statsd-instrument 3.9.0 → 3.9.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 9776ac7b9f3dc12364a1ba286d7b77330c52b0aa8560c29c0ff47c8033c066df
-  data.tar.gz: ecfc23f01f345eabc82d1c696c2fb280cc3185f2d30b47e8e5c75326077f10bd
+  metadata.gz: 436100176f41a4557dfad3169208a34581a4a51ed03532857ce0192ef0586f95
+  data.tar.gz: 888b9d6dcab8741aede0db4044f9a4d5ce0eb146ce016037b5ee278296f1f2a1
 SHA512:
-  metadata.gz: e831388ae5824bd7d1e53b9eefc399dd639210648976adeafd48fabd9d4a37705af2d5814868e75922cc27840cba42d5066054751a8120e4ea8d055efdaea379
-  data.tar.gz: 91f63290ef5d9dcc032e0fcdad8e34acdc5448d563e1eedd71b23a68330060ba58af3999ace24b4a82567f2e5c96182996c3f8dda940684a1df93fe9acee02ef
+  metadata.gz: 181b505253e000ae9f4b457716e4715227dbdc81d85357bedc335296af681d2975e4da0062abd4028b87d37462cd3ae5280f1033886a1b96a9bf566c32cf7930
+  data.tar.gz: cbc3e3cb9ed41abfda94075eeb4e25e29445f42447af826616cc3545419be2e4ceeab273183bbb56af3e166066ef5bb02ed19ac09b4467782c0184e5251e454f

data/.ruby-version

@@ -1 +1 @@
-3.3.1
+3.3.3

data/CHANGELOG.md

@@ -6,6 +6,12 @@ section below.
 
 ## Unreleased changes
 
+## Version 3.9.1
+
+- [#378](https://github.com/Shopify/statsd-instrument/pull/378) - Respect sampling rate when aggregation is enabled, just for timing metrics.
+  Not respecting sampling rate, incurs in a performance penalty, as we will send more metrics than expected.
+  Moreover, it overloads the StatsD server, which has to send out and process more metrics than expected.
+
 ## Version 3.9.0
 
 - Introduced an experimental aggregation feature to improve the efficiency of metrics reporting by aggregating 
@@ -15,7 +21,7 @@ decrease the overhead associated with high-frequency metric reporting. To enable
 - Added support for sending StatsD via Unix domain sockets. This feature is enabled by
 setting the `STATSD_SOCKET` environment variable to the path of the Unix domain socket.
   - :warning: **Possible breaking change**: We removed/renamed some classes and now Sinks are generic, so the classes `UDPSink` and `UDPBatchedSink` are now called
-`StatsD::Instrument::Sink` and `StatsD::Instrument::Sink` respectively.
+`StatsD::Instrument::Sink` and `StatsD::Instrument::BatchedSink` respectively.
 If you used those internal classes, you will need to update your code to use the new classes.
 
 ## Version 3.8.0

data/lib/statsd/instrument/aggregator.rb

@@ -24,12 +24,15 @@ module StatsD
     end
 
     class Aggregator
+      DEFAULT_MAX_CONTEXT_SIZE = 250
+
       CONST_SAMPLE_RATE = 1.0
       COUNT = :c
       DISTRIBUTION = :d
       MEASURE = :ms
       HISTOGRAM = :h
       GAUGE = :g
+      private_constant :COUNT, :DISTRIBUTION, :MEASURE, :HISTOGRAM, :GAUGE, :CONST_SAMPLE_RATE
 
       class << self
         def finalize(aggregation_state, sink, datagram_builders, datagram_builder_class, default_tags)
@@ -78,7 +81,14 @@ module StatsD
       # @param default_tags [Array<String>] The tags to add to all metrics.
       # @param flush_interval [Float] The interval at which to flush the aggregated metrics.
       # @param max_values [Integer] The maximum number of values to aggregate before flushing.
-      def initialize(sink, datagram_builder_class, prefix, default_tags, flush_interval: 5.0, max_values: 100)
+      def initialize(
+        sink,
+        datagram_builder_class,
+        prefix,
+        default_tags,
+        flush_interval: 5.0,
+        max_values: DEFAULT_MAX_CONTEXT_SIZE
+      )
         @sink = sink
         @datagram_builder_class = datagram_builder_class
         @metric_prefix = prefix

data/lib/statsd/instrument/client.rb

@@ -156,7 +156,8 @@ module StatsD
         sink: StatsD::Instrument::NullSink.new,
         datagram_builder_class: self.class.datagram_builder_class_for_implementation(implementation),
         enable_aggregation: false,
-        aggregation_flush_interval: 2.0
+        aggregation_flush_interval: 2.0,
+        aggregation_max_context_size: StatsD::Instrument::Aggregator::DEFAULT_MAX_CONTEXT_SIZE
       )
         @sink = sink
         @datagram_builder_class = datagram_builder_class
@@ -176,6 +177,7 @@ module StatsD
               prefix,
               default_tags,
               flush_interval: @aggregation_flush_interval,
+              max_values: aggregation_max_context_size,
             )
         end
       end
@@ -237,6 +239,19 @@ module StatsD
       # @param tags (see #increment)
       # @return [void]
       def measure(name, value = nil, sample_rate: nil, tags: nil, no_prefix: false, &block)
+        sample_rate ||= @default_sample_rate
+        if sample_rate && !sample?(sample_rate)
+          # For all timing metrics, we have to use the sampling logic.
+          # Not doing so would impact performance and CPU usage.
+          # See Datadog's documentation for more details: https://github.com/DataDog/datadog-go/blob/20af2dbfabbbe6bd0347780cd57ed931f903f223/statsd/aggregator.go#L281-L283
+
+          if block_given?
+            return yield
+          end
+
+          return StatsD::Instrument::VOID
+        end
+
         if block_given?
           return latency(name, sample_rate: sample_rate, tags: tags, metric_type: :ms, no_prefix: no_prefix, &block)
         end
@@ -245,10 +260,7 @@ module StatsD
           @aggregator.aggregate_timing(name, value, tags: tags, no_prefix: no_prefix, type: :ms)
           return StatsD::Instrument::VOID
         end
-        sample_rate ||= @default_sample_rate
-        if sample_rate.nil? || sample?(sample_rate)
-          emit(datagram_builder(no_prefix: no_prefix).ms(name, value, sample_rate, tags))
-        end
+        emit(datagram_builder(no_prefix: no_prefix).ms(name, value, sample_rate, tags))
         StatsD::Instrument::VOID
       end
 
