ls-trace 0.1.2 → 0.2.0

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.logger.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

data/lib/ddtrace/sampling/rule_sampler.rb

@@ -0,0 +1,125 @@
+require 'forwardable'
+
+require 'ddtrace/ext/priority'
+
+require 'ddtrace/ext/sampling'
+require 'ddtrace/sampler'
+require 'ddtrace/sampling/rate_limiter'
+
+module Datadog
+  module Sampling
+    # Span {Sampler} that applies a set of {Rule}s to decide
+    # on sampling outcome. Then, a rate limiter is applied.
+    #
+    # If a span does not conform to any rules, a default
+    # sampling strategy is applied.
+    class RuleSampler
+      extend Forwardable
+
+      AGENT_RATE_METRIC_KEY = '_dd.agent_psr'.freeze
+
+      attr_reader :rules, :rate_limiter, :default_sampler
+
+      # @param rules [Array<Rule>] ordered list of rules to be applied to a span
+      # @param rate_limit [Float] number of traces per second, defaults to +100+
+      # @param rate_limiter [RateLimiter] limiter applied after rule matching
+      # @param default_sample_rate [Float] fallback sample rate when no rules apply to a span,
+      #   between +[0,1]+, defaults to +1+
+      # @param default_sampler [Sample] fallback strategy when no rules apply to a span
+      def initialize(rules = [],
+                     rate_limit: Datadog.configuration.sampling.rate_limit,
+                     rate_limiter: nil,
+                     default_sample_rate: Datadog.configuration.sampling.default_rate,
+                     default_sampler: nil)
+
+        @rules = rules
+
+        @rate_limiter = if rate_limiter
+                          rate_limiter
+                        elsif rate_limit
+                          Datadog::Sampling::TokenBucket.new(rate_limit)
+                        else
+                          Datadog::Sampling::UnlimitedLimiter.new
+                        end
+
+        @default_sampler = if default_sampler
+                             default_sampler
+                           elsif default_sample_rate
+                             # 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.
+                             Datadog::RateSampler.new.tap { |s| s.sample_rate = default_sample_rate }
+                           else
+                             RateByServiceSampler.new(1.0, env: -> { Datadog.tracer.tags[:env] })
+                           end
+      end
+
+      # /RuleSampler's components (it's rate limiter, for example) are
+      # not be guaranteed to be size-effect free.
+      # It is not possible to guarantee that a call to {#sample?} will
+      # return the same result as a successive call to {#sample!} with the same span.
+      #
+      # Use {#sample!} instead
+      def sample?(_span)
+        raise 'RuleSampler cannot be evaluated without side-effects'
+      end
+
+      def sample!(span)
+        sampled = sample_span(span) do |s|
+          @default_sampler.sample!(s).tap do
+            # We want to make sure the span is tagged with the agent-derived
+            # service rate. Retrieve this from the rate by service sampler.
+            # Only do this if it was set by a RateByServiceSampler.
+            if @default_sampler.is_a?(RateByServiceSampler)
+              s.set_metric(AGENT_RATE_METRIC_KEY, @default_sampler.sample_rate(span))
+            end
+          end
+        end
+
+        sampled.tap do
+          span.sampled = sampled
+        end
+      end
+
+      def update(*args)
+        return false unless @default_sampler.respond_to?(:update)
+        @default_sampler.update(*args)
+      end
+
+      private
+
+      def sample_span(span)
+        rule = @rules.find { |r| r.match?(span) }
+
+        return yield(span) if rule.nil?
+
+        sampled = rule.sample?(span)
+        sample_rate = rule.sample_rate(span)
+
+        set_rule_metrics(span, sample_rate)
+
+        return false unless sampled
+
+        rate_limiter.allow?(1).tap do
+          set_limiter_metrics(span, rate_limiter.effective_rate)
+        end
+      rescue StandardError => e
+        Datadog.logger.error("Rule sampling failed. Cause: #{e.message} Source: #{e.backtrace.first}")
+        yield(span)
+      end
+
+      def set_rule_metrics(span, sample_rate)
+        span.set_metric(Ext::Sampling::RULE_SAMPLE_RATE, sample_rate)
+      end
+
+      def set_limiter_metrics(span, limiter_rate)
+        span.set_metric(Ext::Sampling::RATE_LIMITER_RATE, limiter_rate)
+      end
+    end
+  end
+end

data/lib/ddtrace/span.rb

@@ -4,8 +4,10 @@ require 'thread'
 require 'ddtrace/utils'
 require 'ddtrace/ext/errors'
 require 'ddtrace/ext/priority'
+require 'ddtrace/environment'
 require 'ddtrace/analytics'
 require 'ddtrace/forced_tracing'
+require 'ddtrace/diagnostics/health'
 
 module Datadog
   # Represents a logical unit of work in the system. Each trace consists of one or more spans.
@@ -30,6 +32,9 @@ module Datadog
     # parsing 64-bit integers for distributed tracing since an upstream system may generate one
     EXTERNAL_MAX_ID = 2**64
 
