Only report when hits are nonzero

amarziali · amarziali · commit f1bc43ad38cc · 2026-04-29T15:14:59.000+02:00
diff --git a/dd-trace-core/src/main/java/datadog/trace/common/metrics/Aggregator.java b/dd-trace-core/src/main/java/datadog/trace/common/metrics/Aggregator.java
@@ -200,7 +200,9 @@ private CommonKeyCleaner(Set<MetricKey> commonKeys, HealthMetrics healthMetrics)
     @Override
     public void accept(Map.Entry<MetricKey, AggregateMetric> expired) {
       commonKeys.remove(expired.getKey());
-      healthMetrics.onStatsAggregateDropped();
+      if (expired.getValue().getHitCount() > 0) {
+        healthMetrics.onStatsAggregateDropped();
+      }
     }
   }
 }
diff --git a/dd-trace-core/src/test/groovy/datadog/trace/common/metrics/ConflatingMetricAggregatorTest.groovy b/dd-trace-core/src/test/groovy/datadog/trace/common/metrics/ConflatingMetricAggregatorTest.groovy
@@ -978,6 +978,52 @@ class ConflatingMetricAggregatorTest extends DDSpecification {
     aggregator.close()
   }
 
+  def "should not report dropped aggregate when evicted entry was already flushed"() {
+    setup:
+    int maxAggregates = 5
+    MetricWriter writer = Mock(MetricWriter)
+    Sink sink = Stub(Sink)
+    DDAgentFeaturesDiscovery features = Mock(DDAgentFeaturesDiscovery)
+    features.supportsMetrics() >> true
+    features.peerTags() >> []
+    HealthMetrics healthMetrics = Mock(HealthMetrics)
+    ConflatingMetricsAggregator aggregator = new ConflatingMetricsAggregator(empty,
+      features, healthMetrics, sink, writer, maxAggregates, queueSize, reportingInterval, SECONDS, false)
+    aggregator.start()
+
+    when: "fill cache and flush — entries are cleared (hitCount=0) but stay in the LRU"
+    CountDownLatch latch1 = new CountDownLatch(1)
+    for (int i = 0; i < maxAggregates; ++i) {
+      aggregator.publish([
+        new SimpleSpan("service" + i, "operation", "resource", "type", false, true, false, 0, 100, HTTP_OK)
+        .setTag(SPAN_KIND, "baz")
+      ])
+    }
+    aggregator.report()
+    latch1.await(2, SECONDS)
+
+    then:
+    1 * writer.finishBucket() >> { latch1.countDown() }
+
+    when: "publish new distinct spans — LRU evicts the cleared entries before the next report"
+    CountDownLatch latch2 = new CountDownLatch(1)
+    for (int i = maxAggregates; i < maxAggregates * 2; ++i) {
+      aggregator.publish([
+        new SimpleSpan("service" + i, "operation", "resource", "type", false, true, false, 0, 100, HTTP_OK)
+        .setTag(SPAN_KIND, "baz")
+      ])
+    }
+    aggregator.report()
+    latch2.await(2, SECONDS)
+
+    then: "no drop metric because all evicted entries had hitCount=0 (already reported)"
+    1 * writer.finishBucket() >> { latch2.countDown() }
+    0 * healthMetrics.onStatsAggregateDropped()
+
+    cleanup:
+    aggregator.close()
+  }
+
   def "aggregate not updated in reporting interval not reported"() {
     setup:
     int maxAggregates = 10

Original file line number	Diff line number	Diff line change
`@@ -200,7 +200,9 @@ private CommonKeyCleaner(Set<MetricKey> commonKeys, HealthMetrics healthMetrics)`
`200`	`200`	`@Override`
`201`	`201`	`public void accept(Map.Entry<MetricKey, AggregateMetric> expired) {`
`202`	`202`	`commonKeys.remove(expired.getKey());`
`203`		`- healthMetrics.onStatsAggregateDropped();`
	`203`	`+ if (expired.getValue().getHitCount() > 0) {`
	`204`	`+ healthMetrics.onStatsAggregateDropped();`
	`205`	`+ }`
`204`	`206`	`}`
`205`	`207`	`}`
`206`	`208`	`}`