datadog 2.36.0 → 2.37.0

data/lib/datadog/core/configuration/components_state.rb

@@ -5,9 +5,10 @@ module Datadog
     module Configuration
       # Stores the state of component tree when replacing the tree.
       class ComponentsState
-        def initialize(telemetry_enabled:, remote_started:)
+        def initialize(telemetry_enabled:, remote_started:, di_implicitly_enabled: false)
           @telemetry_enabled = !!telemetry_enabled
           @remote_started = !!remote_started
+          @di_implicitly_enabled = !!di_implicitly_enabled
         end
 
         def telemetry_enabled?
@@ -17,6 +18,10 @@ module Datadog
         def remote_started?
           @remote_started
         end
+
+        def di_implicitly_enabled?
+          @di_implicitly_enabled
+        end
       end
     end
   end

data/lib/datadog/core/configuration/settings.rb

@@ -14,7 +14,6 @@ require_relative '../../profiling/ext'
 
 require_relative '../../tracing/configuration/settings'
 require_relative '../../opentelemetry/configuration/settings'
-require_relative '../../symbol_database/configuration'
 
 module Datadog
   module Core
@@ -1130,8 +1129,6 @@ module Datadog
         extend Datadog::Tracing::Configuration::Settings
 
         extend Datadog::OpenTelemetry::Configuration::Settings
-
-        extend Datadog::SymbolDatabase::Configuration::Settings
       end
       # standard:enable Metrics/BlockLength
     end

data/lib/datadog/core/configuration/supported_configurations.rb

@@ -19,6 +19,7 @@ module Datadog
           "DD_API_SECURITY_DOWNSTREAM_BODY_ANALYSIS_SAMPLE_RATE",
           "DD_API_SECURITY_ENABLED",
           "DD_API_SECURITY_ENDPOINT_COLLECTION_ENABLED",
+          "DD_API_SECURITY_MAX_DOWNSTREAM_BODY_BYTES",
           "DD_API_SECURITY_MAX_DOWNSTREAM_REQUEST_BODY_ANALYSIS",
           "DD_API_SECURITY_REQUEST_SAMPLE_RATE",
           "DD_API_SECURITY_SAMPLE_DELAY",
@@ -59,6 +60,7 @@ module Datadog
           "DD_EXPERIMENTAL_FLAGGING_PROVIDER_ENABLED",
           "DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED",
           "DD_EXTERNAL_ENV",
+          "DD_FLAGGING_EVALUATION_COUNTS_ENABLED",
           "DD_GIT_COMMIT_SHA",
           "DD_GIT_REPOSITORY_URL",
           "DD_HEALTH_METRICS_ENABLED",

data/lib/datadog/core/evp.rb

@@ -0,0 +1,11 @@
+# frozen_string_literal: true
+
+module Datadog
+  module Core
+    module EVP
+      SUBDOMAIN_HEADER_NAME = 'X-Datadog-EVP-Subdomain'
+      EVENT_PLATFORM_INTAKE_SUBDOMAIN = 'event-platform-intake'
+      PAYLOAD_SIZE_LIMIT_BYTES = 5 * 1024 * 1024
+    end
+  end
+end

data/lib/datadog/core/remote/client/capabilities.rb

@@ -4,6 +4,7 @@ require_relative '../../utils/base64_codec'
 require_relative '../../../appsec/remote'
 require_relative '../../../tracing/remote'
 require_relative '../../../di/remote'
+require_relative '../../../symbol_database'
 require_relative '../../../symbol_database/remote'
 require_relative '../../../open_feature/remote'
 
@@ -35,13 +36,43 @@ module Datadog
               register_receivers(Datadog::AppSec::Remote.receivers(@telemetry))
             end
 
