Skip to content
Open
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,7 @@
import org.apache.bookkeeper.mledger.Position;
import org.apache.pulsar.common.policies.data.ManagedLedgerInternalStats;

class MockManagedCursor implements ManagedCursor {
public class MockManagedCursor implements ManagedCursor {
ActiveManagedCursorContainer container;
Position markDeletePosition;
Position readPosition;
Expand All @@ -58,7 +58,8 @@ class MockManagedCursor implements ManagedCursor {
this.durable = durable;
}

static MockManagedCursor createCursor(ActiveManagedCursorContainer container, String name, Position position) {
public static MockManagedCursor createCursor(ActiveManagedCursorContainer container, String name,
Position position) {
return new MockManagedCursor(container, name, position, position, false, true);
}

Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,321 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/
package org.apache.pulsar.broker.delayed.bucket;

import io.netty.util.HashedWheelTimer;
import io.netty.util.Timer;
import io.netty.util.concurrent.DefaultThreadFactory;
import java.time.Clock;
import java.util.NavigableSet;
import java.util.concurrent.ThreadLocalRandom;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicLong;
import org.apache.bookkeeper.mledger.Position;
import org.apache.bookkeeper.mledger.PositionFactory;
import org.apache.bookkeeper.mledger.impl.ActiveManagedCursorContainerImpl;
import org.apache.bookkeeper.mledger.impl.MockManagedCursor;
import org.apache.pulsar.broker.delayed.DelayedDeliveryTracker;
import org.apache.pulsar.broker.delayed.NoopDelayedDeliveryContext;
import org.openjdk.jmh.annotations.Benchmark;
import org.openjdk.jmh.annotations.BenchmarkMode;
import org.openjdk.jmh.annotations.Fork;
import org.openjdk.jmh.annotations.Level;
import org.openjdk.jmh.annotations.Measurement;
import org.openjdk.jmh.annotations.Mode;
import org.openjdk.jmh.annotations.OutputTimeUnit;
import org.openjdk.jmh.annotations.Param;
import org.openjdk.jmh.annotations.Scope;
import org.openjdk.jmh.annotations.Setup;
import org.openjdk.jmh.annotations.State;
import org.openjdk.jmh.annotations.TearDown;
import org.openjdk.jmh.annotations.Threads;
import org.openjdk.jmh.annotations.Warmup;

/**
* JMH benchmarks for {@link BucketDelayedDeliveryTracker}.
*
* <p>This benchmark measures tracker throughput under different read/write ratios
* and initial message counts without implying a specific lock implementation.
*
* <p>Run with: mvn exec:java -Dexec.mainClass="org.openjdk.jmh.Main"
* -Dexec.args="BucketDelayedDeliveryTrackerBenchmark"
*/
@BenchmarkMode(Mode.Throughput)
@OutputTimeUnit(TimeUnit.SECONDS)
@State(Scope.Benchmark)
@Warmup(time = 10, timeUnit = TimeUnit.SECONDS, iterations = 1)
@Measurement(time = 10, timeUnit = TimeUnit.SECONDS, iterations = 1)
@Fork(1)
public class BucketDelayedDeliveryTrackerBenchmark {

/**
* Fixed delivery timestamp base that stays far beyond any benchmark trial duration,
* so scheduled tasks will not start firing while throughput is being measured.
*/
private static final long FUTURE_DELIVERY_BASE_TIME_MILLIS = 4102444800000L; // 2100-01-01T00:00:00Z

@Param({"90_10", "80_20", "70_30", "50_50"})
public String readWriteRatio;

