Enable crashtracking without profiler

.gitlab/install_datadog_deps.rb

@@ -63,8 +63,21 @@
 
 puts gem_version_mapping
 
-gem_version_mapping.each do |gem, version|
-  env = {}
+env = {
+  'GEM_HOME' => versioned_path.to_s,
+  # Install `datadog` gem locally without its profiling native extension
+  'DD_PROFILING_NO_EXTENSION' => 'true',
+}
+
+[
+  'debase-ruby_core_source',
+  'ffi',
+  'libddwaf',
+  'msgpack',
+  'libdatadog', # libdatadog MUST be installed before datadog to ensure libdatadog native extension is compiled
+  'datadog',
+].each do |gem|
+  version = gem_version_mapping.delete(gem)
 
   gem_install_cmd = "gem install #{gem} "\
     "--version #{version} "\
@@ -73,19 +86,13 @@
 
   case gem
   when 'ffi'
-    gem_install_cmd << "--install-dir #{versioned_path} "
     # Install `ffi` gem with its built-in `libffi` native extension instead of using system's `libffi`
     gem_install_cmd << '-- --disable-system-libffi '
   when 'datadog'
-    # Install `datadog` gem locally without its profiling native extension
-    env['DD_PROFILING_NO_EXTENSION'] = 'true'
     gem_install_cmd =
       "gem install --local #{ENV.fetch('DATADOG_GEM_LOCATION')} "\
       '--no-document '\
       '--ignore-dependencies '\
-      "--install-dir #{versioned_path} "
-  else
-    gem_install_cmd << "--install-dir #{versioned_path} "
   end
 
   puts "Execute: #{gem_install_cmd}"
@@ -99,6 +106,14 @@
   end
 end
 
+raise "#{gem_version_mapping.keys.join(',')} are not installed." if gem_version_mapping.any?
+
+datadog_gem_path = versioned_path.join("gems/datadog-#{ENV.fetch('RUBY_PACKAGE_VERSION')}")
+libdatadog_so_file = "libdatadog_api.#{RUBY_VERSION[/\d+.\d+/]}_#{RUBY_PLATFORM}.so"
+unless File.exist?("#{datadog_gem_path}/lib/#{libdatadog_so_file}")
+  raise "Missing #{libdatadog_so_file} in #{datadog_gem_path}."
+end
+
 FileUtils.cd(versioned_path.join("extensions/#{Gem::Platform.local}"), verbose: true) do
   # Symlink those directories to be utilized by Ruby compiled with shared libraries
   FileUtils.ln_sf Gem.extension_api_version, ruby_api_version

Matrixfile

@@ -7,6 +7,9 @@
     ''         => '✅ 2.5 / ✅ 2.6 / ✅ 2.7 / ✅ 3.0 / ✅ 3.1 / ✅ 3.2 / ✅ 3.3 / ✅ 3.4 / ✅ jruby',
     'core-old' => '✅ 2.5 / ✅ 2.6 / ✅ 2.7 / ✅ 3.0 / ✅ 3.1 / ✅ 3.2 / ✅ 3.3 / ✅ 3.4 / ✅ jruby'
   },
+  'crashtracking' => {
+    '' => '✅ 2.5 / ✅ 2.6 / ✅ 2.7 / ✅ 3.0 / ✅ 3.1 / ✅ 3.2 / ✅ 3.3 / ✅ 3.4 / ❌ jruby',
+  },
   'appsec:main' => {
     '' => '✅ 2.5 / ✅ 2.6 / ✅ 2.7 / ✅ 3.0 / ✅ 3.1 / ✅ 3.2 / ✅ 3.3 / ✅ 3.4 / ✅ jruby'
   },

Rakefile

@@ -74,12 +74,12 @@ namespace :spec do
   task all: [:main, :benchmark,
              :rails, :railsredis, :railsredis_activesupport, :railsactivejob,
              :elasticsearch, :http, :redis, :sidekiq, :sinatra, :hanami, :hanami_autoinstrument,
-             :profiling]
+             :profiling, :crashtracking]
 
   desc '' # "Explicitly hiding from `rake -T`"
   RSpec::Core::RakeTask.new(:main) do |t, args|
     t.pattern = 'spec/**/*_spec.rb'
-    t.exclude_pattern = 'spec/**/{contrib,benchmark,redis,auto_instrument,opentelemetry,profiling}/**/*_spec.rb,'\
+    t.exclude_pattern = 'spec/**/{contrib,benchmark,redis,auto_instrument,opentelemetry,profiling,crashtracking}/**/*_spec.rb,'\
                         ' spec/**/{auto_instrument,opentelemetry}_spec.rb, spec/datadog/gem_packaging_spec.rb'
     t.rspec_opts = args.to_a.join(' ')
   end
@@ -170,6 +170,15 @@ namespace :spec do
     t.rspec_opts = args.to_a.join(' ')
   end
 
