datadog 2.21.0 → 2.38.0

data/lib/datadog/opentelemetry/signal_configuration.rb

@@ -0,0 +1,53 @@
+# frozen_string_literal: true
+
+require_relative '../core/configuration/ext'
+require_relative '../core/environment/socket'
+
+module Datadog
+  module OpenTelemetry
+    # Shared resource building and signal-specific config fallback logic for Logs and Metrics.
+    module SignalConfiguration
+      private
+
+      def create_resource
+        resource_attributes = {}
+
+        @settings.tags&.each do |key, value| # steep:ignore
+          otel_key = case key
+          when 'service' then 'service.name'
+          when 'env' then 'deployment.environment'
+          when 'version' then 'service.version'
+          else key
+          end
+          resource_attributes[otel_key] = value
+        end
+
+        resource_attributes['service.name'] = @settings.service_without_fallback || resource_attributes['service.name'] || Datadog::Core::Environment::Ext::FALLBACK_SERVICE_NAME # steep:ignore
+        resource_attributes['deployment.environment'] = @settings.env if @settings.env # steep:ignore
+        resource_attributes['service.version'] = @settings.version if @settings.version # steep:ignore
+
+        hostname = Datadog::Core::Environment::Socket.resolved_hostname(@settings) # steep:ignore
+        if hostname
+          if hostname == @settings.hostname # steep:ignore
+            resource_attributes['host.name'] = hostname
+          elsif !resource_attributes.key?('host.name')
+            resource_attributes['host.name'] = hostname
+          end
+        end
+
+        ::OpenTelemetry::SDK::Resources::Resource.create(resource_attributes)
+      end
+
+      # Returns the signal-specific option value when explicitly set,
+      # otherwise falls back to the general OTLP exporter config or computed_default.
+      def config_or_exporter_fallback(signal:, option_name:, computed_default: nil)
+        signal_settings = @settings.opentelemetry.public_send(signal) # steep:ignore
+        if signal_settings.using_default?(option_name)
+          @settings.opentelemetry.exporter.public_send(option_name) || computed_default # steep:ignore
+        else
+          signal_settings.public_send(option_name)
+        end
+      end
+    end
+  end
+end

data/lib/datadog/opentelemetry.rb

@@ -22,6 +22,9 @@ require_relative 'opentelemetry/api/baggage'
 require_relative 'opentelemetry/sdk/configurator' if defined?(OpenTelemetry::SDK)
 require_relative 'opentelemetry/sdk/trace/span' if defined?(OpenTelemetry::SDK)
 
+require_relative 'opentelemetry/metrics' if defined?(OpenTelemetry::SDK::Metrics)
+require_relative 'opentelemetry/logs' if defined?(OpenTelemetry::SDK::Logs)
+
 module Datadog
   # Datadog OpenTelemetry integration.
   module OpenTelemetry
@@ -47,6 +50,7 @@ end
 # Currently, this closely translates to Datadog's partial flushing.
 #
 # @see OpenTelemetry::SDK::Trace::SpanProcessor#on_finish
+
 Datadog.configure do |c|
   c.tracing.partial_flush.enabled = true
 end

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

@@ -14,6 +14,9 @@ module Datadog
       #
       # This class acts both as a collector (collecting data) as well as a recorder (records/serializes it)
       class CodeProvenance
+        # @rbs standard_library_path: ::String
+        # @rbs ruby_native_filename: ::String?
+        # @rbs return: void
         def initialize(
           standard_library_path: RbConfig::CONFIG.fetch("rubylibdir"),
           ruby_native_filename: Datadog::Profiling::Collectors::Stack._native_ruby_native_filename
@@ -22,6 +25,7 @@ module Datadog
           @libraries_by_path = {}
           @seen_files = Set.new
           @seen_libraries = Set.new
+          @executable_paths = [Gem.bindir, bundler_bin_path].uniq.compact.freeze
 
           record_library(
             Library.new(
@@ -29,11 +33,14 @@ module Datadog
               name: "stdlib",
               version: RUBY_VERSION,
               path: standard_library_path,
-              extra_path: ruby_native_filename,
+              extra_paths: [ruby_native_filename],
             )
           )
         end
 