@Param({"1000", "5000", "8000"})
public int initialMessages;

private BucketDelayedDeliveryTracker tracker;
private Timer timer;
private MockBucketSnapshotStorage storage;
private NoopDelayedDeliveryContext context;
private AtomicLong messageIdGenerator;
private int readPercentage;
private long futureDeliveryBaseTimeMillis;
/**
* Maximum number of additional unique (ledgerId, entryId) positions to
* introduce per trial on top of {@link #initialMessages}. This allows
* controlling the memory footprint of the benchmark while still applying
* sustained write pressure to the tracker.
*
* <p>Use {@code -p maxAdditionalUniqueMessages=...} on the JMH command line
* to tune the load. The default value is conservative for local runs.</p>
*/
@Param({"1000000"})
public long maxAdditionalUniqueMessages;
/**
* Upper bound on the absolute message id that will be used to derive
* (ledgerId, entryId) positions during a single trial.
*/
private long maxUniqueMessageId;
/**
* In real Pulsar usage, {@link DelayedDeliveryTracker#addMessage(long, long, long)} is invoked
* by a single dispatcher thread and messages arrive in order of (ledgerId, entryId).
* <p>
* To reflect this invariant in the benchmark, all write operations that end up calling
* {@code tracker.addMessage(...)} are serialized via this mutex so that the tracker only
* ever observes a single writer with monotonically increasing ids, even when JMH runs the
* benchmark method with multiple threads.
*/
private final Object writeMutex = new Object();

@Setup(Level.Trial)
public void setup() throws Exception {
setupMockComponents();
createTracker();
String[] parts = readWriteRatio.split("_");
readPercentage = Integer.parseInt(parts[0]);
futureDeliveryBaseTimeMillis = FUTURE_DELIVERY_BASE_TIME_MILLIS;
preloadMessages();
messageIdGenerator = new AtomicLong(initialMessages + 1);
// Allow a bounded number of additional unique messages per trial to avoid
// unbounded memory growth while still stressing the indexing logic.
maxUniqueMessageId = initialMessages + maxAdditionalUniqueMessages;
}

@TearDown(Level.Trial)
public void tearDown() throws Exception {
if (tracker != null) {
tracker.close();
}
if (timer != null) {
timer.stop();
}
}

private void setupMockComponents() throws Exception {
timer = new HashedWheelTimer(new DefaultThreadFactory("test-delayed-delivery"), 100, TimeUnit.MILLISECONDS);
storage = new MockBucketSnapshotStorage();

ActiveManagedCursorContainerImpl container = new ActiveManagedCursorContainerImpl();
MockManagedCursor cursor = MockManagedCursor.createCursor(container, "test-cursor",
PositionFactory.create(0, 0));
// Use the same "<topic> / <cursor>" naming pattern as real dispatchers,
// so that Bucket.asyncSaveBucketSnapshot can correctly derive topicName.
String dispatcherName = "persistent://public/default/jmh-topic / " + cursor.getName();
context = new NoopDelayedDeliveryContext(dispatcherName, cursor);
}

private void createTracker() throws Exception {
tracker = new BucketDelayedDeliveryTracker(
context, timer, 1000, Clock.systemUTC(), true, storage,
20, 1000, 100, 50
);
}

private void preloadMessages() {
// Preload messages to create realistic test conditions while keeping
// delivery timestamps far beyond the benchmark trial duration so the
// tracker's timer does not start firing during measurement.
long baseTime = futureDeliveryBaseTimeMillis;
for (int i = 1; i <= initialMessages; i++) {
tracker.addMessage(i, i, baseTime + i * 1000L);
}
}

// =============================================================================
// READ-WRITE RATIO BENCHMARKS
// =============================================================================

@Benchmark
public boolean benchmarkMixedOperations() {
if (ThreadLocalRandom.current().nextInt(100) < readPercentage) {
Comment on lines +171 to +173
Copy link

Copilot AI Apr 3, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

benchmarkMixedOperations() parses readWriteRatio (split + Integer.parseInt) on every benchmark invocation. That parsing overhead will be included in the measured throughput and can dominate at high rates. Consider parsing the ratio once in @Setup (e.g., store an int readPercentage field) and using that field in the benchmark methods.

