datadog 2.43.0 → 2.45.0

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

@@ -757,12 +757,15 @@ module Datadog
                 val ||= ""
                 # maps OpenTelemetry semantic attributes to Datadog tags
                 key = case key.downcase
-                when "deployment.environment" then "env"
+                when "deployment.environment.name"
+                  result["env"] = val
+                  nil
+                when "deployment.environment" then "env" unless result.key?("env")
                 when "service.version" then "version"
                 when "service.name" then "service"
                 else key
                 end
-                result[key] = val unless key.empty?
+                result[key] = val unless key.nil? || key.empty?
               end
             end
             result

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

@@ -14,6 +14,7 @@ module Datadog
           "DD_AI_GUARD_ENDPOINT",
           "DD_AI_GUARD_MAX_CONTENT_SIZE",
           "DD_AI_GUARD_MAX_MESSAGES_LENGTH",
+          "DD_AI_GUARD_REDACTION_ENABLED",
           "DD_AI_GUARD_TIMEOUT",
           "DD_API_KEY",
           "DD_API_SECURITY_DOWNSTREAM_BODY_ANALYSIS_SAMPLE_RATE",
@@ -60,10 +61,16 @@ module Datadog
           "DD_ERROR_TRACKING_HANDLED_ERRORS",
           "DD_ERROR_TRACKING_HANDLED_ERRORS_INCLUDE",
           "DD_EXPERIMENTAL_FLAGGING_PROVIDER_ENABLED",
+          "DD_EXPERIMENTAL_FLAGGING_PROVIDER_INITIALIZATION_TIMEOUT_MS",
           "DD_EXPERIMENTAL_FLAGGING_PROVIDER_SPAN_ENRICHMENT_ENABLED",
           "DD_EXPERIMENTAL_NATIVE_TRANSPORT_ENABLED",
           "DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED",
           "DD_EXTERNAL_ENV",
+          "DD_FEATURE_FLAGS_CONFIGURATION_SOURCE",
+          "DD_FEATURE_FLAGS_CONFIGURATION_SOURCE_AGENTLESS_BASE_URL",
+          "DD_FEATURE_FLAGS_CONFIGURATION_SOURCE_AGENTLESS_POLL_INTERVAL_SECONDS",
+          "DD_FEATURE_FLAGS_CONFIGURATION_SOURCE_AGENTLESS_REQUEST_TIMEOUT_SECONDS",
+          "DD_FEATURE_FLAGS_ENABLED",
           "DD_FLAGGING_EVALUATION_COUNTS_ENABLED",
           "DD_GIT_COMMIT_SHA",
           "DD_GIT_REPOSITORY_URL",
@@ -266,6 +273,7 @@ module Datadog
           "DD_TRACE_OPENSEARCH_PEER_SERVICE",
           "DD_TRACE_OPENSEARCH_RESOURCE_PATTERN",
           "DD_TRACE_OPENSEARCH_SERVICE_NAME",
+          "DD_TRACE_OTEL_CTX_ENABLED",
           "DD_TRACE_PEER_SERVICE_DEFAULTS_ENABLED",
           "DD_TRACE_PEER_SERVICE_MAPPING",
           "DD_TRACE_PG_ANALYTICS_ENABLED",

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

@@ -16,14 +16,15 @@ module Datadog
         class Capabilities
           attr_reader :capabilities, :receivers, :base64_capabilities
 
-          def initialize(settings, telemetry)
+          def initialize(settings, telemetry, open_feature_component_provider: nil)
+            open_feature_component_provider ||= -> {}
             @capabilities = []
             @products = []
             @products_mutex = Mutex.new
             @receivers = []
             @telemetry = telemetry
 
-            register(settings)
+            register(settings, open_feature_component_provider)
 
