ARTEMIS-3016 Reduce DuplicateIDCache memory footprint

This commit is contained in:
franz1981 2020-11-30 16:12:54 +01:00 committed by Clebert Suconic
parent 985559d086
commit b3b5d4893c
5 changed files with 385 additions and 66 deletions

View File

@ -0,0 +1,99 @@
/*
* 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.activemq.artemis.api.core;
import java.io.Serializable;
import java.util.Objects;
/**
* A Pair is a holder for 1 object and a >= 0 primitive long value.
* <p>
* This is a utility class.
*/
public class ObjLongPair<A> implements Serializable {
private static final long serialVersionUID = 7749478219139339853L;
public static final long NIL = -1;
public ObjLongPair(final A a, final long b) {
this.a = a;
this.b = b;
if (b < 0 && b != NIL) {
throw new IllegalStateException("b must be >= 0 or == NIL = " + NIL);
}
}
public ObjLongPair(final A a) {
this.a = a;
this.b = NIL;
}
private A a;
private long b;
@Override
public int hashCode() {
if (a == null && b == NIL) {
return super.hashCode();
}
// it's ok to use b to compute hashCode although NIL
return (a == null ? 0 : a.hashCode()) + 37 * Long.hashCode(b);
}
@Override
public boolean equals(final Object other) {
if (other == this) {
return true;
}
if (other instanceof ObjLongPair == false) {
return false;
}
ObjLongPair<A> pother = (ObjLongPair<A>) other;
return (Objects.equals(pother.a, a)) && (pother.b == b);
}
@Override
public String toString() {
return "ObjLongPair[a=" + a + ", b=" + (b == NIL ? "NIL" : b) + "]";
}
public void setA(A a) {
if (this.a == a) {
return;
}
this.a = a;
}
public A getA() {
return a;
}
public void setB(long b) {
if (b < 0 && b != NIL) {
throw new IllegalStateException("b must be >= 0 or == NIL = " + NIL);
}
this.b = b;
}
public long getB() {
return b;
}
}

View File

@ -275,6 +275,52 @@ public class ByteUtil {
}
}
public static int hashCode(byte[] bytes) {
if (PlatformDependent.hasUnsafe() && PlatformDependent.isUnaligned()) {
return unsafeHashCode(bytes);
}
return Arrays.hashCode(bytes);
}
/**
* This hash code computation is borrowed by {@link io.netty.buffer.ByteBufUtil#hashCode(ByteBuf)}.
*/
private static int unsafeHashCode(byte[] bytes) {
if (bytes == null) {
return 0;
}
final int len = bytes.length;
int hashCode = 1;
final int intCount = len >>> 2;
int arrayIndex = 0;
// reading in batch both help hash code computation data dependencies and save memory bandwidth
for (int i = 0; i < intCount; i++) {
hashCode = 31 * hashCode + PlatformDependent.getInt(bytes, arrayIndex);
arrayIndex += Integer.BYTES;
}
final byte remaining = (byte) (len & 3);
if (remaining > 0) {
hashCode = unsafeUnrolledHashCode(bytes, arrayIndex, remaining, hashCode);
}
return hashCode == 0 ? 1 : hashCode;
}
private static int unsafeUnrolledHashCode(byte[] bytes, int index, int bytesCount, int h) {
// there is still the hash data dependency but is more friendly
// then a plain loop, given that we know no loop is needed here
assert bytesCount > 0 && bytesCount < 4;
h = 31 * h + PlatformDependent.getByte(bytes, index);
if (bytesCount == 1) {
return h;
}
h = 31 * h + PlatformDependent.getByte(bytes, index + 1);
if (bytesCount == 2) {
return h;
}
h = 31 * h + PlatformDependent.getByte(bytes, index + 2);
return h;
}
public static boolean equals(final byte[] left, final byte[] right) {
return equals(left, right, 0, right.length);
}

