datadog 2.21.0 → 2.38.0

data/lib/datadog/di/logger.rb

@@ -8,7 +8,7 @@ module Datadog
     #
     # @api private
     class Logger
-      extend Forwardable # steep:ignore
+      extend Forwardable
 
       def initialize(settings, target)
         @settings = settings
@@ -18,7 +18,7 @@ module Datadog
       attr_reader :settings
       attr_reader :target
 
-      def_delegators :target, :debug # steep:ignore
+      def_delegators :target, :debug
 
       def trace(&block)
         if settings.dynamic_instrumentation.internal.trace_logging

data/lib/datadog/di/probe.rb

@@ -17,7 +17,7 @@ module Datadog
     # and remote config code must be prepared to deal with exceptions
     # raised by Probe constructor in particular. Therefore, Probe constructor
     # will raise an exception if it determines that there is not enough
-    # information (or confilcting information) in the arguments to create a
+    # information (or conflicting information) in the arguments to create a
     # functional probe, and upstream code is tasked with not spamming logs
     # with notifications of such errors (and potentially limiting the
     # attempts to construct probe from a given payload).
@@ -34,10 +34,17 @@ module Datadog
     class Probe
       KNOWN_TYPES = %i[log].freeze
 
+      EVALUATE_AT_VALUES = %i[entry exit].freeze
+
       def initialize(id:, type:,
         file: nil, line_no: nil, type_name: nil, method_name: nil,
-        template: nil, capture_snapshot: false, max_capture_depth: nil,
+        template: nil, template_segments: nil,
+        capture_snapshot: false, max_capture_depth: nil,
         max_capture_attribute_count: nil,
+        max_capture_collection_size: nil, max_capture_string_length: nil,
+        capture_expressions: [],
+        evaluate_at: nil,
+        condition: nil,
         rate_limit: nil)
         # Perform some sanity checks here to detect unexpected attribute
         # combinations, in order to not do them in subsequent code.
@@ -45,9 +52,17 @@ module Datadog
           raise ArgumentError, "Unknown probe type: #{type}"
         end
 
-        if line_no && method_name
-          raise ArgumentError, "Probe contains both line number and method name: #{id}"
-        end
+        # Probe should be inferred to be a line probe if the specification
+        # contains a line number. This how Java tracer works and Go tracer
+        # is implementing the same behavior, and Go will have all 3 fields
+        # (file path, line number and method name) for line probes.
+        # Do not raise if line number and method name both exist - instead
+        # treat the probe as a line probe.
+        #
+        # In the future we want to provide type name and method name to line
+        # probes, so that the library can verify that the instrumented line
+        # is in the method that the frontend showed to the user when the
+        # user created the probe.
 
         if line_no && !file
           raise ArgumentError, "Probe contains line number but not file: #{id}"
@@ -57,6 +72,10 @@ module Datadog
           raise ArgumentError, "Partial method probe definition: #{id}"
         end
 
+        if line_no.nil? && method_name.nil?
+          raise ArgumentError, "Unhandled probe type: neither method nor line probe: #{id}"
+        end
+
         @id = id
         @type = type
         @file = file
@@ -64,22 +83,35 @@ module Datadog
         @type_name = type_name
         @method_name = method_name
         @template = template
+        @template_segments = template_segments
         @capture_snapshot = !!capture_snapshot
         @max_capture_depth = max_capture_depth
         @max_capture_attribute_count = max_capture_attribute_count
-
-        # These checks use instance methods that have more complex logic
-        # than checking a single argument value. To avoid duplicating
-        # the logic here, use the methods and perform these checks after
-        # instance variable assignment.
-        unless method? || line?
-          raise ArgumentError, "Unhandled probe type: neither method nor line probe: #{id}"
+        @max_capture_collection_size = max_capture_collection_size
+        @max_capture_string_length = max_capture_string_length
+        @capture_expressions = capture_expressions || []
+        evaluate_at = :exit if evaluate_at.nil?
+        unless EVALUATE_AT_VALUES.include?(evaluate_at)
+          raise ArgumentError, "Unknown evaluate_at value: #{evaluate_at.inspect} (expected one of #{EVALUATE_AT_VALUES.inspect})"
         end
+        @evaluate_at = evaluate_at
+        @condition = condition
 
-        @rate_limit = rate_limit || (@capture_snapshot ? 1 : 5000)
+        @rate_limit = rate_limit || ((@capture_snapshot || !@capture_expressions.empty?) ? 1 : 5000)
         @rate_limiter = Datadog::Core::TokenBucket.new(@rate_limit)
 