             @base64_capabilities = capabilities_to_base64
           end
@@ -46,7 +47,7 @@ module Datadog
 
           private
 
-          def register(settings)
+          def register(settings, open_feature_component_provider)
             if settings.respond_to?(:appsec) && settings.appsec.enabled
               register_capabilities(Datadog::AppSec::Remote.capabilities)
               register_products(Datadog::AppSec::Remote.products)
@@ -98,7 +99,12 @@ module Datadog
             if settings.respond_to?(:open_feature) && settings.open_feature.enabled
               register_capabilities(Datadog::OpenFeature::Remote.capabilities)
               register_products(Datadog::OpenFeature::Remote.products)
-              register_receivers(Datadog::OpenFeature::Remote.receivers(@telemetry))
+              register_receivers(
+                Datadog::OpenFeature::Remote.receivers(
+                  @telemetry,
+                  component_provider: open_feature_component_provider,
+                ),
+              )
             end
           end
 

data/lib/datadog/core/remote/component.rb

@@ -96,6 +96,8 @@ module Datadog
         # Recreates the remote configuration client after a fork.
         # This ensures each forked process has a unique client ID and fresh state.
         def after_fork
+          @worker.after_fork
+          @barrier = Barrier.new(@settings.remote.boot_timeout_seconds)
           @client = Client.new(@transport, @capabilities, settings: @settings, logger: @logger)
           @healthy = false
           logger.debug { "remote configuration client recreated after fork: #{@client.id} products: #{@capabilities.products.sort.join(", ")}" }
@@ -187,10 +189,15 @@ module Datadog
           #
           # Those checks are instead performed inside the worker loop.
           # This allows users to upgrade their agent while keeping their application running.
-          def build(settings, agent_settings, logger:, telemetry:)
+          def build(settings, agent_settings, logger:, telemetry:, open_feature_component_provider: nil)
             return unless settings.remote.enabled
 
-            new(settings, Client::Capabilities.new(settings, telemetry), agent_settings, logger: logger)
+            capabilities = Client::Capabilities.new(
+              settings,
+              telemetry,
+              open_feature_component_provider: open_feature_component_provider,
+            )
+            new(settings, capabilities, agent_settings, logger: logger)
           end
         end
       end

data/lib/datadog/core/remote/worker.rb

@@ -70,6 +70,14 @@ module Datadog
           @started
         end
 
+        # Child-only: discard inherited synchronization without starting a thread.
+        def after_fork
+          @mutex = Mutex.new
+          @thr = nil
+          @starting = false
+          @started = false
+        end
+
         private
 
         def poll(interval)

data/lib/datadog/di/instrumenter.rb

@@ -538,7 +538,7 @@ module Datadog
                 # the context, we won't be able to report anything as
                 # the probe notifier builder requires a context.
                 begin
-                  responder.probe_condition_evaluation_failed_callback(context, exc)
+                  responder.probe_condition_evaluation_failed_callback(context, condition, exc)
                 rescue Exception => nested_exc # standard:disable Lint/RescueException
                   Datadog::DI.reraise_if_fatal(nested_exc)
                   raise if settings.dynamic_instrumentation.internal.propagate_all_exceptions

data/lib/datadog/di/probe_manager.rb

@@ -310,17 +310,22 @@ module Datadog
       #
       # This can happen when the expression references undefined variables,
       # has type mismatches, or encounters runtime errors during evaluation.
-      # Rate-limited to 1 notification per second per probe to avoid
-      # flooding the backend when conditions fail repeatedly.
+      # Rate-limited to 1 notification per second per probe, and subject to
+      # the process-wide global snapshot rate limit, to avoid flooding the
+      # backend when conditions fail repeatedly.
       #
       # @param context [Context] The execution context containing probe and captured data
-      # @param expr [String] The condition expression that failed
+      # @param expr [EL::Expression] The condition expression that failed
       # @param exc [Exception] The exception raised during condition evaluation
       def probe_condition_evaluation_failed_callback(context, expr, exc)
         probe = context.probe
         if probe.condition_evaluation_failed_rate_limiter&.allow?
-          payload = probe_notification_builder.build_condition_evaluation_failed(context, expr, exc)
-          probe_notifier_worker.add_snapshot(payload)
+          if instrumenter.global_snapshot_rate_limiter.allow?
+            payload = probe_notification_builder.build_condition_evaluation_failed(context, expr, exc)
+            probe_notifier_worker.add_snapshot(payload)
+          else
+            logger.trace { "di: #{probe.type} probe #{probe.id}: skipping condition error snapshot due to global rate limit" }
+          end
         end
       end
 