View File

@ -17,17 +17,23 @@
package org.apache.activemq.artemis.utils;
import java.nio.ByteBuffer;
import java.nio.ByteOrder;
import java.nio.ReadOnlyBufferException;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.LinkedHashMap;
import java.util.Map;
import io.netty.util.internal.PlatformDependent;
import org.jboss.logging.Logger;
import org.junit.Assert;
import org.junit.Assume;
import org.junit.Test;
import static org.junit.Assert.assertArrayEquals;
import static org.junit.Assert.assertEquals;
import static org.junit.Assert.assertTrue;
import static org.junit.Assert.fail;
import org.junit.Test;
public class ByteUtilTest {
@ -83,6 +89,46 @@ public class ByteUtilTest {
}
}
@Test
public void testUnsafeUnalignedByteArrayHashCode() {
Assume.assumeTrue(PlatformDependent.hasUnsafe());
Assume.assumeTrue(PlatformDependent.isUnaligned());
Map<byte[], Integer> map = new LinkedHashMap<>();
map.put(new byte[0], 1);
map.put(new byte[]{1}, 32);
map.put(new byte[]{2}, 33);
map.put(new byte[]{0, 1}, 962);
map.put(new byte[]{1, 2}, 994);
if (ByteOrder.nativeOrder() == ByteOrder.BIG_ENDIAN) {
map.put(new byte[]{0, 1, 2, 3, 4, 5}, 63504931);
map.put(new byte[]{6, 7, 8, 9, 0, 1}, -1603953111);
map.put(new byte[]{-1, -1, -1, (byte) 0xE1}, 1);
} else {
map.put(new byte[]{0, 1, 2, 3, 4, 5}, 1250309600);
map.put(new byte[]{6, 7, 8, 9, 0, 1}, -417148442);
map.put(new byte[]{-1, -1, -1, (byte) 0xE1}, -503316450);
}
for (Map.Entry<byte[], Integer> e : map.entrySet()) {
assertEquals("input = " + Arrays.toString(e.getKey()), e.getValue().intValue(), ByteUtil.hashCode(e.getKey()));
}
}
@Test
public void testNoUnsafeAlignedByteArrayHashCode() {
Assume.assumeFalse(PlatformDependent.hasUnsafe());
Assume.assumeFalse(PlatformDependent.isUnaligned());
ArrayList<byte[]> inputs = new ArrayList<>();
inputs.add(new byte[0]);
inputs.add(new byte[]{1});
inputs.add(new byte[]{2});
inputs.add(new byte[]{0, 1});
inputs.add(new byte[]{1, 2});
inputs.add(new byte[]{0, 1, 2, 3, 4, 5});
inputs.add(new byte[]{6, 7, 8, 9, 0, 1});
inputs.add(new byte[]{-1, -1, -1, (byte) 0xE1});
inputs.forEach(input -> assertEquals("input = " + Arrays.toString(input), Arrays.hashCode(input), ByteUtil.hashCode(input)));
}
@Test
public void testTextBytesToLongBytesNegative() {
try {

