datadog 2.39.0 → 2.47.0

data/lib/datadog/open_feature/activation.rb

@@ -0,0 +1,268 @@
+# frozen_string_literal: true
+
+require_relative "agentless/configuration_source"
+require_relative "component"
+require_relative "configuration/agentless_endpoint"
+require_relative "configuration/source"
+require_relative "remote"
+
+module Datadog
+  module OpenFeature
+    # Owns eager Remote Configuration and lazy agentless delivery activation.
+    class Activation
+      attr_reader :component, :failure
+
+      def initialize(settings, agent_settings, remote, logger:, telemetry:)
+        @settings = settings
+        @agent_settings = agent_settings
+        @remote = remote
+        @logger = logger
+        @telemetry = telemetry
+        @component = nil
+        @configuration_source = nil
+        @delivery_source = nil
+        @remote_receivers = nil
+        @failure = nil
+        @providers = {}.compare_by_identity
+        @activated = false
+        @delivery_started = false
+        @shutdown = false
+        @mutex = Mutex.new
+      end
+
+      def activate(provider)
+        @mutex.synchronize do
+          return if @shutdown
+          return unless open_feature_available?
+
+          @providers[provider] = true
+          return @component if @activated && @delivery_started
+          return if @activated
+
+          resolution = Configuration::Source.resolve(@settings, logger: @logger)
+          activate_delivery(resolution)
+        end
+      end
+
+      def startup!
+        start!
+        OpenFeature.reattach(self)
+      rescue => e
+        # Feature Flags is optional and must never interrupt library startup.
+        report_failure(e, "Feature Flags delivery failed to start")
+      end
+
+      def start!
+        @mutex.synchronize do
+          return if @shutdown || @activated
+          return unless open_feature_available?
+
+          resolution = Configuration::Source.resolve(@settings, logger: @logger)
+          return unless resolution.enabled? && resolution.source == Configuration::Source::REMOTE_CONFIG
+
+          activate_delivery(resolution)
+        end
+      end
+
+      def activated?
+        @mutex.synchronize { @activated }
+      end
+
+      def providers
+        @mutex.synchronize { @providers.keys }
+      end
+
+      def after_fork
+        inherited_providers, component = @mutex.synchronize do
+          @shutdown ? [[], nil] : [@providers.keys, @component]
+        end
+        inherited_providers.each { |provider| provider.send(:after_fork, component) }
+
+        configuration_source = @mutex.synchronize do
+          if @shutdown || !@delivery_started
+            nil
+          else
+            (@delivery_source == Configuration::Source::REMOTE_CONFIG) ? @remote : @configuration_source
+          end
+        end
+
+        configuration_source&.start
+        nil
+      rescue => e
+        # Feature Flags is optional and must never interrupt other post-fork handlers.
+        report_failure(e, "Feature Flags delivery failed to restart after fork")
+      end
+
+      def deactivate(provider)
+        configuration_source, component = @mutex.synchronize do
+          return if @shutdown
+          return unless @providers.delete(provider)
+          return unless @providers.empty?
+          if @delivery_source == Configuration::Source::REMOTE_CONFIG && @delivery_started
+            # Remote Configuration is process-scoped and may already hold configuration needed by the next provider.
+            [nil, nil]
+          else
+            previous_source = @configuration_source
+            previous_component = @component
+            @component = nil
+            @configuration_source = nil
+            @delivery_source = nil
+            @activated = false
+            @delivery_started = false
+            @failure = nil
+            [previous_source, previous_component]
+          end
+        end
+
+        configuration_source&.stop
+        component&.shutdown!
+      end
+
+      def shutdown!
+        configuration_source, component = @mutex.synchronize do
+          return if @shutdown
+
+          @shutdown = true
+          [@configuration_source, @component]
+        end
+
+        configuration_source&.stop
+        configuration_received = @mutex.synchronize do
+          !@providers.empty? && !!component&.configuration_received?
+        end
+        configuration_changed(Component::CONFIGURATION_LOST) if configuration_received
+        component&.shutdown!
+      end
+
+      private
+
+      def report_failure(error, description)
+        @logger.error("#{description}: #{error.class}: #{error.message}")
+        @telemetry.report(error, description: description)
+      end
+
+      def open_feature_available?
+        @settings.respond_to?(:open_feature) && @settings.respond_to?(:feature_flags)
+      end
+
+      def activate_delivery(resolution)
+        @activated = true
+        unless resolution.enabled?
+          @failure = "Feature Flags are disabled"
+          return
+        end
+
+        component = Component.build(
+          @settings,
+          @agent_settings,
+          resolution: resolution,
+          on_configuration_change: ->(event) { configuration_changed(event) },
+          logger: @logger,
+          telemetry: @telemetry,
+        )
+        unless component
+          @failure = "Feature Flags are unavailable on this runtime"
+          return
+        end
+
+        @component = component
+        @delivery_source = resolution.source
+        @delivery_started = start_delivery(resolution.source)
+        return component if @delivery_started
+
+        component.shutdown!
+        @component = nil
+        @delivery_source = nil
+        nil
+      rescue
+        configuration_source = @configuration_source
+        failed_component = @component
+        @component = nil
+        @configuration_source = nil
+        @delivery_source = nil
+        @activated = false
+        @delivery_started = false
+        @failure = nil
+        begin
+          configuration_source&.stop
+        ensure
+          failed_component&.shutdown!
+        end
+        raise
+      end
+
+      def start_delivery(source)
+        case source
+        when Configuration::Source::AGENTLESS
+          start_agentless
+        when Configuration::Source::REMOTE_CONFIG
+          start_remote_configuration
+        else
+          @failure = "Feature Flags configuration source is unavailable"
+          false
+        end
+      end
+
+      def start_agentless
+        component = @component
+        return false unless component
+
+        endpoint = Configuration::AgentlessEndpoint.build(
+          site: @settings.site,
+          environment: @settings.env,
+          base_url: @settings.feature_flags.agentless.base_url,
+          logger: @logger,
+        )
+        unless endpoint
+          @failure = "Feature Flags agentless endpoint is unavailable"
+          return false
+        end
+
+        configuration_source = Agentless::ConfigurationSource.build(
+          @settings,
+          endpoint: endpoint,
+          apply: ->(configuration) { component.reconfigure!(configuration) },
+          logger: @logger,
+        )
+        unless configuration_source
+          @failure = "Feature Flags agentless delivery could not start"
+          return false
+        end
+
+        @configuration_source = configuration_source
+        started = configuration_source.start
+        @failure = "Feature Flags agentless delivery could not start" unless started
+        started
+      end
+
+      def start_remote_configuration
+        remote = @remote
+        unless remote
+          @failure = "Feature Flags Remote Configuration is unavailable"
+          @logger.warn(
+            "#{@failure}. To enable Remote Configuration, " \
+              "see https://docs.datadoghq.com/remote_configuration/."
+          )
+          return false
+        end
+
+        receivers = @remote_receivers ||= Remote.receivers(
+          @telemetry,
+          component_provider: -> { @mutex.synchronize { @component } },
+        )
+        remote.register(
+          capabilities: Remote.capabilities,
+          products: Remote.products,
+          receivers: receivers,
+        )
+        remote.start
+        true
+      end
+
+      def configuration_changed(event)
+        providers = @mutex.synchronize { @providers.keys }
+        providers.each { |provider| provider.send(:configuration_changed, event) }
+      end
+    end
+  end
+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