datadog 2.15.0 → 2.17.0

data/lib/datadog/core/utils/only_once_successful.rb

@@ -5,14 +5,25 @@ require_relative 'only_once'
 module Datadog
   module Core
     module Utils
-      # Helper class to execute something with only one success.
+      # Helper class to execute something with only one successful execution.
       #
-      # This is useful for cases where we want to ensure that a block of code is only executed once, and only if it
-      # succeeds. One such example is sending app-started telemetry event.
+      # If limit is not provided to the constructor, +run+ will execute the
+      # block an unlimited number of times until the block indicates that it
+      # executed successfully by returning a truthy value. After a block
+      # executes successfully, subsequent +run+ calls will not invoke the
+      # block.
       #
-      # Successful execution is determined by the return value of the block: any truthy value is considered success.
+      # If a non-zero limit is provided to the constructor, +run+ will
+      # execute the block up to that many times, and will mark the instance
+      # of OnlyOneSuccessful as failed if none of the executions succeeded.
       #
-      # Thread-safe when used correctly (e.g. be careful of races when lazily initializing instances of this class).
+      # One consumer of this class is sending the app-started telemetry event.
+      #
+      # Successful execution is determined by the return value of the block:
+      # any truthy value is considered success.
+      #
+      # This class is thread-safe (however, instances of it must also be
+      # created in a thread-safe manner).
       #
       # Note: In its current state, this class is not Ractor-safe.
       # In https://github.com/DataDog/dd-trace-rb/pull/1398#issuecomment-797378810 we have a discussion of alternatives,

data/lib/datadog/core/utils/time.rb

@@ -31,6 +31,16 @@ module Datadog
         #
         # @param block [Proc] block that returns a `Time` object representing the current wall time
         def now_provider=(block)
+          class << self
+            # Avoid method redefinition warning.
+            # `rescue nil` is added in case customers remove the method
+            # themselves to squelch the warning.
+            begin
+              remove_method(:now)
+            rescue
+              nil
+            end
+          end
           define_singleton_method(:now, &block)
         end
 
@@ -43,6 +53,16 @@ module Datadog
         #
         # @param block [Proc] block that accepts unit and returns timestamp in the requested unit
         def get_time_provider=(block)
+          class << self
+            # Avoid method redefinition warning
+            # `rescue nil` is added in case customers remove the method
+            # themselves to squelch the warning.
+            begin
+              remove_method(:get_time)
+            rescue
+              nil
+            end
+          end
           define_singleton_method(:get_time, &block)
         end
 

data/lib/datadog/core/utils/truncation.rb

@@ -0,0 +1,21 @@
+# frozen_string_literal: true
+
+module Datadog
+  module Core
+    module Utils
+      # Helper methods for truncating data
+      module Truncation
+        module_function
+
+        def truncate_in_middle(string, max_prefix_length, max_suffix_length)
+          max_length = max_prefix_length + 3 + max_suffix_length
+          if string.length > max_length
+            "#{string[0...max_prefix_length]}...#{string[-max_suffix_length..-1]}"
+          else
+            string
+          end
+        end
+      end
+    end
+  end
+end

data/lib/datadog/core/vendor/multipart-post/multipart/post/composite_read_io.rb

@@ -108,7 +108,7 @@ module Datadog
             end
 
             def respond_to?(meth, include_all = false)
-              @io.respond_to?(meth, include_all) || super(meth, include_all)
+              @io.respond_to?(meth, include_all) || super
             end
           end
         end

data/lib/datadog/core/vendor/multipart-post/multipart/post/multipartable.rb

@@ -28,29 +28,29 @@ module Datadog
               "--#{SecureRandom.uuid}"
             end
 
