relative '../core/buffer/cruby'
+require_relative '../core/diagnostics/health'
+
+module Datadog
+  module Tracing
+    # Health metrics for trace buffers.
+    module MeasuredBuffer
+      def initialize(*_)
+        super
+
+        @buffer_accepted = 0
+        @buffer_accepted_lengths = 0
+        @buffer_dropped = 0
+        @buffer_spans = 0
+      end
+
+      def add!(trace)
+        super
+
+        # Emit health metrics
+        measure_accept(trace)
+      end
+
+      def add_all!(traces)
+        super
+
+        # Emit health metrics
+        traces.each { |trace| measure_accept(trace) }
+      end
+
+      def replace!(trace)
+        discarded_trace = super
+
+        # Emit health metrics
+        measure_accept(trace)
+        measure_drop(discarded_trace) if discarded_trace
+
+        discarded_trace
+      end
+
+      # Stored traces are returned and the local buffer is reset.
+      def drain!
+        traces = super
+        measure_pop(traces)
+        traces
+      end
+
+      def measure_accept(trace)
+        @buffer_accepted += 1
+        @buffer_accepted_lengths += trace.length
+
+        @buffer_spans += trace.length
+      rescue StandardError => e
+        Datadog.logger.debug(
+          "Failed to measure queue accept. Cause: #{e.class.name} #{e.message} Source: #{Array(e.backtrace).first}"
+        )
+      end
+
+      def measure_drop(trace)
+        @buffer_dropped += 1
+
+        @buffer_spans -= trace.length
+      rescue StandardError => e
+        Datadog.logger.debug(
+          "Failed to measure queue drop. Cause: #{e.class.name} #{e.message} Source: #{Array(e.backtrace).first}"
+        )
+      end
+
+      def measure_pop(traces)
+        # Accepted, cumulative totals
+        Datadog.health_metrics.queue_accepted(@buffer_accepted)
+        Datadog.health_metrics.queue_accepted_lengths(@buffer_accepted_lengths)
+
+        # Dropped, cumulative totals
+        Datadog.health_metrics.queue_dropped(@buffer_dropped)
+        # TODO: are we missing a +queue_dropped_lengths+ metric?
+
+        # Queue gauges, current values
+        Datadog.health_metrics.queue_max_length(@max_size)
+        Datadog.health_metrics.queue_spans(@buffer_spans)
+        Datadog.health_metrics.queue_length(traces.length)
+
+        # Reset aggregated metrics
+        @buffer_accepted = 0
+        @buffer_accepted_lengths = 0
+        @buffer_dropped = 0
+        @buffer_spans = 0
+      rescue StandardError => e
+        Datadog.logger.debug(
+          "Failed to measure queue. Cause: #{e.class.name} #{e.message} Source: #{Array(e.backtrace).first}"
+        )
+      end
+    end
+
+    # Trace buffer that stores application traces, has a maximum size, and
+    # can be safely used concurrently on any environment.
+    #
+    # @see Datadog::Core::Buffer::ThreadSafe
+    class ThreadSafeTraceBuffer < Core::Buffer::ThreadSafe
+      prepend MeasuredBuffer
+    end
+
+    # Trace buffer that stores application traces, has a maximum size, and
+    # can be safely used concurrently with CRuby.
+    #
+    # @see Datadog::Core::Buffer::CRuby
+    class CRubyTraceBuffer < Core::Buffer::CRuby
+      prepend MeasuredBuffer
+    end
+
+    # Trace buffer that stores application traces. The buffer has a maximum size and when
+    # the buffer is full, a random trace is discarded. This class is thread-safe and is used
+    # automatically by the ``Tracer`` instance when a ``Span`` is finished.
+    #
+    # We choose the default TraceBuffer implementation for current platform dynamically here.
+    #
+    # TODO We should restructure this module, so that classes are not declared at top-level ::Datadog.
+    # TODO Making such a change is potentially breaking for users manually configuring the tracer.
+    TraceBuffer = if Core::Environment::Ext::RUBY_ENGINE == 'ruby'
+                    CRubyTraceBuffer
+                  else
+                    ThreadSafeTraceBuffer
+                  end
+  end
+end

data/lib/datadog/tracing/client_ip.rb

@@ -0,0 +1,61 @@
+# frozen_string_literal: true
+
+require_relative '../core/configuration'
+require_relative '../core/utils/network'
+require_relative 'metadata/ext'
+require_relative 'span'
+
+module Datadog
+  module Tracing
+    # Common functions for supporting the `http.client_ip` span attribute.
+    module ClientIp
+      class << self
+        # Sets the `http.client_ip` tag on the given span.
+        #
+        # This function respects the user's settings: if they disable the client IP tagging,
+        #   or provide a different IP header name.
