revert ddog sampling

dpep · Mar 13, 2024 · ee5acab · ee5acab
1 parent 1351f3c
commit ee5acab
Show file tree

Hide file tree

Showing 4 changed files with 19 additions and 41 deletions.
diff --git a/lib/network_resiliency.rb b/lib/network_resiliency.rb
@@ -22,11 +22,6 @@ module Adapter
   DEFAULT_TIMEOUT_MIN = 10 # ms
   MODE = [ :observe, :resilient ].freeze
   RESILIENCY_THRESHOLD = 100
-  SAMPLE_RATE = {
-    timeout: 0.1,
-    stats: 0.1,
-    sync: 0.1,
-  }
 
   extend self
 
@@ -254,7 +249,6 @@ def record(adapter:, action:, destination:, duration:, error:, timeout:, attempt
         adapter: adapter,
         destination: destination,
       },
-      sample_rate: SAMPLE_RATE[:timeout],
     ) if timeout && timeout > 0
 
     if error
@@ -286,28 +280,23 @@ def record(adapter:, action:, destination:, duration:, error:, timeout:, attempt
       # ensure Syncer is running
       Syncer.start
 
-      if rand < SAMPLE_RATE[:stats]
-        NetworkResiliency.statsd&.distribution(
-          "network_resiliency.#{action}.stats.n",
-          stats.n,
-          tags: tags,
-          sample_rate: SAMPLE_RATE[:stats],
-        )
+      NetworkResiliency.statsd&.distribution(
+        "network_resiliency.#{action}.stats.n",
+        stats.n,
+        tags: tags,
+      )
 
-        NetworkResiliency.statsd&.distribution(
-          "network_resiliency.#{action}.stats.avg",
-          stats.avg,
-          tags: tags,
-          sample_rate: SAMPLE_RATE[:stats],
-        )
+      NetworkResiliency.statsd&.distribution(
+        "network_resiliency.#{action}.stats.avg",
+        stats.avg,
+        tags: tags,
+      )
 
-        NetworkResiliency.statsd&.distribution(
-          "network_resiliency.#{action}.stats.stdev",
-          stats.stdev,
-          tags: tags,
-          sample_rate: SAMPLE_RATE[:stats],
-        )
-      end
+      NetworkResiliency.statsd&.distribution(
+        "network_resiliency.#{action}.stats.stdev",
+        stats.stdev,
+        tags: tags,
+      )
     end
 
     nil
@@ -372,8 +361,7 @@ def timeouts_for(adapter:, action:, destination:, max: nil, units: :ms)
           NetworkResiliency.statsd&.increment(
             "network_resiliency.timeout.raised",
             tags: tags,
-            sample_rate: SAMPLE_RATE[:timeout],
-          ) if rand < SAMPLE_RATE[:timeout]
+          )
         end
       else
         # the specified timeout is less than our expected p99...awkward
@@ -382,8 +370,7 @@ def timeouts_for(adapter:, action:, destination:, max: nil, units: :ms)
         NetworkResiliency.statsd&.increment(
           "network_resiliency.timeout.too_low",
           tags: tags,
-          sample_rate: SAMPLE_RATE[:timeout],
-        ) if rand < SAMPLE_RATE[:timeout]
+        )
       end
     else
       timeouts << p99
@@ -394,8 +381,7 @@ def timeouts_for(adapter:, action:, destination:, max: nil, units: :ms)
       NetworkResiliency.statsd&.increment(
         "network_resiliency.timeout.missing",
         tags: tags,
-        sample_rate: SAMPLE_RATE[:timeout],
-      ) if rand < SAMPLE_RATE[:timeout]
+      )
     end
 
     NetworkResiliency.statsd&.distribution(
@@ -405,8 +391,7 @@ def timeouts_for(adapter:, action:, destination:, max: nil, units: :ms)
         adapter: adapter,
         destination: destination,
       },
-      sample_rate: SAMPLE_RATE[:timeout],
-    ) if rand < SAMPLE_RATE[:timeout]
+    )
 
     case units
     when nil, :ms, :milliseconds

diff --git a/lib/network_resiliency/stats_engine.rb b/lib/network_resiliency/stats_engine.rb
@@ -65,13 +65,11 @@ def sync(redis)
           empty: data.empty?,
           truncated: data.size < dirty_keys.size,
         }.select { |_, v| v },
-        sample_rate: SAMPLE_RATE[:sync],
       )
 
       NetworkResiliency.statsd&.distribution(
         "network_resiliency.sync.keys.dirty",
         dirty_keys.select { |_, n| n > 0 }.count,
-        sample_rate: SAMPLE_RATE[:sync],
       )
 
       return [] if data.empty?

diff --git a/spec/spec_helper.rb b/spec/spec_helper.rb
@@ -49,9 +49,6 @@
     # disable background sync
     allow(NetworkResiliency::Syncer).to receive(:start)
 
-    # disable metric sampling for specs
-    stub_const("NetworkResiliency::SAMPLE_RATE", Hash.new(1))
-
     # since Timecop doesn't work with Process.clock_gettime
     allow(Process).to receive(:clock_gettime).and_return(*(1..1_000))
 

diff --git a/spec/stats_engine_spec.rb b/spec/stats_engine_spec.rb
@@ -132,7 +132,6 @@
           is_expected.to have_received(:distribution).with(
             "network_resiliency.sync.keys.dirty",
             1,
-            anything,
           )
         end
       end
@@ -150,7 +149,6 @@
           is_expected.to have_received(:distribution).with(
             "network_resiliency.sync.keys.dirty",
             1,
-            anything,
           )
         end