-            def initialize(path, params, headers={}, boundary = Multipartable.secure_boundary)
+            def initialize(path, params, headers = {}, boundary = Multipartable.secure_boundary)
               headers = headers.clone # don't want to modify the original variable
               parts_headers = headers.delete(:parts) || {}
               super(path, headers)
-              parts = params.map do |k,v|
+              parts = params.map do |k, v|
                 case v
                 when Array
-                  v.map {|item| Parts::Part.new(boundary, "#{k}[]", item, parts_headers[k]) }
+                  v.map { |item| Parts::Part.new(boundary, "#{k}[]", item, parts_headers[k]) }
                 else
                   Parts::Part.new(boundary, k, v, parts_headers[k])
                 end
               end.flatten
               parts << Parts::EpiloguePart.new(boundary)
-              ios = parts.map {|p| p.to_io }
-              self.set_content_type(headers["Content-Type"] || "multipart/form-data",
-                                    { "boundary" => boundary })
-              self.content_length = parts.inject(0) {|sum,i| sum + i.length }
+              ios = parts.map { |p| p.to_io }
+              set_content_type(headers["Content-Type"] || "multipart/form-data",
+                {"boundary" => boundary})
+              self.content_length = parts.inject(0) { |sum, i| sum + i.length }
               self.body_stream = CompositeReadIO.new(*ios)
 
               @boundary = boundary
             end
 
-            attr :boundary
+            attr_reader :boundary
           end
         end
       end

data/lib/datadog/core/vendor/multipart-post/multipart/post/parts.rb

@@ -58,7 +58,7 @@ module Datadog
                 part = ''
                 part << "--#{boundary}\r\n"
                 part << "Content-ID: #{headers["Content-ID"]}\r\n" if headers["Content-ID"]
-                part << "Content-Disposition: form-data; name=\"#{name.to_s}\"\r\n"
+                part << "Content-Disposition: form-data; name=\"#{name}\"\r\n"
                 part << "Content-Type: #{headers["Content-Type"]}\r\n" if headers["Content-Type"]
                 part << "\r\n"
                 part << "#{value}\r\n"
@@ -76,9 +76,9 @@ module Datadog
               # @param io [IO]
               # @param headers [Hash]
               def initialize(boundary, name, io, headers = {})
-                file_length = io.respond_to?(:length) ?  io.length : File.size(io.local_path)
+                file_length = io.respond_to?(:length) ? io.length : File.size(io.local_path)
                 @head = build_head(boundary, name, io.original_filename, io.content_type, file_length,
-                                  io.respond_to?(:opts) ? io.opts.merge(headers) : headers)
+                  io.respond_to?(:opts) ? io.opts.merge(headers) : headers)
                 @foot = "\r\n"
                 @length = @head.bytesize + file_length + @foot.length
                 @io = CompositeReadIO.new(StringIO.new(@head), io, StringIO.new(@foot))
@@ -98,16 +98,16 @@ module Datadog
 
                 part = ''
                 part << "--#{boundary}\r\n"
-                part << "Content-Disposition: #{content_disposition}; name=\"#{name.to_s}\"; filename=\"#{filename}\"\r\n"
+                part << "Content-Disposition: #{content_disposition}; name=\"#{name}\"; filename=\"#{filename}\"\r\n"
                 part << "Content-Length: #{content_len}\r\n"
                 if content_id = opts.delete("Content-ID")
                   part << "Content-ID: #{content_id}\r\n"
                 end
 
-                if opts["Content-Type"] != nil
-                  part <<  "Content-Type: " + opts["Content-Type"] + "\r\n"
+                part << if !opts["Content-Type"].nil?
+                  "Content-Type: " + opts["Content-Type"] + "\r\n"
                 else
-                  part << "Content-Type: #{type}\r\n"
+                  "Content-Type: #{type}\r\n"
                 end
 
                 part << "Content-Transfer-Encoding: #{trans_encoding}\r\n"

data/lib/datadog/core/worker.rb

@@ -12,7 +12,7 @@ module Datadog
       end
 
       def perform(*args)
-        task.call(*args) unless task.nil?
+        task&.call(*args)
       end
 
       protected

data/lib/datadog/core/workers/async.rb

@@ -47,6 +47,14 @@ module Datadog
             @run_async = false
             Datadog.logger.debug { "Forcibly terminating worker thread for: #{self}" }
             worker.terminate
+            # Wait for the worker thread to end
+            begin
+              Timeout.timeout(SHUTDOWN_TIMEOUT) do
+                worker.join
+              end
+            rescue Timeout::Error
+              Datadog.logger.debug { "Worker thread did not end after #{SHUTDOWN_TIMEOUT} seconds: #{self}" }
+            end
             true
           end
 