+        # @rbs loaded_files: ::Array[::String]
+        # @rbs loaded_specs: ::Array[::Gem::BasicSpecification]
+        # @rbs return: self
         def refresh(loaded_files: $LOADED_FEATURES, loaded_specs: Gem.loaded_specs.values)
           record_loaded_specs(loaded_specs)
           record_loaded_files(loaded_files)
@@ -41,18 +48,20 @@ module Datadog
           self
         end
 
+        #: () -> ::String
         def generate_json
           JSON.generate(v1: seen_libraries.to_a)
         end
 
         private
 
-        attr_reader \
-          :libraries_by_name,
-          :libraries_by_path,
-          :seen_files,
-          :seen_libraries
+        attr_reader :libraries_by_name #: ::Hash[::String, Library]
+        attr_reader :libraries_by_path #: ::Hash[::String, Library]
+        attr_reader :seen_files #: ::Set[::String]
+        attr_reader :seen_libraries #: ::Set[Library]
+        attr_reader :executable_paths #: ::Array[::String]
 
+        #: (Library) -> void
         def record_library(library)
           libraries_by_name[library.name] = library
           libraries_by_path[library.path] = library
@@ -69,23 +78,32 @@ module Datadog
         #
         # Alternatively/in the future we could instead use a trie to match paths, but I doubt for the data sizes we're
         # looking at that a trie is that much faster than using Ruby's built-in native collections.
+        #: () -> void
         def sort_libraries_by_longest_path_first
           @libraries_by_path = @libraries_by_path.sort.reverse!.to_h
         end
 
+        #: (::Array[::Gem::BasicSpecification]) -> void
         def record_loaded_specs(loaded_specs)
           recorded_library = false
 
           loaded_specs.each do |spec|
             next if libraries_by_name.key?(spec.name)
 
+            extra_paths = [(spec.extension_dir if spec.extensions.any?)]
+            spec.executables&.each do |executable|
+              executable_paths.each do |path|
+                extra_paths << File.join(path, executable)
+              end
+            end
+
             record_library(
               Library.new(
                 kind: "library",
                 name: spec.name,
                 version: spec.version,
                 path: spec.gem_dir,
-                extra_path: (spec.extension_dir if spec.extensions.any?),
+                extra_paths: extra_paths,
               )
             )
             recorded_library = true
@@ -94,6 +112,7 @@ module Datadog
           sort_libraries_by_longest_path_first if recorded_library
         end
 
+        #: (::Array[::String]) -> void
         def record_loaded_files(loaded_files)
           loaded_files.each do |file_path|
             next if seen_files.include?(file_path)
@@ -107,6 +126,31 @@ module Datadog
           end
         end
 
