ddtrace 0.29.1 → 0.30.0

data/lib/ddtrace/contrib/sinatra/patcher.rb

@@ -9,19 +9,13 @@ module Datadog
 
         module_function
 
-        def patched?
-          done?(:sinatra)
+        def target_version
+          Integration.version
         end
 
         def patch
-          do_once(:sinatra) do
-            begin
-              require 'ddtrace/contrib/sinatra/tracer'
-              register_tracer
-            rescue StandardError => e
-              Datadog::Tracer.log.error("Unable to apply Sinatra integration: #{e}")
-            end
-          end
+          require 'ddtrace/contrib/sinatra/tracer'
+          register_tracer
         end
 
         def register_tracer

data/lib/ddtrace/contrib/sucker_punch/patcher.rb

@@ -11,23 +11,17 @@ module Datadog
 
         module_function
 
-        def patched?
-          done?(:sucker_punch)
+        def target_version
+          Integration.version
         end
 
         def patch
-          do_once(:sucker_punch) do
-            begin
-              require 'ddtrace/contrib/sucker_punch/exception_handler'
-              require 'ddtrace/contrib/sucker_punch/instrumentation'
+          require 'ddtrace/contrib/sucker_punch/exception_handler'
+          require 'ddtrace/contrib/sucker_punch/instrumentation'
 
-              add_pin!
-              ExceptionHandler.patch!
-              Instrumentation.patch!
-            rescue StandardError => e
-              Datadog::Tracer.log.error("Unable to apply SuckerPunch integration: #{e}")
-            end
-          end
+          add_pin!
+          ExceptionHandler.patch!
+          Instrumentation.patch!
         end
 
         def add_pin!

data/lib/ddtrace/diagnostics/health.rb

@@ -10,14 +10,21 @@ module Datadog
         count :api_errors, Ext::Diagnostics::Health::Metrics::METRIC_API_ERRORS
         count :api_requests, Ext::Diagnostics::Health::Metrics::METRIC_API_REQUESTS
         count :api_responses, Ext::Diagnostics::Health::Metrics::METRIC_API_RESPONSES
+        count :error_context_overflow, Ext::Diagnostics::Health::Metrics::METRIC_ERROR_CONTEXT_OVERFLOW
+        count :error_instrumentation_patch, Ext::Diagnostics::Health::Metrics::METRIC_ERROR_INSTRUMENTATION_PATCH
+        count :error_span_finish, Ext::Diagnostics::Health::Metrics::METRIC_ERROR_SPAN_FINISH
+        count :error_unfinished_spans, Ext::Diagnostics::Health::Metrics::METRIC_ERROR_UNFINISHED_SPANS
+        count :instrumentation_patched, Ext::Diagnostics::Health::Metrics::METRIC_INSTRUMENTATION_PATCHED
         count :queue_accepted, Ext::Diagnostics::Health::Metrics::METRIC_QUEUE_ACCEPTED
         count :queue_accepted_lengths, Ext::Diagnostics::Health::Metrics::METRIC_QUEUE_ACCEPTED_LENGTHS
         count :queue_dropped, Ext::Diagnostics::Health::Metrics::METRIC_QUEUE_DROPPED
+        count :traces_filtered, Ext::Diagnostics::Health::Metrics::METRIC_TRACES_FILTERED
+        count :writer_cpu_time, Ext::Diagnostics::Health::Metrics::METRIC_WRITER_CPU_TIME
+
         gauge :queue_length, Ext::Diagnostics::Health::Metrics::METRIC_QUEUE_LENGTH
         gauge :queue_max_length, Ext::Diagnostics::Health::Metrics::METRIC_QUEUE_MAX_LENGTH
         gauge :queue_spans, Ext::Diagnostics::Health::Metrics::METRIC_QUEUE_SPANS
-        count :traces_filtered, Ext::Diagnostics::Health::Metrics::METRIC_TRACES_FILTERED
-        count :writer_cpu_time, Ext::Diagnostics::Health::Metrics::METRIC_WRITER_CPU_TIME
+        gauge :sampling_service_cache_length, Ext::Diagnostics::Health::Metrics::METRIC_SAMPLING_SERVICE_CACHE_LENGTH
       end
 