@@ -112,23 +120,33 @@ module Datadog
             @worker ||= nil
           end
 
+          # Returns true if worker thread is successfully started,
+          # false if it is not started. Reasons for not starting the worker
+          # thread: it is already running, or the process forked and fork
+          # policy is to stop the workers on fork (which means they are
+          # not started in children, really).
           def start_async(&block)
             mutex.synchronize do
-              return if running?
+              return false if running?
 
               if forked?
                 case fork_policy
                 when FORK_POLICY_STOP
                   stop_fork
+                  false
                 when FORK_POLICY_RESTART
+                  # restart_after_fork should return true
                   restart_after_fork(&block)
                 end
               elsif !run_async?
+                # start_worker should return true
                 start_worker(&block)
               end
             end
           end
 
+          # Returns true if worker thread is successfully started,
+          # which should generally be always.
           def start_worker
             @run_async = true
             @pid = Process.pid
@@ -136,22 +154,21 @@ module Datadog
             Datadog.logger.debug { "Starting thread for: #{self}" }
 
             @worker = ::Thread.new do
-              begin
-                yield
-              # rubocop:disable Lint/RescueException
-              rescue Exception => e
-                @error = e
-                Datadog.logger.debug(
-                  "Worker thread error. Cause: #{e.class.name} #{e.message} Location: #{Array(e.backtrace).first}"
-                )
-                raise
-              end
+              yield
+            # rubocop:disable Lint/RescueException
+            rescue Exception => e
+              @error = e
+              Datadog.logger.debug(
+                "Worker thread error. Cause: #{e.class.name} #{e.message} Location: #{Array(e.backtrace).first}"
+              )
+              raise
+
               # rubocop:enable Lint/RescueException
             end
             @worker.name = self.class.name
             @worker.thread_variable_set(:fork_safe, true)
 
-            nil
+            true
           end
 
           def stop_fork

data/lib/datadog/core/workers/interval_loop.rb

@@ -21,7 +21,18 @@ module Datadog
         # Methods that must be prepended
         module PrependedMethods
           def perform(*args)
-            perform_loop { super(*args) }
+            perform_loop do
+              @in_iteration = true
+              begin
+                super(*args)
+              ensure
+                @in_iteration = false
+              end
+            end
+          end
+
+          def in_iteration?
+            defined?(@in_iteration) && @in_iteration
           end
         end
 

data/lib/datadog/core/workers/runtime_metrics.rb

@@ -20,8 +20,8 @@ module Datadog
         attr_reader \
           :metrics
 
-        def initialize(options = {})
-          @metrics = options.fetch(:metrics) { Core::Runtime::Metrics.new(logger: options[:logger]) }
+        def initialize(telemetry:, **options)
+          @metrics = options.fetch(:metrics) { Core::Runtime::Metrics.new(logger: options[:logger], telemetry: telemetry) }
 
           # Workers::Async::Thread settings
           self.fork_policy = options.fetch(:fork_policy, Workers::Async::Thread::FORK_POLICY_STOP)

data/lib/datadog/core.rb

@@ -11,6 +11,14 @@ module Datadog
   # for higher-level features.
   module Core
     extend Core::Deprecations
+
+    LIBDATADOG_API_FAILURE =
+      begin
+        require "libdatadog_api.#{RUBY_VERSION[/\d+.\d+/]}_#{RUBY_PLATFORM}"
+        nil
+      rescue LoadError => e
+        e.message
+      end
   end
 