-            if settings.respond_to?(:dynamic_instrumentation) && settings.dynamic_instrumentation.enabled
+            # Tracing must register before DI: on a combined RC dispatch,
+            # the APM_TRACING handler must run first to call
+            # Datadog::DI::Remote.handle_rc_enablement and start the
+            # component before the DI receiver processes LIVE_DEBUGGING
+            # changes against `component.started?`. Reversing the order
+            # silently drops the probe — the remote client only
+            # redispatches on content hash changes.
+            register_capabilities(Datadog::Tracing::Remote.capabilities)
+            register_products(Datadog::Tracing::Remote.products)
+            register_receivers(Datadog::Tracing::Remote.receivers(@telemetry))
+
+            # Skip DI registration entirely when DI is explicitly disabled
+            # (DD_DYNAMIC_INSTRUMENTATION_ENABLED=false) or when the runtime
+            # cannot run DI (JRuby, Ruby 2.5): in either case no component will
+            # run, so advertising bit 38 or the LIVE_DEBUGGING product would
+            # invite probe configs and an enable signal the tracer must refuse.
+            # When the env var is unset (default) on a supported runtime, DI is
+            # registered so RC can enable it.
+            if settings.respond_to?(:dynamic_instrumentation) &&
+                !Datadog::DI::Remote.explicitly_disabled?(settings) &&
+                Datadog::DI.supported_runtime?
               register_capabilities(Datadog::DI::Remote.capabilities)
               register_products(Datadog::DI::Remote.products)
               register_receivers(Datadog::DI::Remote.receivers(@telemetry))
+            end
 
-              # Symbol Database
-              if settings.respond_to?(:symbol_database) && settings.symbol_database.enabled
+            # Skip symbol database registration on runtimes that cannot run it
+            # (JRuby, Ruby < 2.7): DI supports Ruby 2.6 but Symbol Database does
+            # not, so advertising LIVE_DEBUGGING_SYMBOL_DB there would invite
+            # symbol-upload configs the component can never serve.
+            if settings.respond_to?(:symbol_database) && Datadog::SymbolDatabase.supported_runtime?
+              # Symbol database follows DI: when unset it advertises whenever DI
+              # advertises (mirror the DI branch above, including the unset/default
+              # case that RC may enable). An explicit symbol_database.enabled wins.
+              di_enabled = settings.respond_to?(:dynamic_instrumentation) &&
+                !Datadog::DI::Remote.explicitly_disabled?(settings)
+              if Datadog::SymbolDatabase.resolve_enabled(settings.symbol_database.enabled, di_enabled)
                 register_capabilities(Datadog::SymbolDatabase::Remote.capabilities)
                 register_products(Datadog::SymbolDatabase::Remote.products)
                 register_receivers(Datadog::SymbolDatabase::Remote.receivers(@telemetry))
@@ -53,10 +84,6 @@ module Datadog
               register_products(Datadog::OpenFeature::Remote.products)
               register_receivers(Datadog::OpenFeature::Remote.receivers(@telemetry))
             end
-
-            register_capabilities(Datadog::Tracing::Remote.capabilities)
-            register_products(Datadog::Tracing::Remote.products)
-            register_receivers(Datadog::Tracing::Remote.receivers(@telemetry))
           end
 
           def register_capabilities(capabilities)

data/lib/datadog/core/telemetry/event/app_started.rb

@@ -57,7 +57,14 @@ module Datadog
                 enabled: !!components.profiler,
               },
               dynamic_instrumentation: {
-                enabled: !!components.dynamic_instrumentation,
+                # With always-build, the DI component is constructed by
+                # default (when the runtime supports DI, RC is available, and
+                # DI is not explicitly disabled) regardless of whether the
+                # customer turned DI on. Its presence therefore no longer
+                # indicates the enabled state. started? reflects whether DI was
+                # actually enabled by env var or by implicit enablement via
+                # remote config.
+                enabled: components.dynamic_instrumentation&.started? || false,
               }
             }
 

data/lib/datadog/di/base.rb

@@ -8,6 +8,7 @@
 # are loaded, and also none of the rest of datadog library which also
 # has contrib code in other products.
 