+        # This is intended to mirror bundler's `Bundler.bin_path` with one key difference: the bundler version of the
+        # method **tries to create the path** if it doesn't exist, whereas our version doesn't.
+        #
+        # This "try to create the path" is annoying because creating the path can fail if e.g. the app doesn't have
+        # permissions to do that, see https://github.com/DataDog/dd-trace-rb/issues/5137.
+        # Thus we have our own version that a) Avoids creating the path, and b) Rescues exceptions to avoid any
+        # bundler complaints here impacting the application. (Bundler tends to go "something is wrong, raise!" which
+        # I think makes a lot of sense given how bundler is intended to be used, but for this our kind of "ask a few
+        # questions usage" it's not what we want.)
+        #: () -> ::String?
+        def bundler_bin_path
+          return unless defined?(Bundler)
+
+          path = Bundler.settings[:bin] || "bin"
+          root = Bundler.root
+          result = Pathname.new(path).expand_path(root).expand_path
+          result.to_s
+        rescue Exception => e # rubocop:disable Lint/RescueException
+          Datadog.logger.debug(
+            "CodeProvenance#bundler_bin_path failed. " \
+            "Cause: #{e.class}: #{e.message} Location: #{Array(e.backtrace).first}"
+          )
+          nil
+        end
+
         # Represents metadata we have for a ruby gem
         #
         # Important note: This class gets encoded to JSON with the built-in JSON gem. But, we've found that in some
@@ -116,21 +160,36 @@ module Datadog
         # Thus, this class was setup to match the JSON output. Take this into consideration if you are adding new
         # fields. (Also, we have a spec for this)
         class Library
-          attr_reader :kind, :name, :version
-
-          def initialize(kind:, name:, version:, path:, extra_path: nil)
-            extra_path = nil if extra_path&.empty?
+          # @rbs @kind: ::String
+          # @rbs @name: ::String
+          # @rbs @version: ::String
+          # @rbs @paths: ::Array[::String]
+
+          attr_reader :kind #: ::String
+          attr_reader :name #: ::String
+          attr_reader :version #: ::String
+
+          # @rbs kind: ::String
+          # @rbs name: ::String
+          # @rbs version: ::String | ::Gem::Version
+          # @rbs path: ::String
+          # @rbs extra_paths: ::Array[::String?]
+          # @rbs return: void
+          def initialize(kind:, name:, version:, path:, extra_paths:)
+            extra_paths = Array(extra_paths).compact.reject(&:empty?).map { |p| p.dup.freeze }
             @kind = kind.freeze
             @name = name.dup.freeze
             @version = version.to_s.dup.freeze
-            @paths = [path.dup.freeze, extra_path.dup.freeze].compact.freeze
+            @paths = [path.dup.freeze, *extra_paths].freeze
             freeze
           end
 
+          #: (?::JSON::State?) -> ::String
           def to_json(arg = nil)
             {kind: @kind, name: @name, version: @version, paths: @paths}.to_json(arg)
           end
 
+          #: () -> ::String
           def path
             @paths.first
           end

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

@@ -9,12 +9,32 @@ module Datadog
       #
       # Methods prefixed with _native_ are implemented in `collectors_cpu_and_wall_time_worker.c`
       class CpuAndWallTimeWorker
+        # @rbs @worker_thread: untyped
+        # @rbs @start_stop_mutex: ::Thread::Mutex
+        # @rbs @failure_exception: ::Exception?
+        # @rbs @idle_sampling_helper: IdleSamplingHelper
+        # @rbs @wait_until_running_mutex: ::Thread::Mutex
+        # @rbs @wait_until_running_condition: ::Thread::ConditionVariable
+
         private
 
-        attr_accessor :failure_exception
+        attr_accessor :failure_exception #: ::Exception?
 
         public
 
+        # @rbs gc_profiling_enabled: bool
+        # @rbs no_signals_workaround_enabled: bool
+        # @rbs thread_context_collector: Datadog::Profiling::Collectors::ThreadContext
+        # @rbs dynamic_sampling_rate_overhead_target_percentage: Float
+        # @rbs cpu_sampling_interval_ms: ::Integer
+        # @rbs idle_sampling_helper: Datadog::Profiling::Collectors::IdleSamplingHelper
+        # @rbs dynamic_sampling_rate_enabled: bool
+        # @rbs allocation_profiling_enabled: bool
+        # @rbs allocation_counting_enabled: bool
+        # @rbs gvl_profiling_enabled: bool
+        # @rbs sighandler_sampling_enabled: bool
+        # @rbs skip_idle_samples_for_testing: false
+        # @rbs return: void
         def initialize(
           gc_profiling_enabled:,
           no_signals_workaround_enabled:,
@@ -24,11 +44,12 @@ module Datadog
           allocation_counting_enabled:,
           gvl_profiling_enabled:,
           sighandler_sampling_enabled:,
+          cpu_sampling_interval_ms:,
           # **NOTE**: This should only be used for testing; disabling the dynamic sampling rate will increase the
           # profiler overhead!
           dynamic_sampling_rate_enabled: true,
           skip_idle_samples_for_testing: false,
-          idle_sampling_helper: IdleSamplingHelper.new
+          idle_sampling_helper: IdleSamplingHelper.new(thread_context_collector: thread_context_collector)
         )
           unless dynamic_sampling_rate_enabled
             Datadog.logger.warn(
@@ -39,6 +60,10 @@ module Datadog
             )
           end
 