   extend Core::Extensions

data/lib/datadog/di/boot.rb

@@ -0,0 +1,34 @@
+# frozen_string_literal: true
+
+require_relative 'logger'
+require_relative 'base'
+require_relative 'error'
+require_relative 'code_tracker'
+require_relative 'component'
+require_relative 'instrumenter'
+require_relative 'probe'
+require_relative 'probe_builder'
+require_relative 'probe_manager'
+require_relative 'probe_notification_builder'
+require_relative 'probe_notifier_worker'
+require_relative 'redactor'
+require_relative 'serializer'
+require_relative 'transport/http'
+require_relative 'utils'
+
+if %w[1 true yes].include?(ENV['DD_DYNAMIC_INSTRUMENTATION_ENABLED']) # steep:ignore
+  # For initial release of Dynamic Instrumentation, activate code tracking
+  # only if DI is explicitly requested in the environment.
+  # Code tracking is required for line probes to work; see the comments
+  # above for the implementation of the method.
+  #
+  # If DI is enabled programmatically, the application can (and must,
+  # for line probes to work) activate tracking in an initializer.
+  # We seem to have Datadog.logger here already
+  Datadog.logger.debug("di: activating code tracking")
+  Datadog::DI.activate_tracking
+end
+
+require_relative 'contrib'
+
+Datadog::DI::Contrib.load_now_or_later

data/lib/datadog/di/remote.rb

@@ -18,6 +18,8 @@ module Datadog
         PRODUCT = 'LIVE_DEBUGGING'
 
         def products
+          # TODO: do not send our product on unsupported runtimes
+          # (Ruby 2.5 / JRuby)
           [PRODUCT]
         end
 

data/lib/datadog/di.rb

@@ -1,24 +1,8 @@
 # frozen_string_literal: true
 
-require_relative 'di/logger'
-require_relative 'di/base'
-require_relative 'di/error'
-require_relative 'di/code_tracker'
-require_relative 'di/component'
 require_relative 'di/configuration'
 require_relative 'di/extensions'
-require_relative 'di/instrumenter'
-require_relative 'di/probe'
-require_relative 'di/probe_builder'
-require_relative 'di/probe_manager'
-require_relative 'di/probe_notification_builder'
-require_relative 'di/probe_notifier_worker'
-require_relative 'di/redactor'
 require_relative 'di/remote'
-require_relative 'di/serializer'
-#require_relative 'di/transport'
-require_relative 'di/transport/http'
-require_relative 'di/utils'
 
 module Datadog
   # Namespace for Datadog dynamic instrumentation.
@@ -52,19 +36,8 @@ module Datadog
   end
 end
 
-if %w(1 true).include?(ENV['DD_DYNAMIC_INSTRUMENTATION_ENABLED']) # steep:ignore
-  # For initial release of Dynamic Instrumentation, activate code tracking
-  # only if DI is explicitly requested in the environment.
-  # Code tracking is required for line probes to work; see the comments
-  # above for the implementation of the method.
-  #
-  # If DI is enabled programmatically, the application can (and must,
-  # for line probes to work) activate tracking in an initializer.
-  # We seem to have Datadog.logger here already
-  Datadog.logger.debug("di: activating code tracking")
-  Datadog::DI.activate_tracking
-end
-
-require_relative 'di/contrib'
-
-Datadog::DI::Contrib.load_now_or_later
+# Line probes will not work on Ruby < 2.6 because of lack of :script_compiled
+# trace point. Activate DI automatically on supported Ruby versions but
+# always load its settings so that, for example, turning DI off when
+# we are on Ruby 2.5 does not produce exceptions.
+require_relative 'di/boot' if RUBY_VERSION >= '2.6' && RUBY_ENGINE != 'jruby'