+require_relative 'fatal_exceptions'
 require_relative 'code_tracker'
 
 # Needed since this file can be loaded without core
@@ -53,7 +54,8 @@ module Datadog
           # Activate code tracking by default because line trace points will not work
           # without it.
           Datadog::DI.activate_tracking!
-        rescue => exc
+        rescue Exception => exc # standard:disable Lint/RescueException
+          Datadog::DI.reraise_if_fatal(exc)
           if defined?(Datadog.logger)
             Datadog.logger.warn { "di: Failed to activate code tracking for DI: #{exc.class}: #{exc.message}" }
           else

data/lib/datadog/di/code_tracker.rb

@@ -3,6 +3,7 @@
 # rubocop:disable Lint/AssignmentInCondition
 
 require_relative 'error'
+require_relative 'fatal_exceptions'
 
 module Datadog
   module DI
@@ -120,7 +121,8 @@ module Datadog
           end
         end
         nil
-      rescue => exc
+      rescue Exception => exc # standard:disable Lint/RescueException
+        Datadog::DI.reraise_if_fatal(exc)
         # Backfill is best-effort — if it fails, line probes on
         # pre-loaded code won't work but everything else is unaffected.
         if component = DI.current_component
@@ -191,7 +193,8 @@ module Datadog
           # Since this method normally is called from customer applications,
           # rescue any exceptions that might not be handled to not break said
           # customer applications.
-          rescue => exc
+          rescue Exception => exc # standard:disable Lint/RescueException
+            Datadog::DI.reraise_if_fatal(exc)
             # Code tracker may be loaded without the rest of DI,
             # in which case DI.component will not yet be defined,
             # but we will have DI.current_component (set to nil).

data/lib/datadog/di/component.rb

@@ -1,5 +1,7 @@
 # frozen_string_literal: true
 
+require_relative 'fatal_exceptions'
+
 module Datadog
   module DI
     # Component for dynamic instrumentation.
@@ -15,45 +17,51 @@ module Datadog
     class Component
       class << self
         def build(settings, agent_settings, logger, telemetry: nil)
-          return unless settings.respond_to?(:dynamic_instrumentation) && settings.dynamic_instrumentation.enabled
+          return unless settings.respond_to?(:dynamic_instrumentation)
 
-          unless settings.respond_to?(:remote) && settings.remote.enabled
-            logger.warn("di: dynamic instrumentation could not be enabled because Remote Configuration Management is not available. To enable Remote Configuration, see https://docs.datadoghq.com/agent/remote_config")
+          # Explicit DD_DYNAMIC_INSTRUMENTATION_ENABLED=false: do not build the
+          # component at all. This is customer intent, not a failure — log at
+          # debug and emit no telemetry error. Capabilities#register skips the
+          # DI RC block under the same condition, so no enable signal arrives.
+          if Remote.explicitly_disabled?(settings)
+            logger.debug("di: dynamic instrumentation is explicitly disabled (DD_DYNAMIC_INSTRUMENTATION_ENABLED=false); not building component")
             return
           end
 
-          return unless environment_supported?(settings, logger)
+          reason = DI.unsupported_reason(settings)
+          if reason
+            # Log level mirrors customer intent: if the customer explicitly
+            # opted in via DD_DYNAMIC_INSTRUMENTATION_ENABLED, warn. Otherwise
+            # debug — with always-build, this path runs on every tracer boot
+            # for every customer, including those who never wanted DI. Spamming
+            # warnings to silent users (especially on JRuby or Ruby 2.5) would
+            # be noise. Customers who later trigger implicit enablement via the
+            # Datadog UI get a symmetric warn from Remote.handle_rc_enablement
+            # when the RC enable signal finds no component to start.
+            level = explicitly_enabled?(settings) ? :warn : :debug
+            logger.public_send(level, "di: dynamic instrumentation is disabled: #{reason}")
+            return
+          end
 
           new(settings, agent_settings, logger, code_tracker: DI.code_tracker, telemetry: telemetry).tap do |component|
             DI.add_current_component(component)
           end
         end
 