       module_function

data/lib/ddtrace/ext/diagnostics.rb

@@ -10,12 +10,18 @@ module Datadog
           METRIC_API_ERRORS = 'datadog.tracer.api.errors'.freeze
           METRIC_API_REQUESTS = 'datadog.tracer.api.requests'.freeze
           METRIC_API_RESPONSES = 'datadog.tracer.api.responses'.freeze
+          METRIC_ERROR_CONTEXT_OVERFLOW = 'datadog.tracer.error.context_overflow'.freeze
+          METRIC_ERROR_INSTRUMENTATION_PATCH = 'datadog.tracer.error.instrumentation_patch'.freeze
+          METRIC_ERROR_SPAN_FINISH = 'datadog.tracer.error.span_finish'.freeze
+          METRIC_ERROR_UNFINISHED_SPANS = 'datadog.tracer.error.unfinished_spans'.freeze
+          METRIC_INSTRUMENTATION_PATCHED = 'datadog.tracer.instrumentation_patched'.freeze
           METRIC_QUEUE_ACCEPTED = 'datadog.tracer.queue.accepted'.freeze
           METRIC_QUEUE_ACCEPTED_LENGTHS = 'datadog.tracer.queue.accepted_lengths'.freeze
           METRIC_QUEUE_DROPPED = 'datadog.tracer.queue.dropped'.freeze
           METRIC_QUEUE_LENGTH = 'datadog.tracer.queue.length'.freeze
           METRIC_QUEUE_MAX_LENGTH = 'datadog.tracer.queue.max_length'.freeze
           METRIC_QUEUE_SPANS = 'datadog.tracer.queue.spans'.freeze
+          METRIC_SAMPLING_SERVICE_CACHE_LENGTH = 'datadog.tracer.sampling.service_cache_length'.freeze
           METRIC_TRACES_FILTERED = 'datadog.tracer.traces.filtered'.freeze
           METRIC_WRITER_CPU_TIME = 'datadog.tracer.writer.cpu_time'.freeze
         end

data/lib/ddtrace/ext/sampling.rb

@@ -0,0 +1,13 @@
+module Datadog
+  module Ext
+    module Sampling
+      # If rule sampling is applied to a span, set this metric the sample rate configured for that rule.
+      # This should be done regardless of sampling outcome.
+      RULE_SAMPLE_RATE = '_dd.rule_psr'.freeze
+
+      # If rate limiting is checked on a span, set this metric the effective rate limiting rate applied.
+      # This should be done regardless of rate limiting outcome.
+      RATE_LIMITER_RATE = '_dd.limit_psr'.freeze
+    end
+  end
+end

data/lib/ddtrace/sampler.rb

@@ -1,6 +1,7 @@
 require 'forwardable'
 
 require 'ddtrace/ext/priority'
+require 'ddtrace/diagnostics/health'
 
 module Datadog
   # \Sampler performs client-side trace sampling.
@@ -12,6 +13,10 @@ module Datadog
     def sample!(_span)
       raise NotImplementedError, 'Samplers must implement the #sample! method'
     end
+
+    def sample_rate(span)
+      raise NotImplementedError, 'Samplers must implement the #sample_rate method'
+    end
   end
 
   # \AllSampler samples all the traces.
@@ -23,6 +28,10 @@ module Datadog
     def sample!(span)
       span.sampled = true
     end
+
+    def sample_rate(*_)
+      1.0
+    end
   end
 
   # \RateSampler is based on a sample rate.
@@ -30,8 +39,6 @@ module Datadog
     KNUTH_FACTOR = 1111111111111111111
     SAMPLE_RATE_METRIC_KEY = '_sample_rate'.freeze
 
-    attr_reader :sample_rate
-
     # Initialize a \RateSampler.
     # This sampler keeps a random subset of the traces. Its main purpose is to
     # reduce the instrumentation footprint.
@@ -48,6 +55,10 @@ module Datadog
       self.sample_rate = sample_rate
     end
 
+    def sample_rate(*_)
+      @sample_rate
+    end
+
     def sample_rate=(sample_rate)
       @sample_rate = sample_rate
       @sampling_id_threshold = sample_rate * Span::MAX_ID
@@ -136,6 +147,10 @@ module Datadog
       end
     end
 
+    def length
+      @samplers.length
+    end
+
     private
 
     def set_rate(key, rate)
@@ -159,6 +174,9 @@ module Datadog
 
       # Update each service rate
       update_all(rate_by_service)
+
+      # Emit metric for service cache size
+      Diagnostics::Health.metrics.sampling_service_cache_length(length)
     end
 
     private
@@ -195,10 +213,8 @@ module Datadog
         # If priority sampling has already been applied upstream, use that, otherwise...
         unless priority_assigned_upstream?(span)
           # Roll the dice and determine whether how we set the priority.
-          # NOTE: We'll want to leave `span.sampled = true` here; all spans for priority sampling must
-          #       be sent to the agent. Otherwise metrics for traces will not be accurate, since the
-          #       agent will have an incomplete dataset.
-          priority = priority_sample(span) ? Datadog::Ext::Priority::AUTO_KEEP : Datadog::Ext::Priority::AUTO_REJECT
+          priority = priority_sample!(span) ? Datadog::Ext::Priority::AUTO_KEEP : Datadog::Ext::Priority::AUTO_REJECT
+
           assign_priority!(span, priority)
         end
       else
@@ -229,8 +245,33 @@ module Datadog
       span.context && !span.context.sampling_priority.nil?
     end
 