+        # At most one report per second.
+        # We create the rate limiter here even though it may never be used,
+        # to avoid having to synchronize the creation since method probes
+        # can be executed on multiple threads concurrently (even if line
+        # probes are never executed concurrently since those are done in a
+        # trace point).
+        if condition
+          @condition_evaluation_failed_rate_limiter = Datadog::Core::TokenBucket.new(1)
+        end
+
         @emitting_notified = false
+        @enabled = true
       end
 
       attr_reader :id
@@ -89,6 +121,10 @@ module Datadog
       attr_reader :type_name
       attr_reader :method_name
       attr_reader :template
+      attr_reader :template_segments
+
+      # The compiled condition for the probe, as a String.
+      attr_reader :condition
 
       # Configured maximum capture depth. Can be nil in which case
       # the global default will be used.
@@ -98,16 +134,57 @@ module Datadog
       # the global default will be used.
       attr_reader :max_capture_attribute_count
 
+      attr_reader :max_capture_collection_size
+
+      attr_reader :max_capture_string_length
+
+      attr_reader :capture_expressions
+
+      attr_reader :evaluate_at
+
       # Rate limit in effect, in invocations per second. Always present.
       attr_reader :rate_limit
 
       # Rate limiter object. For internal DI use only.
       attr_reader :rate_limiter
 
+      # Rate limiter object for sending snapshots with evaluation errors
+      # for when probe condition evaluation fails.
+      # This rate limit is separate from the "base" rate limit for the probe
+      # because when the condition evaluation succeeds we want the "base"
+      # rate limit applied, not tainted by any evaluation errors
+      # (for example, the condition can be highly selective, and when it
+      # does not hold the evaluation may fail - we don't want to use up the
+      # probe rate limit for the errors).
+      attr_reader :condition_evaluation_failed_rate_limiter
+
       def capture_snapshot?
         @capture_snapshot
       end
 
+      def capture_expressions?
+        !@capture_expressions.empty?
+      end
+
+      def evaluate_at_entry?
+        evaluate_at == :entry
+      end
+
+      # Capture-expression mode: the probe captures expressions but not a full
+      # snapshot. Snapshot capture serializes its own values, so expression
+      # capture only runs when a snapshot is not being taken.
+      def capture_expressions_only?
+        capture_expressions? && !capture_snapshot?
+      end
+
+      def capture_entry_expressions?
+        capture_expressions_only? && evaluate_at_entry?
+      end
+
+      def snapshot_serializer_limits(settings)
+        CaptureLimits.resolve(expr_limits: nil, probe: self, settings: settings)
+      end
+
       # Returns whether the probe is a line probe.
       #
       # Method probes may still specify a file name (to aid in locating the
@@ -122,7 +199,7 @@ module Datadog
 
       # Returns whether the probe is a method probe.
       def method?
-        !!(type_name && method_name)
+        line_no.nil?
       end
 
       # Returns the line number associated with the probe, raising
@@ -139,6 +216,20 @@ module Datadog
         line_no
       end
 
+      # Returns the method name associated with the probe, raising
+      # Error::MissingMethodName if the probe does not have a method name
+      # associated with it.
+      #
+      # This method is used by instrumentation driver to ensure a method name
+      # that is passed into the instrumentation logic is actually a method name
+      # and not nil.
+      def method_name!
+        if method_name.nil?
+          raise Error::MissingMethodName, "Probe #{id} does not have a method name associated with it"
+        end
+        method_name
+      end
+
       # Source code location of the probe, for diagnostic reporting.
       def location
         if method?
@@ -186,6 +277,23 @@ module Datadog
       def emitting_notified?
         !!@emitting_notified
       end
+
+      def executed_on_line?
+        !!(defined?(@executed_on_line) && @executed_on_line)
+      end
+
+      def executed_on_line!
+        # TODO lock?
+        @executed_on_line = true
+      end
+
+      def enabled?
+        @enabled
+      end
+
+      def disable!
+        @enabled = false
+      end
     end
   end
 end

data/lib/datadog/di/probe_builder.rb

@@ -1,6 +1,11 @@
 # frozen_string_literal: true
 
+# rubocop:disable Lint/AssignmentInCondition
+
 require_relative "probe"
+require_relative "capture_expression"
+require_relative "capture_limits"
+require_relative 'el'
 
 module Datadog
   module DI
@@ -17,14 +22,39 @@ module Datadog
     #
     # @api private
     module ProbeBuilder
-      PROBE_TYPES = {
+      # Steep: https://github.com/soutaro/steep/issues/363
+      PROBE_TYPES = { # steep:ignore IncompatibleAssignment
         'LOG_PROBE' => :log,
       }.freeze
 