+  # rubocop:disable Style/MultilineBlockChain
+  RSpec::Core::RakeTask.new(:crashtracking) do |t, args|
+    t.pattern = 'spec/datadog/core/crashtracking/**/*_spec.rb'
+    t.rspec_opts = args.to_a.join(' ')
+  end.tap do |t|
+    Rake::Task[t.name].enhance(["compile:libdatadog_api.#{RUBY_VERSION[/\d+.\d+/]}_#{RUBY_PLATFORM}"])
+  end
+  # rubocop:enable Style/MultilineBlockChain
+
   desc '' # "Explicitly hiding from `rake -T`"
   RSpec::Core::RakeTask.new(:contrib) do |t, args|
     contrib_paths = [

ext/libdatadog_api/crashtracker.c

@@ -5,20 +5,21 @@
 
 static VALUE _native_start_or_update_on_fork(int argc, VALUE *argv, DDTRACE_UNUSED VALUE _self);
 static VALUE _native_stop(DDTRACE_UNUSED VALUE _self);
-static void crashtracker_init(VALUE profiling_module);
+static void crashtracker_init(VALUE crashtracking_module);
 
 // Used to report Ruby VM crashes.
 // Once initialized, segfaults will be reported automatically using libdatadog.
 
 void DDTRACE_EXPORT Init_libdatadog_api(void) {
   VALUE datadog_module = rb_define_module("Datadog");
-  VALUE profiling_module = rb_define_module_under(datadog_module, "Profiling");
+  VALUE core_module = rb_define_module_under(datadog_module, "Core");
+  VALUE crashtracking_module = rb_define_module_under(core_module, "Crashtracking");
 
-  crashtracker_init(profiling_module);
+  crashtracker_init(crashtracking_module);
 }
 
-void crashtracker_init(VALUE profiling_module) {
-  VALUE crashtracker_class = rb_define_class_under(profiling_module, "Crashtracker", rb_cObject);
+void crashtracker_init(VALUE crashtracking_module) {
+  VALUE crashtracker_class = rb_define_class_under(crashtracking_module, "Component", rb_cObject);
 
   rb_define_singleton_method(crashtracker_class, "_native_start_or_update_on_fork", _native_start_or_update_on_fork, -1);
   rb_define_singleton_method(crashtracker_class, "_native_stop", _native_stop, 0);

lib/datadog/core/configuration/components.rb

@@ -13,6 +13,7 @@
 require_relative '../../tracing/component'
 require_relative '../../profiling/component'
 require_relative '../../appsec/component'
+require_relative '../crashtracking/component'
 
 module Datadog
   module Core
@@ -58,6 +59,17 @@ def build_runtime_metrics_worker(settings)
           def build_telemetry(settings, agent_settings, logger)
             Telemetry::Component.build(settings, agent_settings, logger)
           end
+
+          def build_crashtracker(settings, agent_settings, logger:)
+            return unless settings.crashtracking.enabled
+
+            if (libdatadog_api_failure = Datadog::Core::Crashtracking::Component::LIBDATADOG_API_FAILURE)
+              logger.debug("Cannot enable crashtracking: #{libdatadog_api_failure}")
+              return
+            end
+
+            Datadog::Core::Crashtracking::Component.build(settings, agent_settings, logger: logger)
+          end
         end
 
         include Datadog::Tracing::Component::InstanceMethods
@@ -70,6 +82,7 @@ def build_telemetry(settings, agent_settings, logger)
           :runtime_metrics,
           :telemetry,
           :tracer,
+          :crashtracker,
           :appsec
 
         def initialize(settings)
@@ -83,11 +96,12 @@ def initialize(settings)
 
           @remote = Remote::Component.build(settings, agent_settings)
           @tracer = self.class.build_tracer(settings, agent_settings, logger: @logger)
+          @crashtracker = self.class.build_crashtracker(settings, agent_settings, logger: @logger)
 
