ddtrace 0.48.0 → 0.51.1

data/lib/ddtrace/opentelemetry/extensions.rb

@@ -6,7 +6,7 @@ module Datadog
     # Defines extensions to ddtrace for OpenTelemetry support
     module Extensions
       def self.extended(base)
-        Datadog::Span.send(:prepend, OpenTelemetry::Span)
+        Datadog::Span.prepend(OpenTelemetry::Span)
       end
     end
   end

data/lib/ddtrace/opentracer.rb

@@ -1,40 +1,21 @@
-module Datadog
-  # Namespace for ddtrace OpenTracing implementation
-  module OpenTracer
-    module_function
+require 'opentracing'
+require 'opentracing/carrier'
+require 'ddtrace'
+require 'ddtrace/opentracer/carrier'
+require 'ddtrace/opentracer/tracer'
+require 'ddtrace/opentracer/span'
+require 'ddtrace/opentracer/span_context'
+require 'ddtrace/opentracer/span_context_factory'
+require 'ddtrace/opentracer/scope'
+require 'ddtrace/opentracer/scope_manager'
+require 'ddtrace/opentracer/thread_local_scope'
+require 'ddtrace/opentracer/thread_local_scope_manager'
+require 'ddtrace/opentracer/distributed_headers'
+require 'ddtrace/opentracer/propagator'
+require 'ddtrace/opentracer/text_map_propagator'
+require 'ddtrace/opentracer/binary_propagator'
+require 'ddtrace/opentracer/rack_propagator'
+require 'ddtrace/opentracer/global_tracer'
 
-    def supported?
-      Gem::Version.new(RUBY_VERSION) >= Gem::Version.new('2.1')
-    end
-
-    def load_opentracer
-      require 'opentracing'
-      require 'opentracing/carrier'
-      require 'ddtrace'
-      require 'ddtrace/opentracer/carrier'
-      require 'ddtrace/opentracer/tracer'
-      require 'ddtrace/opentracer/span'
-      require 'ddtrace/opentracer/span_context'
-      require 'ddtrace/opentracer/span_context_factory'
-      require 'ddtrace/opentracer/scope'
-      require 'ddtrace/opentracer/scope_manager'
-      require 'ddtrace/opentracer/thread_local_scope'
-      require 'ddtrace/opentracer/thread_local_scope_manager'
-      require 'ddtrace/opentracer/distributed_headers'
-      require 'ddtrace/opentracer/propagator'
-      require 'ddtrace/opentracer/text_map_propagator'
-      require 'ddtrace/opentracer/binary_propagator'
-      require 'ddtrace/opentracer/rack_propagator'
-      require 'ddtrace/opentracer/global_tracer'
-
-      # Modify the OpenTracing module functions
-      OpenTracing.module_eval do
-        class << self
-          prepend Datadog::OpenTracer::GlobalTracer
-        end
-      end
-    end
-
-    load_opentracer if supported?
-  end
-end
+# Modify the OpenTracing module functions
+::OpenTracing.singleton_class.prepend(Datadog::OpenTracer::GlobalTracer)

data/lib/ddtrace/patcher.rb

@@ -16,8 +16,8 @@ module Datadog
         )
       end
 
-      base.send(:extend, CommonMethods)
-      base.send(:include, CommonMethods)
+      base.extend(CommonMethods)
+      base.include(CommonMethods)
     end
 
     # Defines some common methods for patching, that can be used

data/lib/ddtrace/profiling.rb

@@ -4,40 +4,70 @@ module Datadog
     module_function
 
     GOOGLE_PROTOBUF_MINIMUM_VERSION = Gem::Version.new('3.0')
+    private_constant :GOOGLE_PROTOBUF_MINIMUM_VERSION
 
     def supported?
-      google_protobuf_supported?
+      unsupported_reason.nil?
     end
 
-    def native_cpu_time_supported?
-      require 'ddtrace/profiling/ext/cpu'
-      Ext::CPU.supported?
+    def unsupported_reason
+      # NOTE: Only the first matching reason is returned, so try to keep a nice order on reasons -- e.g. tell users
+      # first that they can't use this on JRuby before telling them that they are missing protobuf
+
+      ruby_engine_unsupported? || protobuf_gem_unavailable? || protobuf_version_unsupported? || protobuf_failed_to_load?
     end
 
-    def google_protobuf_supported?
-      RUBY_PLATFORM != 'java' \
-        && !Gem.loaded_specs['google-protobuf'].nil? \
-        && Gem.loaded_specs['google-protobuf'].version >= GOOGLE_PROTOBUF_MINIMUM_VERSION \
-        && !defined?(@failed_to_load_protobuf)
+    def self.ruby_engine_unsupported?
+      'JRuby is not supported' if RUBY_ENGINE == 'jruby'
     end
+    private_class_method :ruby_engine_unsupported?
 