View File

@ -16,12 +16,15 @@
*/
package org.apache.activemq.artemis.core.postoffice.impl;
import java.lang.ref.Reference;
import java.lang.ref.WeakReference;
import java.util.ArrayList;
import java.util.List;
import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
import org.apache.activemq.artemis.api.core.ActiveMQDuplicateIdException;
import org.apache.activemq.artemis.api.core.ObjLongPair;
import org.apache.activemq.artemis.api.core.Pair;
import org.apache.activemq.artemis.api.core.SimpleString;
import org.apache.activemq.artemis.core.persistence.StorageManager;
@ -33,6 +36,8 @@ import org.apache.activemq.artemis.core.transaction.TransactionOperationAbstract
import org.apache.activemq.artemis.utils.ByteUtil;
import org.jboss.logging.Logger;
import static org.apache.activemq.artemis.api.core.ObjLongPair.NIL;
/**
* A DuplicateIDCacheImpl
*
@ -41,6 +46,23 @@ import org.jboss.logging.Logger;
public class DuplicateIDCacheImpl implements DuplicateIDCache {
private static final Logger logger = Logger.getLogger(DuplicateIDCacheImpl.class);
// we're not interested into safe publication here: we need to scale, be fast and save "some" GC to happen
private static WeakReference<Integer[]> INDEXES = null;
private static Integer[] boxedInts(int size) {
final Reference<Integer[]> indexesRef = INDEXES;
final Integer[] indexes = indexesRef == null ? null : indexesRef.get();
if (indexes != null && size <= indexes.length) {
return indexes;
}
final int newSize = size + (indexes == null ? 0 : size / 2);
final Integer[] newIndexes = new Integer[newSize];
if (indexes != null) {
System.arraycopy(indexes, 0, newIndexes, 0, indexes.length);
}
INDEXES = new WeakReference<>(newIndexes);
return newIndexes;
}
// ByteHolder, position
private final Map<ByteArrayHolder, Integer> cache = new ConcurrentHashMap<>();
@ -49,7 +71,9 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
// Note - deliberately typed as ArrayList since we want to ensure fast indexed
// based array access
private final ArrayList<Pair<ByteArrayHolder, Long>> ids;
private final ArrayList<ObjLongPair<ByteArrayHolder>> ids;
private final Integer[] cachedBoxedInts;
private int pos;
@ -69,11 +93,24 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
ids = new ArrayList<>(size);
cachedBoxedInts = boxedInts(size);
this.storageManager = storageManager;
this.persist = persist;
}
// best effort caching mechanism
private Integer boxed(int index) {
Integer boxedInt = this.cachedBoxedInts[index];
if (boxedInt == null) {
boxedInt = index;
cachedBoxedInts[index] = boxedInt;
}
assert boxedInt != null;
return boxedInt;
}
@Override
public void load(final List<Pair<byte[], Long>> theIds) throws Exception {
long txID = -1;
@ -98,9 +135,9 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
} else {
ByteArrayHolder bah = new ByteArrayHolder(id.getA());
Pair<ByteArrayHolder, Long> pair = new Pair<>(bah, id.getB());
ObjLongPair<ByteArrayHolder> pair = new ObjLongPair<>(bah, id.getB());
cache.put(bah, ids.size());
cache.put(bah, boxed(ids.size()));
ids.add(pair);
if (logger.isTraceEnabled()) {
@ -133,7 +170,7 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
Integer posUsed = cache.remove(bah);
if (posUsed != null) {
Pair<ByteArrayHolder, Long> id;
ObjLongPair<ByteArrayHolder> id;
synchronized (this) {
id = ids.get(posUsed.intValue());
@ -144,7 +181,7 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
if (logger.isTraceEnabled()) {
logger.trace("DuplicateIDCacheImpl(" + this.address + ")::deleteFromCache deleting id=" + describeID(duplicateID, id.getB()));
}
id.setB(null);
id.setB(NIL);
}
}
}
@ -161,10 +198,16 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
@Override
public boolean contains(final byte[] duplID) {
boolean contains = cache.get(new ByteArrayHolder(duplID)) != null;
return contains(new ByteArrayHolder(duplID));
}
if (contains) {
logger.trace("DuplicateIDCacheImpl(" + this.address + ")::constains found a duplicate " + describeID(duplID, 0));
private boolean contains(final ByteArrayHolder duplID) {
boolean contains = cache.containsKey(duplID);
if (logger.isTraceEnabled()) {
if (contains) {
logger.trace("DuplicateIDCacheImpl(" + this.address + ")::contains found a duplicate " + describeID(duplID.bytes, 0));
}
}
return contains;
}
@ -181,68 +224,68 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
@Override
public synchronized boolean atomicVerify(final byte[] duplID, final Transaction tx) throws Exception {
if (contains(duplID)) {
final ByteArrayHolder holder = new ByteArrayHolder(duplID);
if (contains(holder)) {
if (tx != null) {
tx.markAsRollbackOnly(new ActiveMQDuplicateIdException());
}
return false;
} else {
addToCache(duplID, tx, true);
return true;
}
addToCache(holder, tx, true);
return true;
}
@Override
public synchronized void addToCache(final byte[] duplID, final Transaction tx, boolean instantAdd) throws Exception {
long recordID = -1;
addToCache(new ByteArrayHolder(duplID), tx, instantAdd);
}
private synchronized void addToCache(final ByteArrayHolder holder,
final Transaction tx,