+        #
+        # @param [Span] span The span that's associated with the request.
+        # @param [HeaderCollection, #get, nil] headers A collection with the request headers.
+        # @param [String, nil] remote_ip The remote IP the request associated with the span is sent to.
+        def set_client_ip_tag(span, headers: nil, remote_ip: nil)
+          return unless configuration.enabled
+
+          set_client_ip_tag!(span, headers: headers, remote_ip: remote_ip)
+        end
+
+        # Forcefully sets the `http.client_ip` tag on the given span.
+        #
+        # This function ignores the user's `enabled` setting.
+        #
+        # @param [Span] span The span that's associated with the request.
+        # @param [HeaderCollection, #get, nil] headers A collection with the request headers.
+        # @param [String, nil] remote_ip The remote IP the request associated with the span is sent to.
+        def set_client_ip_tag!(span, headers: nil, remote_ip: nil)
+          ip = extract_client_ip(headers, remote_ip)
+
+          span.set_tag(Tracing::Metadata::Ext::HTTP::TAG_CLIENT_IP, ip) if ip
+        end
+
+        def extract_client_ip(headers, remote_ip)
+          if headers && configuration.header_name
+            return Datadog::Core::Utils::Network.stripped_ip_from_request_headers(
+              headers,
+              ip_headers_to_check: Array(configuration.header_name)
+            )
+          end
+
+          ip_from_headers = Datadog::Core::Utils::Network.stripped_ip_from_request_headers(headers) if headers
+
+          ip_from_headers || Datadog::Core::Utils::Network.stripped_ip(remote_ip)
+        end
+
+        private
+
+        def configuration
+          Datadog.configuration.tracing.client_ip
+        end
+      end
+    end
+  end
+end

data/lib/datadog/tracing/component.rb

@@ -0,0 +1,206 @@
+# frozen_string_literal: true
+
+require_relative 'tracer'
+require_relative 'flush'
+require_relative 'sync_writer'
+require_relative 'sampling/span/rule_parser'
+require_relative 'sampling/span/sampler'
+require_relative 'diagnostics/environment_logger'
+
+module Datadog
+  module Tracing
+    # Tracing component
+    module Component
+      # Methods that interact with component instance fields.
+      module InstanceMethods
+        # Hot-swaps with a new sampler.
+        # This operation acquires the Components lock to ensure
+        # there is no concurrent modification of the sampler.
+        def reconfigure_live_sampler
+          sampler = self.class.build_sampler(Datadog.configuration)
+          Datadog.send(:safely_synchronize) { tracer.sampler.sampler = sampler }
+        end
+      end
+
+      def build_tracer(settings, agent_settings, logger:)
+        # If a custom tracer has been provided, use it instead.
+        # Ignore all other options (they should already be configured.)
+        tracer = settings.tracing.instance
+        return tracer unless tracer.nil?
+
+        # Apply test mode settings if test mode is activated
+        if settings.tracing.test_mode.enabled
+          trace_flush = build_test_mode_trace_flush(settings)
+          sampler = build_test_mode_sampler
+          writer = build_test_mode_writer(settings, agent_settings)
+        else
+          trace_flush = build_trace_flush(settings)
+          sampler = build_sampler(settings)
+          writer = build_writer(settings, agent_settings)
+        end
+
+        # The sampler instance is wrapped in a delegator,
+        # so dynamic instrumentation can hot-swap it.
+        # This prevents full tracer reinitialization on sampling changes.
+        sampler_delegator = SamplerDelegatorComponent.new(sampler)
+
+        subscribe_to_writer_events!(writer, sampler_delegator, settings.tracing.test_mode.enabled)
+
+        Tracing::Tracer.new(
+          default_service: settings.service,
+          enabled: settings.tracing.enabled,
+          trace_flush: trace_flush,
+          sampler: sampler_delegator,
+          span_sampler: build_span_sampler(settings),
+          writer: writer,
+          tags: build_tracer_tags(settings),
+        )
+      end
+
+      def build_trace_flush(settings)
+        if settings.tracing.partial_flush.enabled
+          Tracing::Flush::Partial.new(
+            min_spans_before_partial_flush: settings.tracing.partial_flush.min_spans_threshold
+          )
+        else
+          Tracing::Flush::Finished.new
+        end
+      end
+
+      def build_sampler(settings)
+        # A custom sampler is provided
+        if (sampler = settings.tracing.sampler)
+          return sampler
+        end
+
+        # Sampling rules are provided
+        if (rules = settings.tracing.sampling.rules)
+          post_sampler = Tracing::Sampling::RuleSampler.parse(
+            rules,
+            settings.tracing.sampling.rate_limit,
+            settings.tracing.sampling.default_rate
+          )
+        end
+
+        # The default sampler.