-        # Checks whether the runtime environment is supported by
-        # dynamic instrumentation. Currently we only require that, if Rails
-        # is used, that Rails environment is not development because
-        # DI does not currently support code unloading and reloading.
-        def environment_supported?(settings, logger)
-          # TODO add tests?
-          unless settings.dynamic_instrumentation.internal.development
-            if Datadog::Core::Environment::Execution.development?
-              logger.warn("di: development environment detected; not enabling dynamic instrumentation")
-              return false
-            end
-          end
-          if RUBY_ENGINE != 'ruby'
-            logger.warn("di: cannot enable dynamic instrumentation: MRI is required, but running on #{RUBY_ENGINE}")
-            return false
-          end
-          if RubyVersion.is?('< 2.6')
-            logger.warn("di: cannot enable dynamic instrumentation: Ruby 2.6+ is required, but running on #{RUBY_VERSION}")
-            return false
-          end
-          unless DI.respond_to?(:exception_message)
-            logger.warn("di: cannot enable dynamic instrumentation: C extension is not available")
-            return false
-          end
-          true
+        # True when the customer explicitly set
+        # DD_DYNAMIC_INSTRUMENTATION_ENABLED=true (or its equivalent in code).
+        # Symmetric to {Remote.explicitly_disabled?}.
+        #
+        # Uses {Datadog::Core::Configuration::Options::InstanceMethods#using_default?}
+        # rather than `options[:enabled].default_precedence?` because the option
+        # hash is populated lazily on first access; reading the underlying option
+        # before {Component.build} touches the value would NoMethodError on nil.
+        #
+        # @param settings [Datadog::Core::Configuration::Settings]
+        # @return [Boolean]
+        def explicitly_enabled?(settings)
+          !settings.dynamic_instrumentation.using_default?(:enabled) &&
+            settings.dynamic_instrumentation.enabled
         end
       end
 
@@ -80,7 +88,11 @@ module Datadog
           settings, instrumenter, probe_notification_builder, probe_notifier_worker, logger, probe_repository,
           telemetry: telemetry,
         )
-        probe_notifier_worker.start
+        # @started transitions are serialized by @lifecycle_mutex so that
+        # concurrent RC callbacks (which run on the remote-config thread)
+        # cannot race a foreground start! with a background stop!.
+        @lifecycle_mutex = Mutex.new
+        @started = false
       end
 
       attr_reader :settings
@@ -96,9 +108,61 @@ module Datadog
       attr_reader :redactor
       attr_reader :serializer
 
-      # Shuts down dynamic instrumentation.
+      # Starts the DI component: begins accepting probes and
+      # processing snapshots.
+      #
+      # Starts the probe notifier worker thread before enabling the
+      # definition trace point, so any future status emission from a
+      # trace-point-driven installation has a worker to drain it.
+      # Today {ProbeManager#reopen} re-hooks without emitting statuses, so
+      # the order is defensive rather than load-bearing. No-op if already
+      # started. Serialized by @lifecycle_mutex.
+      #
+      # @return [void]
+      def start!
+        @lifecycle_mutex.synchronize do
+          return if @started
+
+          probe_notifier_worker.start
+          probe_manager.reopen
+          @started = true
+        end
+      end
+
+      # Stops the DI component: removes all probes and stops
+      # background threads.
+      #
+      # The component remains alive and can be restarted with {#start!}.
+      # Does not clear out the code tracker.
+      # No-op if already stopped. Serialized by @lifecycle_mutex.
+      #
+      # @return [void]
+      def stop!
+        @lifecycle_mutex.synchronize do
+          return unless @started
+
+          probe_manager.stop
+          probe_notifier_worker.stop
+          @started = false
+        end
+      end
+
+      # Whether the component is currently started.
+      #
+      # Read by remote config dispatch to decide whether to apply probe
+      # changes (changes received while stopped are dropped, since the
+      # next start! will reconcile from the latest RC state).
+      #
+      # @return [Boolean] true if start! has been called and stop! has not
+      def started?
+        @started
+      end
+
+      # Shuts down dynamic instrumentation permanently.
       #
       # Removes all code hooks and stops background threads.
+      # Called by Components#shutdown! during component destruction.
+      # Unlike {#stop!}, this is not reversible.
       #
       # Does not clear out the code tracker, because it's only populated
       # by code when code is compiled and therefore, if the code tracker
@@ -107,21 +171,31 @@ module Datadog
       def shutdown!(replacement = nil)
         DI.remove_current_component(self)
 