data/lib/datadog/di/proc_responder.rb

@@ -20,7 +20,7 @@ module Datadog
         executed_proc.call(context)
       end
 
-      def probe_condition_evaluation_failed_callback(context, exc)
+      def probe_condition_evaluation_failed_callback(context, _expr, exc)
         if failed_proc.nil?
           raise NotImplementedError, "Failed proc not provided"
         end

data/lib/datadog/open_feature/agentless/configuration_source.rb

@@ -0,0 +1,246 @@
+# frozen_string_literal: true
+
+require_relative "../../core/workers/polling"
+require_relative "../evaluation_engine"
+require_relative "parser"
+require_relative "transport"
+
+module Datadog
+  module OpenFeature
+    module Agentless
+      # Polls the agentless endpoint and applies accepted configuration.
+      class ConfigurationSource
+        MAX_ATTEMPTS = 3
+        FIRST_RETRY_MIN_SECONDS = 2.0
+        FIRST_RETRY_MAX_SECONDS = 10.0
+        SECOND_RETRY_MIN_SECONDS = 5.0
+        SECOND_RETRY_MAX_SECONDS = 30.0
+        RETRY_JITTER = 0.2
+
+        include Core::Workers::Polling
+
+        attr_reader :etag
+
+        def self.build(settings, endpoint:, apply:, logger: Datadog.logger)
+          api_key = settings.api_key.to_s.strip
+          if endpoint.managed? && api_key.empty?
+            logger.warn("Feature Flags agentless delivery requires DD_API_KEY; agentless delivery is disabled")
+            return
+          end
+
+          new(
+            endpoint: endpoint,
+            api_key: api_key.empty? ? nil : api_key,
+            poll_interval_seconds: settings.feature_flags.agentless.poll_interval_seconds,
+            request_timeout_seconds: settings.feature_flags.agentless.request_timeout_seconds,
+            apply: apply,
+            logger: logger,
+          )
+        end
+
+        def initialize(
+          endpoint:,
+          api_key:,
+          poll_interval_seconds:,
+          request_timeout_seconds:,
+          apply:,
+          logger:,
+          transport: nil,
+          random: nil,
+          retry_wait: nil
+        )
+          @transport = transport || Transport.new(
+            endpoint: endpoint,
+            api_key: api_key,
+            timeout_seconds: request_timeout_seconds,
+          )
+          @apply = apply
+          @logger = logger
+          @poll_interval_seconds = poll_interval_seconds
+          @random = random || -> { Kernel.rand }
+          @retry_wait = retry_wait || ->(delay) { wait_for_retry(delay) }
+          @etag = nil
+          @warned = {}
+          @lifecycle_mutex = Mutex.new
+          @started = false
+          @stopped = false
+
+          self.enabled = true
+          self.fork_policy = Core::Workers::Async::Thread::FORK_POLICY_RESTART
+          self.loop_base_interval = poll_interval_seconds
+        end
+
+        def start
+          @lifecycle_mutex.synchronize do
+            return false if @stopped
+            return true if @started && !forked?
+
+            perform
+            @started = true
+          end
+
+          true
+        end
+
+        def stop(force_stop = true, timeout = Core::Workers::Polling::DEFAULT_SHUTDOWN_TIMEOUT)
+          @lifecycle_mutex.synchronize do
+            return false if @stopped
+
+            @stopped = true
+            self.enabled = false
+          end
+
+          super
+          true
+        end
+
+        def perform
+          poll
+        # An exception escaping here ends the unsupervised worker thread and freezes configuration.
+        rescue => error
+          warn_once(:unexpected) do
+            @logger.error("Feature Flags agentless poll failed with #{error.class}; polling continues")
+          end
+        end
+
+        def poll
+          attempt = 1
+
+          while attempt <= MAX_ATTEMPTS
+            return if stopped?
+
+            response = @transport.get(@etag)
+            unless retryable?(response)
+              apply_response(response)
+              return
+            end
+
+            if attempt == MAX_ATTEMPTS
+              warn_failure(response, attempt)
+              return
+            end
+
+            return if wait_before_retry(retry_delay(attempt))
+
+            attempt += 1
+          end
+        end
+
+        protected
+
+        def after_fork
+          # The child must report its own failures, while the inherited ETag still matches its configuration.
+          @warned = {}
+          nil
+        end
+
+        private
+
+        def stopped?
+          @lifecycle_mutex.synchronize { @stopped }
+        end
+
+        def retryable?(response)
+          status = response.status
+          return true unless status
+
+          retryable_status?(status)
+        end
+
+        def retryable_status?(status)
+          status == 408 || status == 429 || status.between?(500, 599)
+        end
+
+        def retry_delay(attempt)
+          base = if attempt == 1
+            clamp(@poll_interval_seconds / 6.0, FIRST_RETRY_MIN_SECONDS, FIRST_RETRY_MAX_SECONDS)
+          else
+            clamp(@poll_interval_seconds / 3.0, SECOND_RETRY_MIN_SECONDS, SECOND_RETRY_MAX_SECONDS)
+          end
+          random = @random.call
+          jitter = 1 - RETRY_JITTER + (random * RETRY_JITTER * 2)
+          [1.0, base * jitter].max
+        end
+
+        def clamp(value, minimum, maximum)
+          value.clamp(minimum, maximum)
+        end
+
+        def wait_before_retry(delay)
+          @retry_wait.call(delay)
+        end
+
+        def wait_for_retry(delay)
+          mutex.synchronize do
+            return true if stopped?
+
+            shutdown.wait(mutex, delay)
+          end
+
+          stopped? || !run_loop?
+        end
+
+        def apply_response(response)
+          case response.status
+          when 200
+            apply_configuration(response)
+          when 304
+            nil
+          when 401, 403
+            warn_once(:authentication) do
+              @logger.warn(
+                "Feature Flags agentless endpoint returned HTTP #{response.status}; verify endpoint authentication"
+              )
+            end
+          else
+            warn_failure(response, 1)
+          end
+        end
+
+        def apply_configuration(response)
+          configuration = response.body && Parser.parse(response.body)
+          unless configuration
+            warn_once(:malformed) do
+              @logger.error("Feature Flags agentless endpoint returned malformed UFC payload")
+            end
+            return
+          end
+
+          @apply.call(configuration)
+          etag = response.etag.to_s.strip
+          @etag = etag.empty? ? nil : etag
+        rescue EvaluationEngine::ReconfigurationError => error
+          warn_once(:application) do
+            @logger.warn("Feature Flags agentless UFC payload could not be applied: #{error.class}: #{error.message}")
+          end
+        end
+
+        def warn_failure(response, attempts)
+          status = response.status
+          category = if status.nil?
+            :request
+          elsif retryable_status?(status)
+            :http_retryable
+          else
+            :http_unexpected
+          end
+          warn_once(category) do
+            if status
+              @logger.warn("Feature Flags agentless endpoint returned HTTP #{status} after #{attempts} attempt(s)")
+            else
+              error_class = response.error ? response.error.class : StandardError
+              @logger.warn("Feature Flags agentless request failed after #{attempts} attempt(s): #{error_class}")
+            end
+          end
+        end
+
+        def warn_once(category)
+          return if @warned[category]
+
+          @warned[category] = true
+          yield
+        end
+      end
+    end
+  end
+end