-    def load_profiling
-      require 'ddtrace/profiling/ext/cpu'
-      require 'ddtrace/profiling/ext/forking'
+    def self.protobuf_gem_unavailable?
+      # NOTE: On environments where protobuf is already loaded, we skip the check. This allows us to support environments
+      # where no Gem.loaded_version is NOT available but customers are able to load protobuf; see for instance
+      # https://github.com/teamcapybara/capybara/commit/caf3bcd7664f4f2691d0ca9ef3be9a2a954fecfb
+      if !defined?(::Google::Protobuf) && Gem.loaded_specs['google-protobuf'].nil?
+        "Missing google-protobuf dependency; please add `gem 'google-protobuf', '~> 3.0'` to your Gemfile or gems.rb file"
+      end
+    end
+    private_class_method :protobuf_gem_unavailable?
 
-      require 'ddtrace/profiling/collectors/stack'
-      require 'ddtrace/profiling/exporter'
-      require 'ddtrace/profiling/recorder'
-      require 'ddtrace/profiling/scheduler'
-      require 'ddtrace/profiling/tasks/setup'
-      require 'ddtrace/profiling/transport/io'
-      require 'ddtrace/profiling/transport/http'
-      require 'ddtrace/profiling/profiler'
+    def self.protobuf_version_unsupported?
+      # See above for why we skip the check when protobuf is already loaded; note that when protobuf was already loaded
+      # we skip the version check to avoid the call to Gem.loaded_specs. Unfortunately, protobuf does not seem to
+      # expose the gem version constant elsewhere, so in that setup we are not able to check the version.
+      if !defined?(::Google::Protobuf) && Gem.loaded_specs['google-protobuf'].version < GOOGLE_PROTOBUF_MINIMUM_VERSION
+        'Your google-protobuf is too old; ensure that you have google-protobuf >= 3.0 by ' \
+        "adding `gem 'google-protobuf', '~> 3.0'` to your Gemfile or gems.rb file"
+      end
+    end
+    private_class_method :protobuf_version_unsupported?
+
+    def self.protobuf_failed_to_load?
+      unless protobuf_loaded_successfully?
+        'There was an error loading the google-protobuf library; see previous warning message for details'
+      end
+    end
+    private_class_method :protobuf_failed_to_load?
+
+    # The `google-protobuf` gem depends on a native component, and its creators helpfully tried to provide precompiled
+    # versions of this extension on rubygems.org.
+    #
+    # Unfortunately, for a long time, the supported Ruby versions metadata on these precompiled versions of the extension
+    # was not correctly set. (This is fixed in newer versions -- but not all Ruby versions we want to support can use
+    # these.)
+    #
+    # Thus, the gem can still be installed, but can be in a broken state. To avoid breaking customer applications, we
+    # use this helper to load it and gracefully handle failures.
+    def self.protobuf_loaded_successfully?
+      return @protobuf_loaded if defined?(@protobuf_loaded)
 
       begin
-        require 'ddtrace/profiling/pprof/pprof_pb' if google_protobuf_supported?
+        require 'google/protobuf'
+        @protobuf_loaded = true
       rescue LoadError => e
-        @failed_to_load_protobuf = true
+        # NOTE: We use Kernel#warn here because this code gets run BEFORE Datadog.logger is actually set up.
+        # In the future it'd be nice to shuffle the logger startup to happen first to avoid this special case.
         Kernel.warn(
           "[DDTRACE] Error while loading google-protobuf gem. Cause: '#{e.message}' Location: '#{e.backtrace.first}'. " \
           'This can happen when google-protobuf is missing its native components. ' \
@@ -46,8 +76,31 @@ module Datadog
           'If the error persists, please contact support via <https://docs.datadoghq.com/help/> or ' \
           'file a bug at <https://github.com/DataDog/dd-trace-rb/blob/master/CONTRIBUTING.md#found-a-bug>.'
         )
+        @protobuf_loaded = false
       end
     end
+    private_class_method :protobuf_loaded_successfully?
+
+    def self.load_profiling
+      return false unless supported?
+
+      require 'ddtrace/profiling/ext/cpu'
+      require 'ddtrace/profiling/ext/forking'
+
+      require 'ddtrace/profiling/collectors/stack'
+      require 'ddtrace/profiling/exporter'
+      require 'ddtrace/profiling/recorder'
+      require 'ddtrace/profiling/scheduler'
+      require 'ddtrace/profiling/tasks/setup'
+      require 'ddtrace/profiling/transport/io'
+      require 'ddtrace/profiling/transport/http'
+      require 'ddtrace/profiling/profiler'
+
+      require 'ddtrace/profiling/pprof/pprof_pb'
+
+      true
+    end
+    private_class_method :load_profiling
 