-    def priority_sample(span)
-      @priority_sampler.sample?(span)
+    def priority_sample!(span)
+      preserving_sampling(span) do
+        @priority_sampler.sample!(span)
+      end
+    end
+
+    # Ensures the span is always propagated to the writer and that
+    # the sample rate metric represents the true client-side sampling.
+    def preserving_sampling(span)
+      pre_sample_rate_metric = span.get_metric(SAMPLE_RATE_METRIC_KEY)
+
+      yield.tap do
+        # NOTE: We'll want to leave `span.sampled = true` here; all spans for priority sampling must
+        #       be sent to the agent. Otherwise metrics for traces will not be accurate, since the
+        #       agent will have an incomplete dataset.
+        #
+        #       We also ensure that the agent knows we that our `post_sampler` is not performing true sampling,
+        #       to avoid erroneous metric upscaling.
+        span.sampled = true
+        if pre_sample_rate_metric
+          # Restore true sampling metric, as only the @pre_sampler can reject traces
+          span.set_metric(SAMPLE_RATE_METRIC_KEY, pre_sample_rate_metric)
+        else
+          # If @pre_sampler is not enable, sending this metric would be misleading
+          span.clear_metric(SAMPLE_RATE_METRIC_KEY)
+        end
+      end
     end
 
     def assign_priority!(span, priority)

data/lib/ddtrace/sampling.rb

@@ -0,0 +1,2 @@
+require 'ddtrace/sampling/rule'
+require 'ddtrace/sampling/rule_sampler'

data/lib/ddtrace/sampling/matcher.rb

@@ -0,0 +1,57 @@
+module Datadog
+  module Sampling
+    # Checks if a span conforms to a matching criteria.
+    class Matcher
+      # Returns `true` if the span should conforms to this rule, `false` otherwise
+      #
+      # @abstract
+      # @param [Span] span
+      # @return [Boolean]
+      def match?(span)
+        raise NotImplementedError
+      end
+    end
+
+    # A \Matcher that supports matching a span by
+    # operation name and/or service name.
+    class SimpleMatcher < Matcher
+      # Returns `true` for case equality (===) with any object
+      MATCH_ALL = Class.new do
+        # DEV: A class that implements `#===` is ~20% faster than
+        # DEV: a `Proc` that always returns `true`.
+        def ===(other)
+          true
+        end
+      end.new
+
+      attr_reader :name, :service
+
+      # @param name [String,Regexp,Proc] Matcher for case equality (===) with the span name, defaults to always match
+      # @param service [String,Regexp,Proc] Matcher for case equality (===) with the service name, defaults to always match
+      def initialize(name: MATCH_ALL, service: MATCH_ALL)
+        @name = name
+        @service = service
+      end
+
+      def match?(span)
+        name === span.name && service === span.service
+      end
+    end
+
+    # A \Matcher that allows for arbitrary span matching
+    # based on the return value of a provided block.
+    class ProcMatcher < Matcher
+      attr_reader :block
+
+      # @yield [name, service] Provides span name and service to the block
+      # @yieldreturn [Boolean] Whether the span conforms to this matcher
+      def initialize(&block)
+        @block = block
+      end
+
+      def match?(span)
+        block.call(span.name, span.service)
+      end
+    end
+  end
+end