data/lib/datadog/open_feature/agentless/parser.rb

@@ -0,0 +1,43 @@
+# frozen_string_literal: true
+
+require "json"
+
+module Datadog
+  module OpenFeature
+    module Agentless
+      # Parses the JSON:API envelope returned by agentless delivery.
+      module Parser
+        RESOURCE_TYPE = "universal-flag-configuration"
+
+        module_function
+
+        def parse(body)
+          payload = JSON.parse(body)
+          return unless payload.is_a?(Hash)
+
+          data = payload["data"]
+          return unless data.is_a?(Hash) && data["type"] == RESOURCE_TYPE
+
+          attributes = data["attributes"]
+          return unless valid_attributes?(attributes)
+
+          JSON.generate(attributes)
+        rescue JSON::ParserError, JSON::GeneratorError, TypeError
+          nil
+        end
+
+        def valid_attributes?(attributes)
+          return false unless attributes.is_a?(Hash)
+
+          environment = attributes["environment"]
+          attributes["format"].is_a?(String) &&
+            attributes["createdAt"].is_a?(String) &&
+            environment.is_a?(Hash) &&
+            environment["name"].is_a?(String) &&
+            attributes["flags"].is_a?(Hash)
+        end
+        private_class_method :valid_attributes?
+      end
+    end
+  end
+end