           @profiler, profiler_logger_extra = Datadog::Profiling::Component.build_profiler_component(
             settings: settings,
             agent_settings: agent_settings,
-            optional_tracer: @tracer,
+            optional_tracer: @tracer
           )
           @environment_logger_extra.merge!(profiler_logger_extra) if profiler_logger_extra
 

lib/datadog/core/configuration/settings.rb

@@ -451,17 +451,16 @@ def initialize(*_)
               o.default 60
             end
 
-            # Enables reporting of information when the Ruby VM crashes.
-            #
-            # This feature is no longer experimental, and we plan to deprecate this setting and replace it with a
-            # properly-named one soon.
-            #
-            # @default `DD_PROFILING_EXPERIMENTAL_CRASH_TRACKING_ENABLED` environment variable as a boolean,
-            # otherwise `true`
+            # DEV-3.0: Remove `experimental_crash_tracking_enabled` option
             option :experimental_crash_tracking_enabled do |o|
-              o.type :bool
-              o.env 'DD_PROFILING_EXPERIMENTAL_CRASH_TRACKING_ENABLED'
-              o.default true
+              o.after_set do |_, _, precedence|
+                unless precedence == Datadog::Core::Configuration::Option::Precedence::DEFAULT
+                  Core.log_deprecation(key: :experimental_crash_tracking_enabled) do
+                    'The profiling.advanced.experimental_crash_tracking_enabled setting has been deprecated for removal '\
+                    'and no longer does anything. Please remove it from your Datadog.configure block.'
+                  end
+                end
+              end
             end
           end
 
@@ -833,6 +832,15 @@ def initialize(*_)
           option :service
         end
 
+        settings :crashtracking do
+          # Enables reporting of information when Ruby VM crashes.
+          option :enabled do |o|
+            o.type :bool
+            o.default true
+            o.env 'DD_CRASHTRACKING_ENABLED'
+          end
+        end
+
         # TODO: Tracing should manage its own settings.
         #       Keep this extension here for now to keep things working.
         extend Datadog::Tracing::Configuration::Settings

lib/datadog/core/crashtracking/agent_base_url.rb

@@ -0,0 +1,21 @@
+# frozen_string_literal: true
+
+require_relative '../configuration/ext'
+
+module Datadog
+  module Core
+    module Crashtracking
+      # This module provides a method to resolve the base URL of the agent
+      module AgentBaseUrl
+        def self.resolve(agent_settings)
+          case agent_settings.adapter
+          when Datadog::Core::Configuration::Ext::Agent::HTTP::ADAPTER
+            "#{agent_settings.ssl ? 'https' : 'http'}://#{agent_settings.hostname}:#{agent_settings.port}/"
+          when Datadog::Core::Configuration::Ext::Agent::UnixSocket::ADAPTER
+            "unix://#{agent_settings.uds_path}"
+          end
+        end
+      end
+    end
+  end
+end