     load_profiling if supported?
   end

data/lib/ddtrace/profiling/collectors/stack.rb

@@ -11,6 +11,7 @@ module Datadog
       # Collects stack trace samples from Ruby threads for both CPU-time (if available) and wall-clock.
       # Runs on its own background thread.
       #
+      # rubocop:disable Metrics/ClassLength
       class Stack < Worker
         include Workers::Polling
 
@@ -27,7 +28,7 @@ module Datadog
 
         def initialize(
           recorder,
-          max_frames: nil,
+          max_frames:,
           ignore_thread: nil,
           max_time_usage_pct: DEFAULT_MAX_TIME_USAGE_PCT,
           thread_api: Thread,
@@ -36,8 +37,7 @@ module Datadog
           enabled: true
         )
           @recorder = recorder
-          # TODO: Make this a required named argument after we drop support for Ruby 2.0
-          @max_frames = max_frames || raise(ArgumentError, 'missing keyword :max_frames')
+          @max_frames = max_frames
           @ignore_thread = ignore_thread
           @max_time_usage_pct = max_time_usage_pct
           @thread_api = thread_api
@@ -52,6 +52,11 @@ module Datadog
           self.enabled = enabled
 
           @warn_about_missing_cpu_time_instrumentation_only_once = Datadog::Utils::OnlyOnce.new
+
+          # Cache this proc, since it's pretty expensive to keep recreating it
+          @build_backtrace_location = method(:build_backtrace_location).to_proc
+          # Cache this buffer, since it's pretty expensive to keep accessing it
+          @stack_sample_event_recorder = recorder[Events::StackSample]
         end
 
         def start
@@ -117,7 +122,7 @@ module Datadog
           # Convert backtrace locations into structs
           locations = convert_backtrace_locations(locations)
 
-          thread_id = thread.respond_to?(:native_thread_id) ? thread.native_thread_id : thread.object_id
+          thread_id = thread.respond_to?(:pthread_thread_id) ? thread.pthread_thread_id : thread.object_id
           trace_id, span_id = get_trace_identifiers(thread)
           cpu_time = get_cpu_time_interval!(thread)
 
@@ -183,7 +188,7 @@ module Datadog
         def convert_backtrace_locations(locations)
           locations.collect do |location|
             # Re-use existing BacktraceLocation if identical copy, otherwise build a new one.
-            recorder[Events::StackSample].cache(:backtrace_locations).fetch(
+            @stack_sample_event_recorder.cache(:backtrace_locations).fetch(
               # Function name
               location.base_label,
               # Line number
@@ -191,13 +196,13 @@ module Datadog
               # Filename
               location.path,
               # Build function
-              &method(:build_backtrace_location)
+              &@build_backtrace_location
             )
           end
         end
 
         def build_backtrace_location(_id, base_label, lineno, path)
-          string_table = recorder[Events::StackSample].string_table
+          string_table = @stack_sample_event_recorder.string_table
 
           Profiling::BacktraceLocation.new(
             string_table.fetch_string(base_label),
@@ -219,6 +224,7 @@ module Datadog
             #    In this case, the fix is to make sure ddtrace gets loaded before any other parts of the application.
             #
             # b) The thread was started using the Ruby native APIs (e.g. from a C extension such as ffi).
+            #    Known cases right now that trigger this are the ethon/typhoeus gems.
             #    We currently have no solution for this case; these threads will always be missing our CPU instrumentation.
             #
             # c) The thread was started with `Thread.start`/`Thread.fork` and hasn't yet enabled the instrumentation.
@@ -227,7 +233,9 @@ module Datadog
             #    it to run and our instrumentation to be applied.
             #
             if thread_api.current.respond_to?(:cpu_time) && thread_api.current.cpu_time
-              Datadog.logger.debug("Detected thread ('#{thread}') with missing CPU profiling instrumentation.")
+              Datadog.logger.debug(
+                "Thread ('#{thread}') is missing profiling instrumentation; other threads should be unaffected"
+              )
             end
           end
         end
@@ -248,6 +256,7 @@ module Datadog
           end
         end
       end
+      # rubocop:enable Metrics/ClassLength
     end
   end
 end