+          if cpu_sampling_interval_ms < 1
+            raise ArgumentError, "cpu_sampling_interval_ms must be a positive integer, got #{cpu_sampling_interval_ms}"
+          end
+
           self.class._native_initialize(
             self_instance: self,
             thread_context_collector: thread_context_collector,
@@ -52,6 +77,7 @@ module Datadog
             gvl_profiling_enabled: gvl_profiling_enabled,
             sighandler_sampling_enabled: sighandler_sampling_enabled,
             skip_idle_samples_for_testing: skip_idle_samples_for_testing,
+            cpu_sampling_interval_ms: cpu_sampling_interval_ms,
           )
           @worker_thread = nil
           @failure_exception = nil
@@ -61,6 +87,8 @@ module Datadog
           @wait_until_running_condition = ConditionVariable.new
         end
 
+        # @rbs on_failure_proc: (^(?log_failure: bool) -> void)?
+        # @rbs return: bool?
         def start(on_failure_proc: nil)
           @start_stop_mutex.synchronize do
             return if @worker_thread&.alive?
@@ -75,14 +103,22 @@ module Datadog
               self.class._native_sampling_loop(self)
 
               Datadog.logger.debug("CpuAndWallTimeWorker thread stopping cleanly")
+            rescue Profiling::ExistingSignalHandler => e
+              @failure_exception = e
+              Datadog.logger.warn(
+                "Profiling was not started as another profiler or gem is already using the SIGPROF signal. " \
+                "Please disable the other profiler to use Datadog profiling."
+              )
+              on_failure_proc&.call(log_failure: false)
             rescue Exception => e # rubocop:disable Lint/RescueException
               @failure_exception = e
+              operation_name = self.class._native_failure_exception_during_operation(self).inspect
               Datadog.logger.warn(
                 "CpuAndWallTimeWorker thread error. " \
-                "Cause: #{e.class.name} #{e.message} Location: #{Array(e.backtrace).first}"
+                "Operation: #{operation_name} Cause: #{e.class}: #{e.message} Location: #{Array(e.backtrace).first}"
               )
               on_failure_proc&.call
-              Datadog::Core::Telemetry::Logger.report(e, description: "CpuAndWallTimeWorker thread error", pii_safe: true)
+              Datadog::Core::Telemetry::Logger.report(e, description: "CpuAndWallTimeWorker thread error: #{operation_name}")
             end
             @worker_thread.name = self.class.name # Repeated from above to make sure thread gets named asap
             @worker_thread.thread_variable_set(:fork_safe, true)
@@ -91,6 +127,7 @@ module Datadog
           true
         end
 
+        #: () -> void
         def stop
           @start_stop_mutex.synchronize do
             Datadog.logger.debug("Requesting CpuAndWallTimeWorker thread shut down")
@@ -107,14 +144,17 @@ module Datadog
           end
         end
 
+        #: () -> true
         def reset_after_fork
           self.class._native_reset_after_fork(self)
         end
 
+        #: () -> ::Hash[::Symbol, untyped]
         def stats
           self.class._native_stats(self)
         end
 
+        #: () -> ::Hash[::Symbol, untyped]
         def stats_and_reset_not_thread_safe
           stats = self.stats
           self.class._native_stats_reset_not_thread_safe(self)
@@ -122,22 +162,30 @@ module Datadog
         end
 
         # Useful for testing, to e.g. make sure the profiler is running before we start running some code we want to observe
+        # @rbs timeout_seconds: ::Integer?
+        # @rbs return: true
         def wait_until_running(timeout_seconds: 5)
           @wait_until_running_mutex.synchronize do
             return true if self.class._native_is_running?(self)
 
