ddtrace 0.48.0 → 0.49.0

data/lib/ddtrace/error.rb

@@ -1,23 +1,99 @@
+# frozen_string_literal: true
+
 # Datadog global namespace
 module Datadog
   # Error is a value-object responsible for sanitizing/encapsulating error data
   class Error
     attr_reader :type, :message, :backtrace
 
-    def self.build_from(value)
-      case value
-      when Error then value
-      when Array then new(*value)
-      when Exception then new(value.class, value.message, value.backtrace)
-      when ContainsMessage then new(value.class, value.message)
-      else BlankError
+    class << self
+      def build_from(value)
+        case value
+        when Error then value
+        when Array then new(*value)
+        when Exception then new(value.class, value.message, full_backtrace(value))
+        when ContainsMessage then new(value.class, value.message)
+        else BlankError
+        end
+      end
+
+      private
+
+      if Gem::Version.new(RUBY_VERSION) < Gem::Version.new('2.1.0')
+        # Ruby 2.0 exceptions don't have `cause`.
+        # Only current exception stack trace is reported.
+        # This is the same behavior as before.
+        def full_backtrace(ex)
+          backtrace = ex.backtrace
+          backtrace.join("\n") if backtrace
+        end
+      else
+        # Returns a stack trace with nested error causes and details.
+        #
+        # This manually implements Ruby >= 2.6 error output for two reasons:
+        #
+        # 1. It is not available in Ruby < 2.6.
+        # 2. It's measurably faster to manually implement it in Ruby.
+        #
+        # This method mimics the exact output of
+        # `ex.full_message(highlight: false, order: :top)`
+        # but it's around 3x faster in our benchmark test
+        # at `error_spec.rb`.
+        def full_backtrace(ex)
+          backtrace = String.new
+          backtrace_for(ex, backtrace)
+
+          # Avoid circular causes
+          causes = {}
+          causes[ex] = true
+
+          while (cause = ex.cause) && !causes.key?(cause)
+            backtrace_for(cause, backtrace)
+            causes[cause] = true
+          end
+
+          backtrace
+        end
+
+        # Outputs the following format for exceptions:
+        #
+        # ```
+        # error_spec.rb:55:in `wrapper': wrapper layer (RuntimeError)
+        # 	from error_spec.rb:40:in `wrapper'
+        # 	from error_spec.rb:61:in `caller'
+        #   ...
+        # ```
+        def backtrace_for(ex, backtrace)
+          trace = ex.backtrace
+          return unless trace
+
+          if trace[0]
+            # Add Exception information to error line
+            backtrace << trace[0]
+            backtrace << ': '
+            backtrace << ex.message
+            backtrace << ' ('
+            backtrace << ex.class.to_s
+            backtrace << ')'
+          end
+
+          if trace[1]
+            # Ident stack trace for caller lines, to separate
+            # them from the main error lines.
+            trace[1..-1].each do |line|
+              backtrace << "\n\tfrom "
+              backtrace << line
+            end
+          end
+
+          backtrace << "\n"
+        end
       end
     end
 
     def initialize(type = nil, message = nil, backtrace = nil)
-      backtrace = Array(backtrace).join("\n")
+      backtrace = Array(backtrace).join("\n") unless backtrace.is_a?(String)
 
-      # DEV: We should measure if `Utils.utf8_encode` is still needed in practice.
       @type = Utils.utf8_encode(type)
       @message = Utils.utf8_encode(message)
       @backtrace = Utils.utf8_encode(backtrace)

data/lib/ddtrace/ext/profiling.rb

@@ -29,6 +29,7 @@ module Datadog
           FORM_FIELD_TAG_ENV = 'env'.freeze
           FORM_FIELD_TAG_HOST = 'host'.freeze
           FORM_FIELD_TAG_LANGUAGE = 'language'.freeze
+          FORM_FIELD_TAG_PID = 'pid'.freeze
           FORM_FIELD_TAG_PROFILER_VERSION = 'profiler_version'.freeze
           FORM_FIELD_TAG_RUNTIME = 'runtime'.freeze
           FORM_FIELD_TAG_RUNTIME_ENGINE = 'runtime_engine'.freeze

data/lib/ddtrace/ext/transport.rb

@@ -4,6 +4,7 @@ module Datadog
       module HTTP
         DEFAULT_HOST = '127.0.0.1'.freeze
         DEFAULT_PORT = 8126
+        DEFAULT_TIMEOUT_SECONDS = 1
         ENV_DEFAULT_HOST = 'DD_AGENT_HOST'.freeze
         ENV_DEFAULT_PORT = 'DD_TRACE_AGENT_PORT'.freeze
         ENV_DEFAULT_URL = 'DD_TRACE_AGENT_URL'.freeze

data/lib/ddtrace/profiling.rb

@@ -4,40 +4,68 @@ 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? || ruby_version_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.ruby_version_unsupported?
+      'Ruby >= 2.1 is required' if Gem::Version.new(RUBY_VERSION) < Gem::Version.new('2.1')
+    end
+    private_class_method :ruby_version_unsupported?
 
-      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_gem_unavailable?
+      if 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?
+
+    def self.protobuf_version_unsupported?
+      if 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
         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 +74,31 @@ 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
@@ -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),
@@ -248,6 +253,7 @@ module Datadog
           end
         end
       end
+      # rubocop:enable Metrics/ClassLength
     end
   end
 end

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

@@ -40,8 +40,6 @@ 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. " \

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_native_thread_id(thread_a)` from within
+        #       `thread_b` will return `thread_b`'s thread ID, not `thread_a`'s.
+        def self.get_native_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
@@ -27,8 +47,30 @@ 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 :native_thread_id
+        else
+          def native_thread_id
+            defined?(@native_thread_id) && @native_thread_id
+          end
+        end
 
         def initialize(*args)
           @pid = ::Process.pid
@@ -48,11 +90,6 @@ 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)
 
@@ -71,36 +108,22 @@ 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
+          @native_thread_id = NativePthread.get_native_thread_id(self)
+          @clock_id = NativePthread.get_clock_id(self, @native_thread_id)
         end
       end