data/lib/ddtrace/profiling/encoding/profile.rb

@@ -1,4 +1,5 @@
 require 'set'
+require 'time'
 
 require 'ddtrace/profiling/flush'
 require 'ddtrace/profiling/pprof/template'
@@ -21,6 +22,11 @@ module Datadog
             # Add all events to the pprof
             flush.event_groups.each { |event_group| template.add_events!(event_group.event_class, event_group.events) }
 
+            Datadog.logger.debug do
+              "Encoding profile covering #{flush.start.iso8601} to #{flush.finish.iso8601}, " \
+              "events: #{flush.event_count} (#{template.debug_statistics})"
+            end
+
             # Build the profile and encode it
             template.to_pprof
           end

data/lib/ddtrace/profiling/ext/cpu.rb

@@ -24,8 +24,8 @@ module Datadog
           # Applying CThread to Thread will ensure any new threads
           # will provide a thread/clock ID for CPU timing.
           require 'ddtrace/profiling/ext/cthread'
-          ::Thread.send(:prepend, Profiling::Ext::CThread)
-          ::Thread.singleton_class.send(:prepend, Datadog::Profiling::Ext::WrapThreadStartFork)
+          ::Thread.prepend(Profiling::Ext::CThread)
+          ::Thread.singleton_class.prepend(Datadog::Profiling::Ext::WrapThreadStartFork)
         end
 
         def self.unsupported_reason
@@ -40,14 +40,27 @@ module Datadog
             'Feature requires Linux; Windows is not supported'
           elsif !RUBY_PLATFORM.include?('linux')
             "Feature requires Linux; #{RUBY_PLATFORM} is not supported"
-          elsif Gem::Version.new(RUBY_VERSION) < Gem::Version.new('2.1')
-            'Ruby >= 2.1 is required'
           elsif Gem::Specification.find_all_by_name('rollbar', ROLLBAR_INCOMPATIBLE_VERSIONS).any?
             'You have an incompatible rollbar gem version installed; ensure that you have rollbar >= 3.1.2 by ' \
             "adding `gem 'rollbar', '>= 3.1.2'` to your Gemfile or gems.rb file. " \
-            'See https://github.com/rollbar/rollbar-gem/pull/1018 for details.'
+            'See https://github.com/rollbar/rollbar-gem/pull/1018 for details'
+          elsif Gem::Specification.find_all_by_name('logging').any? && logging_inherit_context_enabled?
+            'The `logging` gem is installed and its thread inherit context feature is enabled. ' \
+            "Please add LOGGING_INHERIT_CONTEXT=false to your application's environment variables to disable the " \
+            'conflicting `logging` gem feature. ' \
+            'See https://github.com/TwP/logging/pull/230 for details'
           end
         end
+
+        private_class_method def self.logging_inherit_context_enabled?
+          # The logging gem provides a mechanism to disable the conflicting behavior, see
+          # https://github.com/TwP/logging/blob/ae9872d093833b2a5a34cbe1faa4e895a81f6845/lib/logging/diagnostic_context.rb#L418
+          # Here we check if the behavior is enabled
+          inherit_context_configuration = ENV['LOGGING_INHERIT_CONTEXT']
+
+          inherit_context_configuration.nil? ||
+          (inherit_context_configuration && !%w[false no 0].include?(inherit_context_configuration.downcase))
+        end
       end
     end
   end

data/lib/ddtrace/profiling/ext/cthread.rb

@@ -8,13 +8,33 @@ module Datadog
         layout :value, :int
       end
 
-      # Extension used to enable CPU-time profiling via use of Pthread's `getcpuclockid`.
-      module CThread
+      # Enables interfacing with pthread via FFI
+      module NativePthread
         extend FFI::Library
         ffi_lib ['pthread', 'libpthread.so.0']
         attach_function :pthread_self, [], :ulong
         attach_function :pthread_getcpuclockid, [:ulong, CClockId], :int
 