             @wait_until_running_condition.wait(@wait_until_running_mutex, timeout_seconds)
 
+            # Not using `_native_is_running?(self) || raise(...)` here: `_native_is_running?` is declared
+            # to return `bool`, so steep infers the `||` expression as `bool` too and rejects it against
+            # this method's declared `true` return type. The explicit `if`/`else` type-checks correctly.
+            # rubocop:disable Style/RedundantCondition
             if self.class._native_is_running?(self)
               true
             else
               raise "Timeout waiting for #{self.class.name} to start (waited for #{timeout_seconds} seconds)"
             end
+            # rubocop:enable Style/RedundantCondition
           end
         end
 
         private
 
+        #: () -> void
         def signal_running
           @wait_until_running_mutex.synchronize { @wait_until_running_condition.broadcast }
         end

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

@@ -8,17 +8,24 @@ module Datadog
       #
       # Methods prefixed with _native_ are implemented in `collectors_idle_sampling_helper.c`
       class IdleSamplingHelper
+        # @rbs @worker_thread: untyped
+        # @rbs @start_stop_mutex: ::Thread::Mutex
+        # @rbs @thread_context_collector: Datadog::Profiling::Collectors::ThreadContext?
+
         private
 
-        attr_accessor :failure_exception
+        attr_accessor :failure_exception #: ::Exception?
 
         public
 
-        def initialize
+        #: (thread_context_collector: Datadog::Profiling::Collectors::ThreadContext) -> void
+        def initialize(thread_context_collector:)
           @worker_thread = nil
           @start_stop_mutex = Mutex.new
+          @thread_context_collector = thread_context_collector
         end
 
+        #: () -> (nil | true)
         def start
           @start_stop_mutex.synchronize do
             return if @worker_thread&.alive?
@@ -32,16 +39,16 @@ module Datadog
             @worker_thread = Thread.new do
               Thread.current.name = self.class.name
 
-              self.class._native_idle_sampling_loop(self)
+              self.class._native_idle_sampling_loop(self, @thread_context_collector)
 
               Datadog.logger.debug("IdleSamplingHelper thread stopping cleanly")
             rescue Exception => e # rubocop:disable Lint/RescueException
               @failure_exception = e
               Datadog.logger.warn(
                 "IdleSamplingHelper thread error. " \
-                "Cause: #{e.class.name} #{e.message} Location: #{Array(e.backtrace).first}"
+                "Cause: #{e.class}: #{e.message} Location: #{Array(e.backtrace).first}"
               )
-              Datadog::Core::Telemetry::Logger.report(e, description: "IdleSamplingHelper thread error", pii_safe: true)
+              Datadog::Core::Telemetry::Logger.report(e, description: "IdleSamplingHelper thread error")
             end
             @worker_thread.name = self.class.name # Repeated from above to make sure thread gets named asap
             @worker_thread.thread_variable_set(:fork_safe, true)
@@ -50,6 +57,7 @@ module Datadog
           true
         end
 
+        #: () -> void
         def stop
           @start_stop_mutex.synchronize do
             Datadog.logger.debug("Requesting IdleSamplingHelper thread shut down")

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

@@ -14,8 +14,17 @@ module Datadog
       # could be seen as overkill for this case but it allows us to centralize information
       # gathering and easily support more flexible/dynamic info collection in the future.
       class Info
+        # @rbs @info: ::Hash[::Symbol, ::Hash[::Symbol, untyped]]
+        # @rbs @platform_info: ::Hash[::Symbol, untyped]
+        # @rbs @runtime_info: ::Hash[::Symbol, untyped]
+        # @rbs @application_info: ::Hash[::Symbol, untyped]
+        # @rbs @profiler_info: ::Hash[::Symbol, untyped]?
+        # @rbs @gc_tuning_info: ::Hash[::Symbol, ::String]
+
+        #: (untyped) -> void
         def initialize(settings)
           @profiler_info = nil
+
           @info = {
             platform: collect_platform_info,
             runtime: collect_runtime_info,
@@ -24,7 +33,7 @@ module Datadog
           }.freeze
         end
 
