datadog 2.14.0 → 2.16.0

data/lib/datadog/core/telemetry/transport/http.rb

@@ -0,0 +1,63 @@
+# frozen_string_literal: true
+
+require_relative 'telemetry'
+require_relative 'http/api'
+require_relative '../../transport/http'
+
+module Datadog
+  module Core
+    module Telemetry
+      module Transport
+        # Namespace for HTTP transport components
+        module HTTP
+          module_function
+
+          # Builds a new Transport::HTTP::Client with default settings
+          # Pass a block to override any settings.
+          def agentless_telemetry(
+            agent_settings:,
+            logger:,
+            api_key: nil,
+            api_version: nil,
+            headers: nil
+          )
+            Core::Transport::HTTP.build(api_instance_class: Telemetry::API::Instance,
+              logger: logger,
+              agent_settings: agent_settings,
+              api_version: api_version,
+              headers: headers) do |transport|
+              apis = API.defaults
+
+              transport.api API::AGENTLESS_TELEMETRY, apis[API::AGENTLESS_TELEMETRY]
+
+              # Call block to apply any customization, if provided
+              yield(transport) if block_given?
+            end.to_transport(Core::Telemetry::Transport::Telemetry::Transport).tap do |transport|
+              transport.api_key = api_key
+            end
+          end
+
+          # Builds a new Transport::HTTP::Client with default settings
+          # Pass a block to override any settings.
+          def agent_telemetry(
+            agent_settings:,
+            logger:,
+            api_version: nil,
+            headers: nil
+          )
+            Core::Transport::HTTP.build(api_instance_class: Telemetry::API::Instance,
+              logger: logger,
+              agent_settings: agent_settings, api_version: api_version, headers: headers) do |transport|
+              apis = API.defaults
+
+              transport.api API::AGENT_TELEMETRY, apis[API::AGENT_TELEMETRY]
+
+              # Call block to apply any customization, if provided
+              yield(transport) if block_given?
+            end.to_transport(Core::Telemetry::Transport::Telemetry::Transport)
+          end
+        end
+      end
+    end
+  end
+end

data/lib/datadog/core/telemetry/transport/telemetry.rb

@@ -0,0 +1,52 @@
+# frozen_string_literal: true
+
+require_relative '../../transport/parcel'
+require_relative 'http/client'
+require_relative 'http/telemetry'
+
+module Datadog
+  module Core
+    module Telemetry
+      module Transport
+        module Telemetry
+          class EncodedParcel
+            include Datadog::Core::Transport::Parcel
+          end
+
+          class Request < Datadog::Core::Transport::Request
+
+            attr_reader :request_type
+            attr_reader :api_key
+
+            def initialize(request_type, parcel, api_key)
+              @request_type = request_type
+              super(parcel)
+              @api_key = api_key
+            end
+          end
+
+          class Transport
+            attr_reader :client, :apis, :default_api, :current_api_id, :logger
+            attr_accessor :api_key
+
+            def initialize(apis, default_api, logger:)
+              @apis = apis
+              @logger = logger
+
+              @client = HTTP::Client.new(@apis[default_api], logger: logger)
+            end
+
+            def send_telemetry(request_type:, payload:)
+              json = JSON.dump(payload)
+              parcel = EncodedParcel.new(json)
+              request = Request.new(request_type, parcel, api_key)
+
+              @client.send_telemetry_payload(request)
+              # Perform no error checking here
+            end
+          end
+        end
+      end
+    end
+  end
+end

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

@@ -52,10 +52,15 @@ module Datadog
 
         attr_reader :logger
 
+        # Returns true if worker thread is successfully started,
+        # false if worker thread was not started but telemetry is enabled,
+        # nil if telemetry is disabled.
         def start
           return if !enabled? || forked?
 
           # starts async worker
+          # perform should return true if thread was actually started,
+          # false otherwise
           perform
         end
 
@@ -65,10 +70,15 @@ module Datadog
           super
         end
 
+        # Returns true if event was enqueued, nil if not.
+        # While returning false may seem more reasonable, the only reason
+        # for not enqueueing event (presently) is that telemetry is disabled
+        # altogether, and in this case other methods return nil.
         def enqueue(event)
           return if !enabled? || forked?
 
           buffer.push(event)
+          true
         end
 
         def sent_started_event?
@@ -79,6 +89,39 @@ module Datadog
           TELEMETRY_STARTED_ONCE.failed?
         end
 
+        # Wait for the worker to send out all events that have already
+        # been queued, up to 15 seconds. Returns whether all events have
+        # been flushed.
+        #
+        # @api private
+        def flush
+          return true unless enabled? || !run_loop?
+
+          started = Utils::Time.get_time
+          loop do
+            # The AppStarted event is triggered by the worker itself,
+            # from the worker thread. As such the main thread has no way
+            # to delay itself until that event is queued and we need some
+            # way to wait until that event is sent out to assert on it in
+            # the test suite. Check the run once flag which *should*
+            # indicate the event has been queued (at which point our queue
+            # depth check should waint until it's sent).
+            # This is still a hack because the flag can be overridden
+            # either way with or without the event being sent out.
+            # Note that if the AppStarted sending fails, this check
+            # will return false and flushing will be blocked until the
+            # 15 second timeout.
+            # Note that the first wait interval between telemetry event
+            # sending is 10 seconds, the timeout needs to be strictly
+            # greater than that.
+            return true if buffer.empty? && !in_iteration? && TELEMETRY_STARTED_ONCE.success?
+
+            sleep 0.5
+
+            return false if Utils::Time.get_time - started > 15
+          end
+        end
+
         private
 
         def perform(*events)
@@ -99,6 +142,8 @@ module Datadog
 
         def flush_events(events)
           return if events.empty?
+          # TODO: can this method silently drop events which are
+          # generated prior to the started event being submitted?
           return if !enabled? || !sent_started_event?
 
           events = deduplicate_logs(events)

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

@@ -31,6 +31,12 @@ 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.
+            remove_method(:now) rescue nil
+          end
           define_singleton_method(:now, &block)
         end
 
@@ -43,6 +49,12 @@ 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.
+            remove_method(:get_time) rescue nil
+          end
           define_singleton_method(:get_time, &block)
         end
 

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
@@ -151,7 +169,7 @@ module Datadog
             @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/probe_notification_builder.rb

@@ -191,7 +191,7 @@ module Datadog
       end
 
       def timestamp_now
-        (Time.now.to_f * 1000).to_i
+        (Core::Utils::Time.now.to_f * 1000).to_i
       end
 
       def get_local_variables(trace_point)

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/transport/http/diagnostics.rb

@@ -38,7 +38,6 @@ module Datadog
               end
             end
 
-            # Endpoint for negotiation
             class Endpoint < Datadog::Core::Transport::HTTP::API::Endpoint
               attr_reader :encoder
 

data/lib/datadog/di/transport/http/input.rb

@@ -38,7 +38,6 @@ module Datadog
               end
             end
 
-            # Endpoint for negotiation
             class Endpoint < Datadog::Core::Transport::HTTP::API::Endpoint
               HEADER_CONTENT_TYPE = 'Content-Type'
 

data/lib/datadog/di/transport/http.rb

@@ -1,15 +1,9 @@
 # frozen_string_literal: true
 
-require 'uri'
-
-require_relative '../../core/environment/container'
-require_relative '../../core/environment/ext'
-require_relative '../../core/transport/ext'
 require_relative 'diagnostics'
 require_relative 'input'
 require_relative 'http/api'
 require_relative '../../core/transport/http'
-require_relative '../../../datadog/version'
 
 module Datadog
   module DI

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