+        # NOTE: Only returns thread ID for thread that evaluates this call.
+        #       a.k.a. evaluating `get_pthread_thread_id(thread_a)` from within
+        #       `thread_b` will return `thread_b`'s thread ID, not `thread_a`'s.
+        def self.get_pthread_thread_id(thread)
+          return unless ::Thread.current == thread
+
+          pthread_self
+        end
+
+        def self.get_clock_id(thread, pthread_id)
+          return unless ::Thread.current == thread && pthread_id
+
+          clock = CClockId.new
+          clock[:value] = 0
+          pthread_getcpuclockid(pthread_id, clock).zero? ? clock[:value] : nil
+        end
+      end
+
+      # Extension used to enable CPU-time profiling via use of pthread's `getcpuclockid`.
+      module CThread
         def self.prepended(base)
           # Threads that have already been created, will not have resolved
           # a thread/clock ID. This is because these IDs can only be resolved
@@ -27,12 +47,34 @@ module Datadog
           base.current.send(:update_native_ids) if base.current.is_a?(CThread)
         end
 
-        attr_reader \
-          :native_thread_id
+        # Process::Waiter crash workaround:
+        #
+        # This is a workaround for a Ruby VM segfault (usually something like
+        # "[BUG] Segmentation fault at 0x0000000000000008") in the affected Ruby versions.
+        # See https://bugs.ruby-lang.org/issues/17807 and the regression tests added to this module's specs for details.
+        #
+        # In those Ruby versions, there's a very special subclass of `Thread` called `Process::Waiter` that causes VM
+        # crashes whenever something tries to read its instance variables. This subclass of thread only shows up when
+        # the `Process.detach` API gets used.
+        # In this module's specs you can find crash regression tests that include a way of reproducing it.
+        #
+        # The workaround is to use `defined?` to check first if the instance variable exists. This seems to be fine
+        # with Ruby.
+        # Note that this crash doesn't affect `@foo ||=` nor instance variable writes (after the first write ever of any
+        # instance variable on a `Process::Waiter`, then further reads and writes to that or any other instance are OK;
+        # it looks like there's some lazily-created structure that is missing and did not get created).
+        if Gem::Version.new(RUBY_VERSION) < Gem::Version.new('2.3') &&
+           Gem::Version.new(RUBY_VERSION) >= Gem::Version.new('2.7')
+          attr_reader :pthread_thread_id
+        else
+          def pthread_thread_id
+            defined?(@pthread_thread_id) && @pthread_thread_id
+          end
+        end
 
         def initialize(*args)
           @pid = ::Process.pid
-          @native_thread_id = nil
+          @pthread_thread_id = nil
           @clock_id = nil
 
           # Wrap the work block with our own
@@ -48,15 +90,8 @@ module Datadog
         end
         ruby2_keywords :initialize if respond_to?(:ruby2_keywords, true)
 
-        def clock_id
-          update_native_ids if forked?
-          defined?(@clock_id) && @clock_id
-        end
-
         def cpu_time(unit = :float_second)
-          return unless clock_id && ::Process.respond_to?(:clock_gettime)
-
-          ::Process.clock_gettime(clock_id, unit)
+          ::Process.clock_gettime(clock_id, unit) if clock_id
         end
 
         def cpu_time_instrumentation_installed?
@@ -71,36 +106,22 @@ module Datadog
 
         private
 
-        # Retrieves number of classes from runtime
+        def clock_id
+          update_native_ids if forked?
+          defined?(@clock_id) && @clock_id
+        end
+
         def forked?
           ::Process.pid != (@pid ||= nil)
         end
 
         def update_native_ids
-          # Can only resolve if invoked from same thread.
+          # Can only resolve if invoked from same thread
           return unless ::Thread.current == self
 
           @pid = ::Process.pid
-          @native_thread_id = get_native_thread_id
-          @clock_id = get_clock_id(@native_thread_id)
-        end
-
-        def get_native_thread_id
-          return unless ::Thread.current == self
-
-          # NOTE: Only returns thread ID for thread that evaluates this call.
-          #       a.k.a. evaluating `thread_a.get_native_thread_id` from within
-          #       `thread_b` will return `thread_b`'s thread ID, not `thread_a`'s.
-          pthread_self
-        end
-
-        def get_clock_id(pthread_id)
-          return unless pthread_id && alive?
-
-          # Build a struct, pass it to Pthread's getcpuclockid function.
-          clock = CClockId.new
-          clock[:value] = 0
-          pthread_getcpuclockid(pthread_id, clock).zero? ? clock[:value] : nil
+          @pthread_thread_id = NativePthread.get_pthread_thread_id(self)
+          @clock_id = NativePthread.get_clock_id(self, @pthread_thread_id)
         end
       end