+        # Used if no custom sampler is provided, or if sampling rule parsing fails.
+        post_sampler ||= Tracing::Sampling::RuleSampler.new(
+          rate_limit: settings.tracing.sampling.rate_limit,
+          default_sample_rate: settings.tracing.sampling.default_rate
+        )
+
+        Tracing::Sampling::PrioritySampler.new(
+          base_sampler: Tracing::Sampling::AllSampler.new,
+          post_sampler: post_sampler
+        )
+      end
+
+      # TODO: Writer should be a top-level component.
+      # It is currently part of the Tracer initialization
+      # process, but can take a variety of options (including
+      # a fully custom instance) that makes the Tracer
+      # initialization process complex.
+      def build_writer(settings, agent_settings, options = settings.tracing.writer_options)
+        if (writer = settings.tracing.writer)
+          return writer
+        end
+
+        Tracing::Writer.new(agent_settings: agent_settings, **options)
+      end
+
+      def subscribe_to_writer_events!(writer, sampler_delegator, test_mode)
+        return unless writer.respond_to?(:events) # Check if it's a custom, external writer
+
+        writer.events.after_send.subscribe(&WRITER_RECORD_ENVIRONMENT_INFORMATION_CALLBACK)
+
+        # DEV: We need to ignore priority sampling updates coming from the agent in test mode
+        # because test mode wants to *unconditionally* sample all traces.
+        #
+        # This can cause trace metrics to be overestimated, but that's a trade-off we take
+        # here to achieve 100% sampling rate.
+        return if test_mode
+
+        writer.events.after_send.subscribe(&writer_update_priority_sampler_rates_callback(sampler_delegator))
+      end
+
+      WRITER_RECORD_ENVIRONMENT_INFORMATION_CALLBACK = lambda do |_, responses|
+        Tracing::Diagnostics::EnvironmentLogger.collect_and_log!(responses: responses)
+      end
+
+      # Create new lambda for writer callback,
+      # capture the current sampler in the callback closure.
+      def writer_update_priority_sampler_rates_callback(sampler)
+        lambda do |_, responses|
+          response = responses.last
+
+          next unless response && !response.internal_error? && response.service_rates
+
+          sampler.update(response.service_rates, decision: Tracing::Sampling::Ext::Decision::AGENT_RATE)
+        end
+      end
+
+      def build_span_sampler(settings)
+        rules = Tracing::Sampling::Span::RuleParser.parse_json(settings.tracing.sampling.span_rules)
+        Tracing::Sampling::Span::Sampler.new(rules || [])
+      end
+
+      # Configure non-privileged components.
+      def configure_tracing(settings)
+        Datadog::Tracing::Contrib::Component.configure(settings)
+      end
+
+      # Sampler wrapper component, to allow for hot-swapping
+      # the sampler instance used by the tracer.
+      # Swapping samplers happens during Dynamic Configuration.
+      class SamplerDelegatorComponent
+        attr_accessor :sampler
+
+        def initialize(sampler)
+          @sampler = sampler
+        end
+
+        def sample!(trace)
+          @sampler.sample!(trace)
+        end
+
+        def update(*args, **kwargs)
+          return unless @sampler.respond_to?(:update)
+
+          @sampler.update(*args, **kwargs)
+        end
+      end
+
+      private
+
+      def build_tracer_tags(settings)
+        settings.tags.dup.tap do |tags|
+          tags[Core::Environment::Ext::TAG_ENV] = settings.env unless settings.env.nil?
+          tags[Core::Environment::Ext::TAG_VERSION] = settings.version unless settings.version.nil?
+        end
+      end
+
+      def build_test_mode_trace_flush(settings)
+        # If context flush behavior is provided, use it instead.
+        settings.tracing.test_mode.trace_flush || build_trace_flush(settings)
+      end
+
+      def build_test_mode_sampler
+        # Do not sample any spans for tests; all must be preserved.
+        # Set priority sampler to ensure the agent doesn't drop any traces.
+        Tracing::Sampling::PrioritySampler.new(
+          base_sampler: Tracing::Sampling::AllSampler.new,
+          post_sampler: Tracing::Sampling::AllSampler.new
+        )
+      end
+
+      def build_test_mode_writer(settings, agent_settings)
+        writer_options = settings.tracing.test_mode.writer_options || {}
+
+        return build_writer(settings, agent_settings, writer_options) if settings.tracing.test_mode.async
+
+        # Flush traces synchronously, to guarantee they are written.
+        Tracing::SyncWriter.new(agent_settings: agent_settings, **writer_options)
+      end
+    end
+  end
+end