Copilot uses AI. Check for mistakes.
// Read operations
return performReadOperation();
} else {
// Write operations
return performWriteOperation();
}
}

/**
* Serialize calls to {@link BucketDelayedDeliveryTracker#addMessage(long, long, long)} and
* ensure (ledgerId, entryId) are generated in a strictly increasing sequence, matching the
* real dispatcher single-threaded behaviour.
*/
private boolean addMessageSequential(long deliverAt, int entryIdModulo) {
synchronized (writeMutex) {
long id = messageIdGenerator.getAndIncrement();
// Limit the number of distinct positions that are introduced into the tracker
// to keep memory usage bounded. Once the upper bound is reached, we re-use
// the last position id so that subsequent calls behave like updates to
// existing messages and are short-circuited by containsMessage checks.
long boundedId = Math.min(id, maxUniqueMessageId);
long ledgerId = boundedId;
long entryId = boundedId % entryIdModulo;
return tracker.addMessage(ledgerId, entryId, deliverAt);
}
}

private boolean performReadOperation() {
int operation = ThreadLocalRandom.current().nextInt(3);
switch (operation) {
case 0:
// containsMessage
long ledgerId = ThreadLocalRandom.current().nextLong(1, initialMessages + 100);
long entryId = ThreadLocalRandom.current().nextLong(1, 1000);
return tracker.containsMessage(ledgerId, entryId);
case 1:
// nextDeliveryTime
try {
tracker.nextDeliveryTime();
return true;
} catch (Exception e) {
return false;
}
case 2:
// getNumberOfDelayedMessages
long count = tracker.getNumberOfDelayedMessages();
return count >= 0;
default:
return false;
}
}

private boolean performWriteOperation() {
long deliverAt = futureDeliveryBaseTimeMillis + ThreadLocalRandom.current().nextLong(5000, 30000);
return addMessageSequential(deliverAt, 1000);
}
Comment on lines +226 to +229
Copy link

Copilot AI Apr 3, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

performWriteOperation() calls System.currentTimeMillis() on the hot path. Besides adding unrelated overhead, it also makes the scheduled timeout firing time depend on wall-clock timing/jitter. Consider deriving deliverAt from a precomputed base timestamp (set up in @Setup) to reduce noise and improve benchmark repeatability.

Copilot uses AI. Check for mistakes.

// =============================================================================
// SPECIFIC OPERATION BENCHMARKS
// =============================================================================

@Benchmark
@Threads(8)
public boolean benchmarkConcurrentContainsMessage() {
long ledgerId = ThreadLocalRandom.current().nextLong(1, initialMessages + 100);
long entryId = ThreadLocalRandom.current().nextLong(1, 1000);
return tracker.containsMessage(ledgerId, entryId);
}

@Benchmark
@Threads(4)
public boolean benchmarkConcurrentAddMessage() {
long deliverAt = futureDeliveryBaseTimeMillis + ThreadLocalRandom.current().nextLong(10000, 60000);
return addMessageSequential(deliverAt, 1000);
}

@Benchmark
@Threads(2)
public NavigableSet<Position> benchmarkConcurrentGetScheduledMessages() {
// Create some messages ready for delivery
long currentTime = System.currentTimeMillis();
for (int i = 0; i < 5; i++) {
addMessageSequential(currentTime - 1000, 100);
}
return tracker.getScheduledMessages(10);
}

@Benchmark
@Threads(16)
public long benchmarkConcurrentNextDeliveryTime() {
try {
return tracker.nextDeliveryTime();
} catch (Exception e) {
return -1;
}
}

@Benchmark
@Threads(1)
public long benchmarkGetNumberOfDelayedMessages() {
return tracker.getNumberOfDelayedMessages();
}

// =============================================================================
// HIGH CONTENTION SCENARIOS
// =============================================================================

@Benchmark
@Threads(32)
public boolean benchmarkHighContentionMixedOperations() {
return benchmarkMixedOperations();
}

@Benchmark
@Threads(16)
public boolean benchmarkContentionReads() {
return performReadOperation();
}

@Benchmark
@Threads(8)
public boolean benchmarkContentionWrites() {
return performWriteOperation();
}

// =============================================================================
// THROUGHPUT BENCHMARKS
// =============================================================================

@Benchmark
@Threads(1)
public boolean benchmarkSingleThreadedThroughput() {
return benchmarkMixedOperations();
}

@Benchmark
@Threads(4)
public boolean benchmarkMediumConcurrencyThroughput() {
return benchmarkMixedOperations();
}

@Benchmark
@Threads(8)
public boolean benchmarkHighConcurrencyThroughput() {
return benchmarkMixedOperations();
}

}
Loading
Loading