+    # This limit is for numeric tags because uint64 could end up rounded.
+    NUMERIC_TAG_SIZE_RANGE = (-2**53..2**53)
+
     attr_accessor :name, :service, :resource, :span_type,
                   :start_time, :end_time,
                   :span_id, :trace_id, :parent_id,
@@ -79,29 +84,70 @@ module Datadog
     #
     #   span.set_tag('http.method', request.method)
     def set_tag(key, value = nil)
-      @meta[key] = value.to_s
+      # Keys must be unique between tags and metrics
+      @metrics.delete(key)
+
+      # Ensure `http.status_code` is always a string so it is added to
+      #   @meta instead of @metrics
+      # DEV: This is necessary because the agent looks to `meta['http.status_code']` for
+      #   tagging necessary metrics
+      value = value.to_s if key == Ext::HTTP::STATUS_CODE
+
+      # NOTE: Adding numeric tags as metrics is stop-gap support
+      #       for numeric typed tags. Eventually they will become
+      #       tags again.
+      # Any numeric that is not an integer greater than max size is logged as a metric.
+      # Everything else gets logged as a tag.
+      if value.is_a?(Numeric) && !(value.is_a?(Integer) && !NUMERIC_TAG_SIZE_RANGE.cover?(value))
+        set_metric(key, value)
+      else
+        @meta[key] = value.to_s
+      end
     rescue StandardError => e
-      Datadog::Tracer.log.debug("Unable to set the tag #{key}, ignoring it. Caused by: #{e}")
+      Datadog.logger.debug("Unable to set the tag #{key}, ignoring it. Caused by: #{e}")
+    end
+
+    # Sets tags from given hash, for each key in hash it sets the tag with that key
+    # and associated value from the hash. It is shortcut for `set_tag`. Keys and values
+    # of the hash must be strings. Note that nested hashes are not supported.
+    # A valid example is:
+    #
+    #   span.set_tags({ "http.method" => "GET", "user.id" => "234" })
+    def set_tags(tags)
+      tags.each { |k, v| set_tag(k, v) }
+    end
+
+    # This method removes a tag for the given key.
+    def clear_tag(key)
+      @meta.delete(key)
     end
 
     # Return the tag with the given key, nil if it doesn't exist.
     def get_tag(key)
-      @meta[key]
+      @meta[key] || @metrics[key]
     end
 
     # This method sets a tag with a floating point value for the given key. It acts
     # like `set_tag()` and it simply add a tag without further processing.
     def set_metric(key, value)
+      # Keys must be unique between tags and metrics
+      @meta.delete(key)
+
       # enforce that the value is a floating point number
       value = Float(value)
       @metrics[key] = value
     rescue StandardError => e
-      Datadog::Tracer.log.debug("Unable to set the metric #{key}, ignoring it. Caused by: #{e}")
+      Datadog.logger.debug("Unable to set the metric #{key}, ignoring it. Caused by: #{e}")
+    end
+
+    # This method removes a metric for the given key. It acts like {#remove_tag}.
+    def clear_metric(key)
+      @metrics.delete(key)
     end
 
     # Return the metric with the given key, nil if it doesn't exist.
     def get_metric(key)
-      @metrics[key]
+      @metrics[key] || @meta[key]
     end
 
     # Mark the span with the given error.
@@ -137,13 +183,14 @@ module Datadog
       # spans without a service would be dropped, so here we provide a default.
       # This should really never happen with integrations in contrib, as a default
       # service is always set. It's only for custom instrumentation.
-      @service ||= @tracer.default_service unless @tracer.nil?
+      @service ||= (@tracer && @tracer.default_service)
 
       begin
         @context.close_span(self)
         @tracer.record(self)
       rescue StandardError => e
-        Datadog::Tracer.log.debug("error recording finished trace: #{e}")
+        Datadog.logger.debug("error recording finished trace: #{e}")
+        Datadog.health_metrics.error_span_finish(1, tags: ["error:#{e.class.name}"])
       end
       self
     end

data/lib/ddtrace/sync_writer.rb

@@ -4,9 +4,10 @@ require 'ddtrace/runtime/metrics'
 
 module Datadog
   # SyncWriter flushes both services and traces synchronously
+  # DEV: To be replaced by Datadog::Workers::TraceWriter.
   class SyncWriter
     attr_reader \
-      :runtime_metrics,
+      :priority_sampler,
       :transport
 
     def initialize(options = {})
@@ -15,16 +16,13 @@ module Datadog
         Transport::HTTP.default(transport_options)
       end
 
-      # Runtime metrics
-      @runtime_metrics = options.fetch(:runtime_metrics) do
-        Runtime::Metrics.new
-      end
+      @priority_sampler = options.fetch(:priority_sampler, nil)
     end
 
     def write(trace, services = nil)
       unless services.nil?
         Datadog::Patcher.do_once('SyncWriter#write') do
-          Datadog::Tracer.log.warn(%(
+          Datadog.logger.warn(%(
             write: Writing services has been deprecated and no longer need to be provided.
             write(traces, services) can be updted to write(traces)
           ))
@@ -35,7 +33,13 @@ module Datadog
         proc { flush_trace(trace) }
       )
     rescue => e
-      Tracer.log.debug(e)
+      Datadog.logger.debug(e)
+    end
+
+    # Added for interface completeness
+    def stop
+      # No cleanup to do for the SyncWriter
+      true
     end
 
     private
@@ -46,8 +50,9 @@ module Datadog
 
     def flush_trace(trace)
       processed_traces = Pipeline.process!([trace])
+      return if processed_traces.empty?
       inject_hostname!(processed_traces.first) if Datadog.configuration.report_hostname
-      transport.send(:traces, processed_traces)
+      transport.send_traces(processed_traces)
     end
 
     def inject_hostname!(trace)