data/lib/datadog/error_tracking/collector.rb

@@ -0,0 +1,87 @@
+# frozen_string_literal: true
+
+require_relative 'ext'
+
+module Datadog
+  module ErrorTracking
+    # The Collector is in charge, for a SpanOperation of storing the span events
+    # created when an error is handled. Each SpanOperation has a Collector as soon
+    # as a span event is created and the Collector has the same life time as the SpanOp.
+    #
+    # If an error is handled then rethrown, the SpanEvent corresponding to the error
+    # will be deleted. That is why we do not add directly the SpanEvent to the SpanOp.
+    #
+    # @api private
+    class Collector
+      SPAN_EVENTS_LIMIT = 100
+      LOCK = Mutex.new
+      # Proc called when the span_operation :after_stop event is published
+      def self.after_stop
+        @after_stop ||= proc do |span_op, error|
+          # if this proc is called, we are sure that span_op has a collector
+          collector = span_op.get_collector_or_initialize
+          # if an error exited the scope of the span, we delete the corresponding SpanEvent.
+          collector.on_error(span_op, error) if error
+
+          span_events = collector.span_events
+          span_op.span_events.concat(span_events)
+        end
+      end
+
+      def initialize
+        @span_event_per_error = {}
+      end
+
+      def add_span_event(span_op, span_event, error)
+        # When this is the first time we add a span event for a span,
+        # we suscribe to the :after_stop event
+        if @span_event_per_error.empty?
+          events = span_op.send(:events)
+          events.after_stop.subscribe(&self.class.after_stop)
+
+          # This tag is used by the Error Tracking product to report
+          # the error in Error Tracking
+          span_op.set_tag(Ext::SPAN_EVENTS_HAS_EXCEPTION, true)
+        end
+        # Set a limit to the number of span event we can store per SpanOp
+        # If an error has been handled several times in the same span we can still
+        # modify the event (even if the capacity is reached) in order to report
+        # the information of the last rescue
+        if @span_event_per_error.key?(error) || @span_event_per_error.length < SPAN_EVENTS_LIMIT
+          @span_event_per_error[error] = span_event
+        end
+      end
+
+      if RUBY_VERSION >= Ext::RUBY_VERSION_WITH_RESCUE_EVENT
+        # Starting from ruby3.3, as we are listening to :rescue event,
+        # we just want to remove the span event if the error was
+        # previously handled
+        def on_error(_span_op, error)
+          @span_event_per_error.delete(error)
+        end
+      else
+        # Up to ruby3.2, we are listening to :raise event. We need to ensure
+        # that an error exiting the scope of a span is not handled in a parent span.
+        # This function will propagate the span event to the parent span. If the
+        # error is not handled in the parent span, it will be deleted by design.
+        def on_error(span_op, error)
+          return unless @span_event_per_error.key?(error)
+
+          unless span_op.root?
+            parent = span_op.send(:parent)
+            LOCK.synchronize do
+              parent_collector = parent.get_collector_or_initialize { Collector.new }
+              parent_collector.add_span_event(parent, @span_event_per_error[error], error)
+            end
+          end
+
+          @span_event_per_error.delete(error)
+        end
+      end
+
+      def span_events
+        @span_event_per_error.values
+      end
+    end
+  end
+end

