From 660e1483fcc4c6fb7b38f9adba8fd672d7e67112 Mon Sep 17 00:00:00 2001
From: David Schlosnagle <davids@palantir.com>
Date: Sat, 7 Sep 2024 08:41:04 -0400
Subject: [PATCH] memoize gauge snapshots

---
 .../AutobatcherTelemetryComponents.java       | 29 ++++++++++++++-----
 .../autobatch/DisruptorAutobatcherTest.java   | 21 +++++++++-----
 2 files changed, 35 insertions(+), 15 deletions(-)

diff --git a/atlasdb-autobatch/src/main/java/com/palantir/atlasdb/autobatch/AutobatcherTelemetryComponents.java b/atlasdb-autobatch/src/main/java/com/palantir/atlasdb/autobatch/AutobatcherTelemetryComponents.java
index f29101de1e6..d77b537b62a 100644
--- a/atlasdb-autobatch/src/main/java/com/palantir/atlasdb/autobatch/AutobatcherTelemetryComponents.java
+++ b/atlasdb-autobatch/src/main/java/com/palantir/atlasdb/autobatch/AutobatcherTelemetryComponents.java
@@ -18,8 +18,11 @@
 
 import com.codahale.metrics.Gauge;
 import com.codahale.metrics.Histogram;
+import com.codahale.metrics.Snapshot;
+import com.google.common.base.Suppliers;
 import com.palantir.tritium.metrics.registry.TaggedMetricRegistry;
 import java.time.Duration;
+import java.util.function.Supplier;
 
 public final class AutobatcherTelemetryComponents {
     private final String safeLoggablePurpose;
@@ -83,24 +86,34 @@ public static AutobatcherTelemetryComponents create(
     }
 
     private void registerGauges() {
+        // capturing a snapshot is expensive, so memoize for 10 milliseconds to reduce overhead during metrics read 50%.
+        Duration memoizeDuration = Duration.ofMillis(10);
+        Supplier<Snapshot> waitTimerSnapshot = Suppliers.memoizeWithExpiration(waitTimer::getSnapshot, memoizeDuration);
+        Supplier<Snapshot> waitTimeHistogramSnapshot =
+                Suppliers.memoizeWithExpiration(waitTimeHistogram::getSnapshot, memoizeDuration);
+        Supplier<Snapshot> runningTimerSnapshot =
+                Suppliers.memoizeWithExpiration(runningTimer::getSnapshot, memoizeDuration);
+        Supplier<Snapshot> totalTimerSnapshot =
+                Suppliers.memoizeWithExpiration(totalTimer::getSnapshot, memoizeDuration);
+
         overheadMetrics.waitTimeNanosP1(
-                (Gauge<Double>) () -> waitTimer.getSnapshot().getValue(0.01));
+                (Gauge<Double>) () -> waitTimerSnapshot.get().getValue(0.01));
         overheadMetrics.waitTimeNanosMedian(
-                (Gauge<Double>) () -> waitTimer.getSnapshot().getValue(0.5));
+                (Gauge<Double>) () -> waitTimerSnapshot.get().getValue(0.5));
 
         overheadMetrics.waitTimePercentageP1(
-                (Gauge<Double>) () -> waitTimeHistogram.getSnapshot().getValue(0.01));
+                (Gauge<Double>) () -> waitTimeHistogramSnapshot.get().getValue(0.01));
         overheadMetrics.waitTimePercentageMedian(
-                (Gauge<Double>) () -> waitTimeHistogram.getSnapshot().getValue(0.5));
+                (Gauge<Double>) () -> waitTimeHistogramSnapshot.get().getValue(0.5));
 
         overheadMetrics.runningTimeNanosP1(
-                (Gauge<Double>) () -> runningTimer.getSnapshot().getValue(0.01));
+                (Gauge<Double>) () -> runningTimerSnapshot.get().getValue(0.01));
         overheadMetrics.runningTimeNanosMedian(
-                (Gauge<Double>) () -> runningTimer.getSnapshot().getValue(0.5));
+                (Gauge<Double>) () -> runningTimerSnapshot.get().getValue(0.5));
 
         overheadMetrics.totalTimeNanosP1(
-                (Gauge<Double>) () -> totalTimer.getSnapshot().getValue(0.01));
+                (Gauge<Double>) () -> totalTimerSnapshot.get().getValue(0.01));
         overheadMetrics.totalTimeNanosMedian(
-                (Gauge<Double>) () -> totalTimer.getSnapshot().getValue(0.5));
+                (Gauge<Double>) () -> totalTimerSnapshot.get().getValue(0.5));
     }
 }
diff --git a/atlasdb-autobatch/src/test/java/com/palantir/atlasdb/autobatch/DisruptorAutobatcherTest.java b/atlasdb-autobatch/src/test/java/com/palantir/atlasdb/autobatch/DisruptorAutobatcherTest.java
index 8f3f79cc1f7..a6b16cb082f 100644
--- a/atlasdb-autobatch/src/test/java/com/palantir/atlasdb/autobatch/DisruptorAutobatcherTest.java
+++ b/atlasdb-autobatch/src/test/java/com/palantir/atlasdb/autobatch/DisruptorAutobatcherTest.java
@@ -76,17 +76,24 @@ private void assertWaitTimeAndRunningTimeAndTotalTimeMetricsAreProduced(
     }
 
     private void assertNoWaitTimeAndRunningTimeMetricsAreProduced(TaggedMetricRegistry registry) {
-        assertThat(getWaitTimeHistogram(registry)).isNull();
-        assertThat(getWaitTimePercentageHistogram(registry)).isNull();
-        assertThat(getRunningTimeHistogram(registry)).isNull();
-        assertThat(getTotalTimeHistogram(registry)).isNull();
+        assertThat(getWaitTimeHistogram(registry))
+                .satisfies(histogram -> assertThat(histogram.getCount()).isZero());
+        assertThat(getWaitTimePercentageHistogram(registry))
+                .satisfies(histogram -> assertThat(histogram.getCount()).isZero());
+        assertThat(getRunningTimeHistogram(registry))
+                .satisfies(histogram -> assertThat(histogram.getCount()).isZero());
+        assertThat(getTotalTimeHistogram(registry))
+                .satisfies(histogram -> assertThat(histogram.getCount()).isZero());
     }
 
     private void assertOnlyWaitTimeMetricsAreProduced(TaggedMetricRegistry registry, int waitTimeNanos) {
         assertThat(getWaitTimeHistogram(registry).getSnapshot().getValues()).containsExactly(waitTimeNanos);
-        assertThat(getWaitTimePercentageHistogram(registry)).isNull();
-        assertThat(getRunningTimeHistogram(registry)).isNull();
-        assertThat(getTotalTimeHistogram(registry)).isNull();
+        assertThat(getWaitTimePercentageHistogram(registry))
+                .satisfies(histogram -> assertThat(histogram.getCount()).isZero());
+        assertThat(getRunningTimeHistogram(registry))
+                .satisfies(histogram -> assertThat(histogram.getCount()).isZero());
+        assertThat(getTotalTimeHistogram(registry))
+                .satisfies(histogram -> assertThat(histogram.getCount()).isZero());
     }
 
     private static Histogram getWaitTimeHistogram(TaggedMetricRegistry registry) {