-        probe_manager.clear_hooks
-        probe_manager.close
-        probe_notifier_worker.stop
+        # Hold the lifecycle mutex so all transitions of @started are
+        # serialized — start! / stop! / shutdown! cannot interleave with
+        # one another. Without the mutex an in-flight stop! from an RC
+        # callback could complete after shutdown!'s probe_manager.close,
+        # producing an inconsistent state.
+        @lifecycle_mutex.synchronize do
+          @started = false
+          probe_manager.clear_hooks
+          probe_manager.close
+          probe_notifier_worker.stop
+        end
       end
 
       def parse_probe_spec_and_notify(probe_spec)
         probe = ProbeBuilder.build_from_remote_config(probe_spec)
-      rescue => exc
+      rescue Exception => exc # standard:disable Lint/RescueException
+        Datadog::DI.reraise_if_fatal(exc)
         begin
           probe = Struct.new(:id).new(
             probe_spec['id'],
           )
           payload = probe_notification_builder.build_errored(probe, exc)
           probe_notifier_worker.add_status(payload)
-        rescue => nested_exc
+        rescue Exception => nested_exc # standard:disable Lint/RescueException
+          Datadog::DI.reraise_if_fatal(nested_exc)
           logger.debug { "di: failed to build error notification: #{nested_exc.class}: #{nested_exc.message}" }
           telemetry&.report(nested_exc, description: 'Error building probe error notification')
           raise

data/lib/datadog/di/error.rb

@@ -18,6 +18,10 @@ module Datadog
       class MissingLineNumber < Error
       end
 
+      # Probe does not contain a method name (i.e., is not a method probe).
+      class MissingMethodName < Error
+      end
+
       # Failed to communicate to the local Datadog agent (e.g. to send
       # probe status or a snapshot).
       class AgentCommunicationError < Error
@@ -47,6 +51,12 @@ module Datadog
       class AlreadyInstrumented < Error
       end
 
+      # Raised when a probe targets a class or module that the tracer
+      # refuses to instrument (for example, anything in the Datadog
+      # namespace, which the tracer itself uses internally).
+      class ProbeTargetForbidden < Error
+      end
+
       # Raised when installing a line probe and multiple files match the
       # specified path suffix.
       # A probe must be installed into one file only, since UI only

data/lib/datadog/di/fatal_exceptions.rb

@@ -0,0 +1,26 @@
+# frozen_string_literal: true
+
+# Defined under DI rather than Core because the DI preload path
+# (datadog/di/preload -> base) activates code tracking before the rest of the
+# library, including Datadog::Core, is loaded. Catch-all rescues on that path
+# must be able to re-raise fatal exceptions without referencing Core.
+
+module Datadog
+  module DI
+    # Exception classes that catch-all rescues must never swallow: they signal
+    # that the process is being torn down or has run out of memory and have to
+    # propagate. SignalException covers Interrupt (Ctrl+C) and every other
+    # signal delivered as an exception.
+    FATAL_EXCEPTION_CLASSES = [SystemExit, SignalException, NoMemoryError].freeze
+
+    # Re-raise +exc+ when it is fatal (see FATAL_EXCEPTION_CLASSES). Call this as
+    # the first statement of a `rescue Exception` handler so that fatal
+    # conditions are not accidentally swallowed by a broad rescue.
+    #
+    # @param exc [Exception] the currently-handled exception
+    # @return [nil] when +exc+ is not fatal; otherwise re-raises +exc+
+    def self.reraise_if_fatal(exc)
+      raise exc if FATAL_EXCEPTION_CLASSES.any? { |klass| exc.is_a?(klass) }
+    end
+  end
+end