@@ -306,6 +318,19 @@ module StatsD
       # @param tags (see #increment)
       # @return [void]
       def distribution(name, value = nil, sample_rate: nil, tags: nil, no_prefix: false, &block)
+        sample_rate ||= @default_sample_rate
+        if sample_rate && !sample?(sample_rate)
+          # For all timing metrics, we have to use the sampling logic.
+          # Not doing so would impact performance and CPU usage.
+          # See Datadog's documentation for more details: https://github.com/DataDog/datadog-go/blob/20af2dbfabbbe6bd0347780cd57ed931f903f223/statsd/aggregator.go#L281-L283
+
+          if block_given?
+            return yield
+          end
+
+          return StatsD::Instrument::VOID
+        end
+
         if block_given?
           return latency(name, sample_rate: sample_rate, tags: tags, metric_type: :d, no_prefix: no_prefix, &block)
         end
@@ -315,10 +340,7 @@ module StatsD
           return StatsD::Instrument::VOID
         end
 
-        sample_rate ||= @default_sample_rate
-        if sample_rate.nil? || sample?(sample_rate)
-          emit(datagram_builder(no_prefix: no_prefix).d(name, value, sample_rate, tags))
-        end
+        emit(datagram_builder(no_prefix: no_prefix).d(name, value, sample_rate, tags))
         StatsD::Instrument::VOID
       end
 
@@ -334,14 +356,20 @@ module StatsD
       # @param tags (see #increment)
       # @return [void]
       def histogram(name, value, sample_rate: nil, tags: nil, no_prefix: false)
+        sample_rate ||= @default_sample_rate
+        if sample_rate && !sample?(sample_rate)
+          # For all timing metrics, we have to use the sampling logic.
+          # Not doing so would impact performance and CPU usage.
+          # See Datadog's documentation for more details: https://github.com/DataDog/datadog-go/blob/20af2dbfabbbe6bd0347780cd57ed931f903f223/statsd/aggregator.go#L281-L283
+          return StatsD::Instrument::VOID
+        end
+
         if @enable_aggregation
           @aggregator.aggregate_timing(name, value, tags: tags, no_prefix: no_prefix, type: :h)
+          return StatsD::Instrument::VOID
         end
 
-        sample_rate ||= @default_sample_rate
-        if sample_rate.nil? || sample?(sample_rate)
-          emit(datagram_builder(no_prefix: no_prefix).h(name, value, sample_rate, tags))
-        end
+        emit(datagram_builder(no_prefix: no_prefix).h(name, value, sample_rate, tags))
         StatsD::Instrument::VOID
       end
 

data/lib/statsd/instrument/environment.rb

@@ -125,6 +125,13 @@ module StatsD
         Float(env.fetch("STATSD_AGGREGATION_INTERVAL", 2.0))
       end
 
+      def aggregation_max_context_size
+        Integer(env.fetch(
+          "STATSD_AGGREGATION_MAX_CONTEXT_SIZE",
+          StatsD::Instrument::Aggregator::DEFAULT_MAX_CONTEXT_SIZE,
+        ))
+      end
+
       def client
         StatsD::Instrument::Client.from_env(self)
       end

data/lib/statsd/instrument/version.rb

@@ -2,6 +2,6 @@
 
 module StatsD
   module Instrument
-    VERSION = "3.9.0"
+    VERSION = "3.9.1"
   end
 end

data/test/client_test.rb

@@ -245,6 +245,18 @@ class ClientTest < Minitest::Test
     5.times { client.increment("metric") }
   end
 
+  def test_sampling_with_aggregation
+    mock_sink = mock("sink")
+    mock_sink.stubs(:sample?).returns(false, true, false, false, true)
+    # Since we are aggregating, we only expect a single datagram.
+    mock_sink.expects(:<<).with("metric:60:60|d").once
+    mock_sink.expects(:flush).once
+
+    client = StatsD::Instrument::Client.new(sink: mock_sink, default_sample_rate: 0.5, enable_aggregation: true)
+    5.times { client.distribution("metric", 60) }
+    client.force_flush
+  end
+
   def test_clone_with_prefix_option
     # Both clients will use the same sink.
     mock_sink = mock("sink")

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: statsd-instrument
 version: !ruby/object:Gem::Version
-  version: 3.9.0
+  version: 3.9.1
 platform: ruby
 authors:
 - Jesse Storimer
@@ -10,7 +10,7 @@ authors:
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2024-08-19 00:00:00.000000000 Z
+date: 2024-09-16 00:00:00.000000000 Z
 dependencies: []
 description: A StatsD client for Ruby apps. Provides metaprogramming methods to inject
   StatsD instrumentation into your code.
@@ -130,7 +130,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.5.17
+rubygems_version: 3.5.18
 signing_key:
 specification_version: 4
 summary: A StatsD client for Ruby apps