-      module_function def build_from_remote_config(config)
+      module_function
+
+      CAPTURE_EXPRESSION_NAME_PATTERN = /\A[a-zA-Z0-9_?]+\z/
+
+      EVALUATE_AT_STRINGS = {
+        "ENTRY" => :entry,
+        "EXIT" => :exit,
+        "DEFAULT" => :exit,
+      }.freeze
+
+      def build_from_remote_config(config, logger:)
         # The validations here are not yet comprehensive.
         type = config.fetch('type')
         type_symbol = PROBE_TYPES[type] or raise ArgumentError, "Unrecognized probe type: #{type}"
+        cond = if cond_spec = config['when']
+          unless cond_spec['dsl'] && cond_spec['json']
+            raise ArgumentError, "Malformed condition specification for probe: #{config}"
+          end
+          compiled = EL::Compiler.new.compile(cond_spec['json'])
+          EL::Expression.new(cond_spec['dsl'], compiled)
+        end
+        capture_expressions = build_capture_expressions(config['captureExpressions'])
+        capture_expressions = dedup_capture_expressions(capture_expressions, config["id"], logger)
+        if !!config["captureSnapshot"] && !capture_expressions.empty?
+          logger.debug do
+            "di: probe #{config["id"]}: captureSnapshot=true wins over captureExpressions (n=#{capture_expressions.size})"
+          end
+        end
         Probe.new(
           id: config.fetch("id"),
           type: type_symbol,
@@ -34,15 +64,106 @@ module Datadog
           line_no: config["where"]&.[]("lines")&.compact&.map(&:to_i)&.first,
           type_name: config["where"]&.[]("typeName"),
           method_name: config["where"]&.[]("methodName"),
+          # We should not be using the template for anything - we instead
+          # use +segments+ - but keep the template for debugging.
           template: config["template"],
+          template_segments: build_template_segments(config['segments']),
           capture_snapshot: !!config["captureSnapshot"],
           max_capture_depth: config["capture"]&.[]("maxReferenceDepth"),
           max_capture_attribute_count: config["capture"]&.[]("maxFieldCount"),
+          max_capture_collection_size: config["capture"]&.[]("maxCollectionSize"),
+          max_capture_string_length: config["capture"]&.[]("maxLength"),
+          capture_expressions: capture_expressions,
+          evaluate_at: parse_evaluate_at(config["evaluateAt"], config["id"], logger),
           rate_limit: config["sampling"]&.[]("snapshotsPerSecond"),
+          condition: cond,
         )
       rescue KeyError => exc
-        raise ArgumentError, "Malformed remote configuration entry for probe: #{exc.class}: #{exc}: #{config}"
+        raise ArgumentError, "Malformed remote configuration entry for probe: #{exc.class}: #{exc.message}: #{config}"
+      end
+
+      def build_capture_expressions(raw)
+        return [] if raw.nil?
+        unless Array === raw
+          raise ArgumentError, "captureExpressions must be an array, got: #{raw.class}"
+        end
+        return [] if raw.empty?
+        raw.map do |entry|
+          unless Hash === entry
+            raise ArgumentError, "captureExpressions entry must be a hash, got: #{entry.class}"
+          end
+          name = entry['name']
+          unless String === name && CAPTURE_EXPRESSION_NAME_PATTERN.match?(name)
+            raise ArgumentError, "captureExpressions entry name missing or invalid (must match #{CAPTURE_EXPRESSION_NAME_PATTERN.inspect}): #{name.inspect}"
+          end
+          expr_spec = entry['expr']
+          unless Hash === expr_spec && expr_spec['dsl'] && expr_spec['json']
+            raise ArgumentError, "captureExpressions entry #{name}: missing or malformed expr"
+          end
+          compiled = EL::Compiler.new.compile(expr_spec['json'])
+          expr = EL::Expression.new(expr_spec['dsl'], compiled)
+          limits = build_capture_limits(entry['capture'])
+          CaptureExpression.new(name: name, expr: expr, limits: limits)
+        end
+      end
+
+      def dedup_capture_expressions(capture_expressions, probe_id, logger)
+        return capture_expressions if capture_expressions.size < 2
+        by_name = {}
+        capture_expressions.each do |capture_expression|
+          by_name[capture_expression.name] = capture_expression
+        end
+        return capture_expressions if by_name.size == capture_expressions.size
+        logger.debug do
+          "di: probe #{probe_id}: collapsed duplicate captureExpressions names, kept last per name (#{capture_expressions.size} -> #{by_name.size})"
+        end
+        by_name.values
+      end
+
+      def build_capture_limits(raw)
+        return nil if raw.nil?
+        unless Hash === raw
+          raise ArgumentError, "capture-expression entry capture must be a hash, got: #{raw.class}"
+        end
+        CaptureLimits.new(
+          max_reference_depth: raw['maxReferenceDepth'],
+          max_collection_size: raw['maxCollectionSize'],
+          max_length: raw['maxLength'],
+          max_field_count: raw['maxFieldCount'],
+        )
+      end
+
+      def parse_evaluate_at(raw, probe_id, logger)
+        return :exit if raw.nil?
+        EVALUATE_AT_STRINGS[raw] || begin
+          logger.debug do
+            "di: probe #{probe_id}: unrecognized evaluateAt value #{raw.inspect}, defaulting to :exit"
+          end
+          :exit
+        end
+      end
+
+      def build_template_segments(segments)
+        segments&.map do |segment|
+          if Hash === segment
+            if str = segment['str']
+              str
+            elsif ast = segment['json']
+              unless dsl = segment['dsl']
+                raise ArgumentError, "Missing dsl for json in segment: #{segment}"
+              end
+              compiled = EL::Compiler.new.compile(ast)
+              EL::Expression.new(dsl, compiled)
+            else
+              # TODO report to telemetry?
+            end
+          else
+            # TODO report to telemetry?
+          end
+        end&.compact
       end
     end
   end
 end
+
+# rubocop:enable Lint/AssignmentInCondition