data/lib/datadog/error_tracking/component.rb

@@ -0,0 +1,167 @@
+# frozen_string_literal: true
+
+require 'set'
+require_relative 'collector'
+require_relative 'filters'
+
+module Datadog
+  module ErrorTracking
+    # Component for Error Tracking.
+    #
+    # Only one instance of the Component should ever be active.
+    #
+    # The component instance records every handled exceptions from the configured scopes
+    # (user, third_party packages, specified files or everything).
+    class Component
+      LOCK = Mutex.new
+
+      class << self
+        def build(settings, tracer, logger)
+          return if !settings.respond_to?(:error_tracking) || (settings.error_tracking.handled_errors.nil? &&
+            settings.error_tracking.handled_errors_include.empty?)
+
+          return unless environment_supported?(logger)
+
+          new(
+            tracer: tracer,
+            handled_errors: settings.error_tracking.handled_errors,
+            handled_errors_include: settings.error_tracking.handled_errors_include,
+          ).tap(&:start)
+        end
+
+        def environment_supported?(logger)
+          if RUBY_ENGINE != 'ruby'
+            logger.warn("error tracking: cannot enable error tracking: MRI is required, but running on #{RUBY_ENGINE}")
+            false
+          elsif RUBY_VERSION < '2.7'
+            logger.warn(
+              "error tracking: cannot enable error tracking: Ruby 2.7+ is required, but running
+              on #{RUBY_VERSION}"
+            )
+            false
+          else
+            true
+          end
+        end
+      end
+
+      def initialize(tracer:, handled_errors:, handled_errors_include:)
+        @tracer = tracer
+
+        # Hash containing the paths to the instrumented files
+        @instrumented_files = Set.new unless handled_errors_include.empty?
+        # Array containing file paths, file names and gems names to instrument.
+        # This is coming from the DD_ERROR_TRACKING_HANDLED_ERRORS_INCLUDE env variable
+        @handled_errors_include = handled_errors_include
+
+        # Filter function is used to filter out the exception
+        # we do not want to report. For instance exception from gems.
+        @filter_function = Filters.generate_filter(handled_errors, @instrumented_files)
+
+        # :rescue event was added in Ruby 3.3
+        #
+        # Before Ruby3.3 the TracePoint listen for :raise events.
+        # If an error is not handled, we will delete the according
+        # span event in the collector.
+        event = (RUBY_VERSION >= '3.3') ? :rescue : :raise
+
+        # This TracePoint is in charge of capturing the handled exceptions
+        # and of adding the corresponding span events to the collector
+        @handled_exc_tracker = create_exc_tracker_trace_point(event)
+
+        if @instrumented_files
+          # The only thing we know about the handled errors is the path of the file
+          # in which the error was rescued. Therefore, we need to retrieve the path
+          # of the files the user want to instrument. This TracePoint is used for that
+          # purpose
+          @include_path_getter = create_script_compiled_trace_point
+        end
+      end
+
+      def create_exc_tracker_trace_point(event)
+        TracePoint.new(event) do |tp|
+          active_span = @tracer.active_span
+          if active_span
+            raised_exception = tp.raised_exception
+            # Note that in 3.2, this will give the path of where the error was raised
+            # which may cause the handled_error_include env variable to malfunction.
+            rescue_file_path = tp.path
+            if @filter_function.call(rescue_file_path)
+              span_event = generate_span_event(raised_exception)
+              LOCK.synchronize do
+                collector = active_span.get_collector_or_initialize { Collector.new }
+                collector.add_span_event(active_span, span_event, raised_exception)
+              end
+            end
+          end
+        end
+      end
+
+      def create_script_compiled_trace_point
+        TracePoint.new(:script_compiled) do |tp|
+          next if tp.eval_script
+
+          path = tp.instruction_sequence.path
+          next if path.nil?
+
+          @handled_errors_include.each do |file_to_instr|
+            # The user can provide either
+            # - absolute_path starting with '/'. In that case the path of the file
+            #   should begin with file_to_instr
+            # - a relative_path starting with './'. In that case, we extend the path
+            #   and it is the same as above
+            # - otherwise we just check if the name provided is in the path and is
+            #   either the name of a folder or of a ruby file.
+            regex =
+              if file_to_instr.start_with?('/')
+                %r{\A#{Regexp.escape(file_to_instr)}(?:/|\.rb\z|\z)}
+              elsif file_to_instr.start_with?('./')
+                abs_path = File.expand_path(file_to_instr)
+                %r{\A#{Regexp.escape(abs_path)}(?:/|\.rb\z|\z)}
+              else
+                %r{/#{Regexp.escape(file_to_instr)}(?:/|\.rb\z|\z)}
+              end
+
+            add_instrumented_file(path) if path.match?(regex)
+          end
+        end
+      end
+
+      # Starts the TracePoints.
+      #
+      # Enables the script_compiled TracePoint if handled_errors_include is not empty.
+      def start
+        @handled_exc_tracker.enable
+        @include_path_getter&.enable
+      end
+
+      # Shuts down error tracker.
+      #
+      # Disables the TracePoints.
+      def shutdown!
+        @handled_exc_tracker.disable
+        @include_path_getter&.disable
+      end
+
+      private
+
+      # Generates a span event from the exception info.
+      #
+      # The event follows the otel semantics.
+      # https://opentelemetry.io/docs/specs/otel/trace/exceptions/
+      def generate_span_event(exception)
+        formatted_exception = Datadog::Core::Error.build_from(exception)
+        attributes = {
+          'exception.type' => formatted_exception.type,
+          'exception.message' => formatted_exception.message,
+          'exception.stacktrace' => formatted_exception.backtrace
+        }
+        Datadog::Tracing::SpanEvent.new('exception', attributes: attributes)
+      end
+
+      def add_instrumented_file(file_path)
+        @instrumented_files&.add(file_path)
+      end
+    end
+  end
+end