lib/datadog/core/crashtracking/component.rb

@@ -0,0 +1,111 @@
+# frozen_string_literal: true
+
+require 'libdatadog'
+
+require_relative 'tag_builder'
+require_relative 'agent_base_url'
+require_relative '../utils/only_once'
+require_relative '../utils/at_fork_monkey_patch'
+
+module Datadog
+  module Core
+    module Crashtracking
+      # Used to report Ruby VM crashes.
+      #
+      # NOTE: The crashtracker native state is a singleton;
+      # so even if you create multiple instances of `Crashtracking::Component` and start them,
+      # it only works as "last writer wins". Same for stop -- there's only one state, so calling stop
+      # on it will stop the crash tracker, regardless of which instance started it.
+      #
+      # Methods prefixed with _native_ are implemented in `crashtracker.c`
+      class Component
+        LIBDATADOG_API_FAILURE =
+          begin
+            require "libdatadog_api.#{RUBY_VERSION[/\d+.\d+/]}_#{RUBY_PLATFORM}"
+            nil
+          rescue LoadError => e
+            e.message
+          end
+
+        ONLY_ONCE = Core::Utils::OnlyOnce.new
+
+        def self.build(settings, agent_settings, logger:)
+          tags = TagBuilder.call(settings)
+          agent_base_url = AgentBaseUrl.resolve(agent_settings)
+          logger.warn('Missing agent base URL; cannot enable crash tracking') unless agent_base_url
+
+          ld_library_path = ::Libdatadog.ld_library_path
+          logger.warn('Missing ld_library_path; cannot enable crash tracking') unless ld_library_path
+
+          path_to_crashtracking_receiver_binary = ::Libdatadog.path_to_crashtracking_receiver_binary
+          unless path_to_crashtracking_receiver_binary
+            logger.warn('Missing path_to_crashtracking_receiver_binary; cannot enable crash tracking')
+          end
+
+          return unless agent_base_url
+          return unless ld_library_path
+          return unless path_to_crashtracking_receiver_binary
+
+          new(
+            tags: tags,
+            agent_base_url: agent_base_url,
+            ld_library_path: ld_library_path,
+            path_to_crashtracking_receiver_binary: path_to_crashtracking_receiver_binary,
+            logger: logger
+          ).tap(&:start)
+        end
+
+        def initialize(tags:, agent_base_url:, ld_library_path:, path_to_crashtracking_receiver_binary:, logger:)
+          @tags = tags
+          @agent_base_url = agent_base_url
+          @ld_library_path = ld_library_path
+          @path_to_crashtracking_receiver_binary = path_to_crashtracking_receiver_binary
+          @logger = logger
+        end
+
+        def start
+          Utils::AtForkMonkeyPatch.apply!
+
+          start_or_update_on_fork(action: :start)
+          ONLY_ONCE.run do
+            Utils::AtForkMonkeyPatch.at_fork(:child) do
+              # Must NOT reference `self` here, as only the first instance will
+              # be captured by the ONLY_ONCE and we want to pick the latest active one
+              # (which may have different tags or agent config)
+              Datadog.send(:components).crashtracker&.update_on_fork
+            end
+          end
+        end
+
+        def update_on_fork
+          start_or_update_on_fork(action: :update_on_fork)
+        end
+
+        def stop
+          self.class._native_stop
+          logger.debug('Crash tracking stopped successfully')
+        rescue => e
+          logger.error("Failed to stop crash tracking: #{e.message}")
+        end
+
+        private
+
+        attr_reader :tags, :agent_base_url, :ld_library_path, :path_to_crashtracking_receiver_binary, :logger
+
+        def start_or_update_on_fork(action:)
+          self.class._native_start_or_update_on_fork(
+            action: action,
+            exporter_configuration: [:agent, agent_base_url],
+            path_to_crashtracking_receiver_binary: path_to_crashtracking_receiver_binary,
+            ld_library_path: ld_library_path,
+            tags_as_array: tags.to_a,
+            upload_timeout_seconds: 1
+          )
+          logger.debug("Crash tracking #{action} successfully")
+        rescue => e
+          logger.error("Failed to #{action} crash tracking: #{e.message}")
+        end
+      end
+    end
+  end
+end

lib/datadog/core/crashtracking/tag_builder.rb

@@ -0,0 +1,37 @@
+# frozen_string_literal: true
+
+require_relative '../utils'
+require_relative '../environment/socket'
+require_relative '../environment/identity'
+require_relative '../environment/git'
+
+module Datadog
+  module Core
+    module Crashtracking
+      # This module builds a hash of tags
+      module TagBuilder
+        def self.call(settings)
+          hash = {
+            'host' => Environment::Socket.hostname,
+            'process_id' => Process.pid.to_s,
+            'runtime_engine' => Environment::Identity.lang_engine,
+            'runtime-id' => Environment::Identity.id,
+            'runtime_platform' => Environment::Identity.lang_platform,
+            'runtime_version' => Environment::Identity.lang_version,
+            'env' => settings.env,
+            'service' => settings.service,
+            'version' => settings.version,
+            'git.repository_url' => Environment::Git.git_repository_url,
+            'git.commit.sha' => Environment::Git.git_commit_sha,
+            'is_crash' => true
+          }.compact
+
+          # Make sure everything is an utf-8 string, to avoid encoding issues in downstream
+          settings.tags.merge(hash).each_with_object({}) do |(key, value), h|
+            h[Utils.utf8_encode(key)] = Utils.utf8_encode(value)
+          end
+        end
+      end
+    end
+  end
+end

lib/datadog/profiling.rb

@@ -144,7 +144,6 @@ def self.allocation_count # rubocop:disable Lint/NestedMethodDefinition (On purp
       require_relative 'profiling/collectors/idle_sampling_helper'
       require_relative 'profiling/collectors/stack'
       require_relative 'profiling/collectors/thread_context'
-      require_relative 'profiling/crashtracker'
       require_relative 'profiling/stack_recorder'
       require_relative 'profiling/exporter'
       require_relative 'profiling/flush'