-        attr_reader :info
+        attr_reader :info #: ::Hash[::Symbol, ::Hash[::Symbol, untyped]]
 
         private
 
@@ -67,6 +76,7 @@ module Datadog
         # gets initialized before a user has a chance to configure the library.
         START_TIME = Time.now.utc.freeze
 
+        #: () -> ::Hash[::Symbol, untyped]
         def collect_platform_info
           @platform_info ||= {
             container_id: Datadog::Core::Environment::Container.container_id,
@@ -77,6 +87,7 @@ module Datadog
           }.freeze
         end
 
+        #: () -> ::Hash[::Symbol, untyped]
         def collect_runtime_info
           @runtime_info ||= {
             engine: Datadog::Core::Environment::Identity.lang_engine,
@@ -86,6 +97,7 @@ module Datadog
           }.freeze
         end
 
+        #: (untyped) -> ::Hash[::Symbol, untyped]
         def collect_application_info(settings)
           @application_info ||= {
             start_time: START_TIME.iso8601,
@@ -95,8 +107,9 @@ module Datadog
           }.freeze
         end
 
+        #: (untyped) -> ::Hash[::Symbol, untyped]
         def collect_profiler_info(settings)
-          unless @profiler_info
+          @profiler_info ||= begin
             lib_datadog_gem = ::Gem.loaded_specs["libdatadog"]
 
             libdatadog_version =
@@ -108,13 +121,12 @@ module Datadog
                 "#{Libdatadog::VERSION}-(unknown)"
               end
 
-            @profiler_info = {
+            {
               version: Datadog::Core::Environment::Identity.gem_datadog_version,
               libdatadog: libdatadog_version,
               settings: collect_settings_recursively(settings.profiling),
             }.freeze
           end
-          @profiler_info
         end
 
         # The settings/option model isn't directly serializable because
@@ -122,6 +134,7 @@ module Datadog
         # instances without proper serialization.
         # This method navigates a settings object recursively, converting
         # it into more basic types that are trivially convertible to JSON.
+        #: (untyped) -> untyped
         def collect_settings_recursively(v)
           v = v.options_hash if v.respond_to?(:options_hash)
 
@@ -142,6 +155,7 @@ module Datadog
           end
         end
 
+        #: () -> ::Hash[::Symbol, ::String]
         def collect_gc_tuning_info
           return @gc_tuning_info if defined?(@gc_tuning_info)
 

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

@@ -19,7 +19,6 @@ module Datadog
           max_frames:,
           tracer:,
           endpoint_collection_enabled:,
-          timeline_enabled:,
           waiting_for_gvl_threshold_ns:,
           otel_context_enabled:,
           native_filenames_enabled:
@@ -31,10 +30,10 @@ module Datadog
             max_frames: max_frames,
             tracer_context_key: tracer_context_key,
             endpoint_collection_enabled: endpoint_collection_enabled,
-            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),
+            overhead_filename: __FILE__,
           )
         end
 
@@ -43,23 +42,29 @@ module Datadog
           max_frames: 400,
           tracer: nil,
           endpoint_collection_enabled: false,
-          timeline_enabled: false,
           waiting_for_gvl_threshold_ns: 10_000_000,
           otel_context_enabled: false,
           native_filenames_enabled: true,
+          trigger_global_reset: true,
           **options
         )
-          new(
+          collector = new(
             recorder: recorder,
             max_frames: max_frames,
             tracer: tracer,
             endpoint_collection_enabled: endpoint_collection_enabled,
-            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,
           )
+
+          # By default, mirror what the CpuAndWallTimeWorker does when profiling starts: reset the global per-thread
+          # context state, which (among other things) resizes the per-thread sampling buffers to this collector's
+          # max_frames. Tests that need to control this explicitly can pass `trigger_global_reset: false`.
+          Testing._native_global_reset_per_thread_context(collector) if trigger_global_reset
+
+          collector
         end
 
         def inspect