data/lib/datadog/open_feature/agentless/response.rb

@@ -0,0 +1,19 @@
+# frozen_string_literal: true
+
+module Datadog
+  module OpenFeature
+    module Agentless
+      # Result of one agentless configuration request.
+      class Response
+        attr_reader :status, :etag, :body, :error
+
+        def initialize(status: nil, etag: nil, body: nil, error: nil)
+          @status = status
+          @etag = etag
+          @body = body
+          @error = error
+        end
+      end
+    end
+  end
+end

data/lib/datadog/open_feature/agentless/transport.rb

@@ -0,0 +1,103 @@
+# frozen_string_literal: true
+
+require "net/http"
+require "stringio"
+require "timeout"
+require "uri"
+require "zlib"
+
+require_relative "../../core/environment/identity"
+require_relative "../../core/transport/ext"
+require_relative "response"
+
+module Datadog
+  module OpenFeature
+    module Agentless
+      # HTTP transport for agentless configuration delivery.
+      class Transport
+        def initialize(endpoint:, api_key:, timeout_seconds:)
+          @endpoint = endpoint
+          @api_key = endpoint.managed? ? api_key : nil
+          @timeout_seconds = timeout_seconds
+        end
+
+        def get(etag)
+          request = Net::HTTP::Get.new(@endpoint.uri.request_uri, headers(etag))
+          apply_custom_authentication(request)
+          response = request(request)
+          status = response.code.to_i
+
+          Response.new(
+            status: status,
+            etag: response["ETag"],
+            # Decoding another status can erase the status the poller needs to classify the response.
+            body: (status == 200) ? response_body(response) : nil,
+          )
+        # Net::HTTP uses several version-specific exception types. None may end
+        # the delivery worker; the poller classifies every transport failure.
+        rescue => error
+          Response.new(error: error)
+        end
+
+        private
+
+        def apply_custom_authentication(request)
+          return if @endpoint.managed?
+
+          uri = @endpoint.uri
+          username = uri.user
+          return unless username
+
+          request.basic_auth(
+            URI::RFC2396_PARSER.unescape(username),
+            URI::RFC2396_PARSER.unescape(uri.password.to_s),
+          )
+        end
+
+        def headers(etag)
+          headers = {
+            "Accept-Encoding" => "gzip",
+            "DD-Client-Library-Language" => Core::Environment::Identity.lang,
+            "DD-Client-Library-Version" => Core::Environment::Identity.gem_datadog_version_semver2,
+            Core::Transport::Ext::HTTP::HEADER_DD_INTERNAL_UNTRACED_REQUEST => "1",
+          }
+          api_key = @api_key
+          headers[Core::Transport::Ext::HTTP::HEADER_DD_API_KEY] = api_key if api_key
+          headers["If-None-Match"] = etag if etag
+          headers
+        end
+
+        def request(request)
+          uri = @endpoint.uri
+          hostname = uri.hostname
+          raise ArgumentError, "Feature Flags agentless endpoint must have a host" unless hostname
+
+          # Agentless delivery requires public egress, so use Ruby's standard proxy discovery.
+          http = Net::HTTP.new(hostname, uri.port)
+          http.max_retries = 0
+          http.use_ssl = uri.scheme == "https"
+          http.open_timeout = @timeout_seconds
+          http.read_timeout = @timeout_seconds
+          http.write_timeout = @timeout_seconds if http.respond_to?(:write_timeout=)
+
+          # Bounds total request time, including DNS resolution, which the per-phase timeouts do not cover.
+          Timeout.timeout(@timeout_seconds) do
+            http.start { |connection| connection.request(request) }
+          end
+        end
+
+        def response_body(response)
+          body = response.body.to_s
+          return body unless response["Content-Encoding"].to_s.strip.casecmp("gzip") == 0
+
+          reader = Zlib::GzipReader.new(StringIO.new(body))
+          begin
+            reader.read.to_s
+          ensure
+            reader.close
+          end
+        end
+      end
+    end
+  end
+end