data/lib/ddtrace/sampling/rate_limiter.rb

@@ -0,0 +1,127 @@
+require 'ddtrace/utils/time'
+
+module Datadog
+  module Sampling
+    # Checks for rate limiting on a resource.
+    class RateLimiter
+      # Checks if resource of specified size can be
+      # conforms with the current limit.
+      #
+      # Implementations of this method are not guaranteed
+      # to be side-effect free.
+      #
+      # @return [Boolean] whether a resource conforms with the current limit
+      def allow?(size); end
+
+      # The effective rate limiting ratio based on
+      # recent calls to `allow?`.
+      #
+      # @return [Float] recent allowance ratio
+      def effective_rate; end
+    end
+
+    # Implementation of the Token Bucket metering algorithm
+    # for rate limiting.
+    #
+    # @see https://en.wikipedia.org/wiki/Token_bucket Token bucket
+    class TokenBucket < RateLimiter
+      attr_reader :rate, :max_tokens
+
+      # @param rate [Numeric] Allowance rate, in units per second
+      #  if rate is negative, always allow
+      #  if rate is zero, never allow
+      # @param max_tokens [Numeric] Limit of available tokens
+      def initialize(rate, max_tokens = rate)
+        @rate = rate
+        @max_tokens = max_tokens
+
+        @tokens = max_tokens
+        @total_messages = 0
+        @conforming_messages = 0
+        @last_refill = Utils::Time.get_time
+      end
+
+      # Checks if a message of provided +size+
+      # conforms with the current bucket limit.
+      #
+      # If it does, return +true+ and remove +size+
+      # tokens from the bucket.
+      # If it does not, return +false+ without affecting
+      # the tokens form the bucket.
+      #
+      # @return [Boolean] +true+ if message conforms with current bucket limit
+      def allow?(size)
+        return false if @rate.zero?
+        return true if @rate < 0
+
+        refill_since_last_message
+
+        increment_total_count
+
+        return false if @tokens < size
+
+        increment_conforming_count
+
+        @tokens -= size
+
+        true
+      end
+
+      # Ratio of 'conformance' per 'total messages' checked
+      # on this bucket.
+      #
+      # Returns +1.0+ when no messages have been checked yet.
+      #
+      # @return [Float] Conformance ratio, between +[0,1]+
+      def effective_rate
+        return 0.0 if @rate.zero?
+        return 1.0 if @rate < 0 || @total_messages.zero?
+
+        @conforming_messages.to_f / @total_messages
+      end
+
+      # @return [Numeric] number of tokens currently available
+      def available_tokens
+        @tokens
+      end
+
+      private
+
+      def refill_since_last_message
+        now = Utils::Time.get_time
+        elapsed = now - @last_refill
+
+        refill_tokens(@rate * elapsed)
+
+        @last_refill = now
+      end
+
+      def refill_tokens(size)
+        @tokens += size
+        @tokens = @max_tokens if @tokens > @max_tokens
+      end
+
+      def increment_total_count
+        @total_messages += 1
+      end
+
+      def increment_conforming_count
+        @conforming_messages += 1
+      end
+    end
+
+    # \RateLimiter that accepts all resources,
+    # with no limits.
+    class UnlimitedLimiter < RateLimiter
+      # @return [Boolean] always +true+
+      def allow?(_)
+        true
+      end
+
+      # @return [Float] always 100%
+      def effective_rate
+        1.0
+      end
+    end
+  end
+end

data/lib/ddtrace/sampling/rule.rb

@@ -0,0 +1,61 @@
+require 'forwardable'
+
+require 'ddtrace/sampling/matcher'
+require 'ddtrace/sampler'
+
+module Datadog
+  module Sampling
+    # Sampling rule that dictates if a span matches
+    # a specific criteria and what sampling strategy to
+    # apply in case of a positive match.
+    class Rule
+      extend Forwardable
+
+      attr_reader :matcher, :sampler
+
+      # @param [Matcher] matcher A matcher to verify span conformity against
+      # @param [Sampler] sampler A sampler to be consulted on a positive match
+      def initialize(matcher, sampler)
+        @matcher = matcher
+        @sampler = sampler
+      end
+
+      # Evaluates if the provided `span` conforms to the `matcher`.
+      #
+      # @param [Span] span
+      # @return [Boolean] whether this rules applies to the span
+      # @return [NilClass] if the matcher fails errs during evaluation
+      def match?(span)
+        @matcher.match?(span)
+      rescue => e
+        Datadog::Tracer.log.error("Matcher failed. Cause: #{e.message} Source: #{e.backtrace.first}")
+        nil
+      end
+
+      def_delegators :@sampler, :sample?, :sample_rate
+    end
+
+    # A \Rule that matches a span based on
+    # operation name and/or service name and
+    # applies a fixed sampling to matching spans.
+    class SimpleRule < Rule
+      # @param name [String,Regexp,Proc] Matcher for case equality (===) with the span name, defaults to always match
+      # @param service [String,Regexp,Proc] Matcher for case equality (===) with the service name, defaults to always match
+      # @param sample_rate [Float] Sampling rate between +[0,1]+
+      def initialize(name: SimpleMatcher::MATCH_ALL, service: SimpleMatcher::MATCH_ALL, sample_rate: 1.0)
+        # We want to allow 0.0 to drop all traces, but \RateSampler
+        # considers 0.0 an invalid rate and falls back to 100% sampling.
+        #
+        # We address that here by not setting the rate in the constructor,
+        # but using the setter method.
+        #
+        # We don't want to make this change directly to \RateSampler
+        # because it breaks its current contract to existing users.
+        sampler = Datadog::RateSampler.new
+        sampler.sample_rate = sample_rate
+
+        super(SimpleMatcher.new(name: name, service: service), sampler)
+      end
+    end
+  end
+end