data/lib/datadog/di/probe_file_loader/railtie.rb

@@ -6,7 +6,7 @@ module Datadog
       # Railtie class initializes dynamic instrumentation contrib code
       # in Rails environments.
       class Railtie < Rails::Railtie
-        initializer 'datadog.dynamic_instrumentation.load_probe_file' do |app| # steep:ignore
+        initializer 'datadog.dynamic_instrumentation.load_probe_file' do |app|
           ProbeFileLoader.load_now
         end
       end

data/lib/datadog/di/probe_file_loader.rb

@@ -2,6 +2,8 @@
 
 require 'json'
 
+require_relative 'fatal_exceptions'
+
 module Datadog
   module DI
     module ProbeFileLoader
@@ -46,28 +48,31 @@ module Datadog
               component.telemetry&.report(exc, description: "Line probe is targeting a loaded file that is not in code tracker")
 
               payload = component.probe_notification_builder.build_errored(probe, exc)
-              component.probe_notifier_worker.add_status(payload)
-            rescue => exc
+              component.probe_notifier_worker.add_status(payload, probe: probe)
+            rescue Exception => exc # standard:disable Lint/RescueException
+              Datadog::DI.reraise_if_fatal(exc)
               raise if component.settings.dynamic_instrumentation.internal.propagate_all_exceptions
 
-              component.logger.debug { "di: unhandled exception adding #{probe.type} probe at #{probe.location} (#{probe.id}) in DI probe file loader: #{exc.class}: #{exc}" }
+              component.logger.debug { "di: unhandled exception adding #{probe.type} probe at #{probe.location} (#{probe.id}) in DI probe file loader: #{exc.class}: #{exc.message}" }
               component.telemetry&.report(exc, description: "Unhandled exception adding probe in DI probe file loader")
 
               # TODO test this path
               payload = component.probe_notification_builder.build_errored(probe, exc)
-              component.probe_notifier_worker.add_status(payload)
+              component.probe_notifier_worker.add_status(payload, probe: probe)
             end
           end
-        rescue => exc
+        rescue Exception => exc # standard:disable Lint/RescueException
+          Datadog::DI.reraise_if_fatal(exc)
           if component.settings.dynamic_instrumentation.internal.propagate_all_exceptions
             should_propagate = true
             raise
           end
 
-          component.logger.debug { "di: unhandled exception handling a probe in DI probe file loader: #{exc.class}: #{exc}" }
+          component.logger.debug { "di: unhandled exception handling a probe in DI probe file loader: #{exc.class}: #{exc.message}" }
           component.telemetry&.report(exc, description: "Unhandled exception handling probe in DI probe file loader")
         end
-      rescue
+      rescue Exception => exc # standard:disable Lint/RescueException
+        Datadog::DI.reraise_if_fatal(exc)
         # We should generally never get here, but if component tree
         # initialization fails for some unexpected reason, don't nuke
         # the customer application.