boolean instantAdd) throws Exception {
long recordID = -1;
if (tx == null) {
if (persist) {
recordID = storageManager.generateID();
storageManager.storeDuplicateID(address, duplID, recordID);
storageManager.storeDuplicateID(address, holder.bytes, recordID);
}
addToCacheInMemory(duplID, recordID);
addToCacheInMemory(holder, recordID);
} else {
if (persist) {
recordID = storageManager.generateID();
storageManager.storeDuplicateIDTransactional(tx.getID(), address, duplID, recordID);
storageManager.storeDuplicateIDTransactional(tx.getID(), address, holder.bytes, recordID);
tx.setContainsPersistent();
}
if (logger.isTraceEnabled()) {
logger.trace("DuplicateIDCacheImpl(" + this.address + ")::addToCache Adding duplicateID TX operation for " + describeID(duplID, recordID) + ", tx=" + tx);
logger.trace("DuplicateIDCacheImpl(" + this.address + ")::addToCache Adding duplicateID TX operation for " + describeID(holder.bytes, recordID) + ", tx=" + tx);
}
if (instantAdd) {
tx.addOperation(new AddDuplicateIDOperation(duplID, recordID, false));
tx.addOperation(new AddDuplicateIDOperation(holder, recordID, false));
} else {
// For a tx, it's important that the entry is not added to the cache until commit
// since if the client fails then resends them tx we don't want it to get rejected
tx.afterStore(new AddDuplicateIDOperation(duplID, recordID, true));
tx.afterStore(new AddDuplicateIDOperation(holder, recordID, true));
}
}
}
@Override
public void load(final Transaction tx, final byte[] duplID) {
tx.addOperation(new AddDuplicateIDOperation(duplID, tx.getID(), true));
tx.addOperation(new AddDuplicateIDOperation(new ByteArrayHolder(duplID), tx.getID(), true));
}
private synchronized void addToCacheInMemory(final byte[] duplID, final long recordID) {
private synchronized void addToCacheInMemory(final ByteArrayHolder holder, final long recordID) {
if (logger.isTraceEnabled()) {
logger.trace("DuplicateIDCacheImpl(" + this.address + ")::addToCacheInMemory Adding " + describeID(duplID, recordID));
logger.trace("DuplicateIDCacheImpl(" + this.address + ")::addToCacheInMemory Adding " + describeID(holder.bytes, recordID));
}
ByteArrayHolder holder = new ByteArrayHolder(duplID);
cache.put(holder, boxed(pos));
cache.put(holder, pos);
Pair<ByteArrayHolder, Long> id;
ObjLongPair<ByteArrayHolder> id;
if (pos < ids.size()) {
// Need fast array style access here -hence ArrayList typing
@ -260,7 +303,7 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
// Note we can't use update since journal update doesn't let older records get
// reclaimed
if (id.getB() != null) {
if (id.getB() != NIL) {
try {
storageManager.deleteDuplicateID(id.getB());
} catch (Exception e) {
@ -273,15 +316,14 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
// The recordID could be negative if the duplicateCache is configured to not persist,
// -1 would mean null on this case
id.setB(recordID >= 0 ? recordID : null);
id.setB(recordID >= 0 ? recordID : NIL);
if (logger.isTraceEnabled()) {
logger.trace("DuplicateIDCacheImpl(" + this.address + ")::addToCacheInMemory replacing old duplicateID by " + describeID(id.getA().bytes, id.getB()));
}
holder.pos = pos;
} else {
id = new Pair<>(holder, recordID >= 0 ? recordID : null);
id = new ObjLongPair<>(holder, recordID >= 0 ? recordID : NIL);
if (logger.isTraceEnabled()) {
logger.trace("DuplicateIDCacheImpl(" + this.address + ")::addToCacheInMemory Adding new duplicateID " + describeID(id.getA().bytes, id.getB()));
@ -289,7 +331,6 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
ids.add(id);
holder.pos = pos;
}
if (pos++ == cacheSize - 1) {
@ -301,10 +342,12 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
public void clear() throws Exception {
logger.debug("DuplicateIDCacheImpl(" + this.address + ")::clear removing duplicate ID data");
synchronized (this) {
if (ids.size() > 0 && persist) {
final int idsSize = ids.size();
if (idsSize > 0 && persist) {
long tx = storageManager.generateID();
for (Pair<ByteArrayHolder, Long> id : ids) {
if (id != null) {
for (int i = 0; i < idsSize; i++) {
final ObjLongPair<ByteArrayHolder> id = ids.get(i);
if (id != null && id.getB() != NIL) {
storageManager.deleteDuplicateIDTransactional(tx, id.getB());
}
}
@ -319,16 +362,18 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
@Override
public List<Pair<byte[], Long>> getMap() {
List<Pair<byte[], Long>> list = new ArrayList<>();
for (Pair<ByteArrayHolder, Long> id : ids) {
list.add(new Pair<>(id.getA().bytes, id.getB()));
final int idsSize = ids.size();
List<Pair<byte[], Long>> copy = new ArrayList<>(idsSize);
for (int i = 0; i < idsSize; i++) {
final ObjLongPair<ByteArrayHolder> id = ids.get(i);
copy.add(new Pair<>(id.getA().bytes, id.getB() == NIL ? null : id.getB()));
}
return list;
return copy;
}
private final class AddDuplicateIDOperation extends TransactionOperationAbstract {
final byte[] duplID;
final ByteArrayHolder holder;
final long recordID;
@ -336,15 +381,15 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
private final boolean afterCommit;
AddDuplicateIDOperation(final byte[] duplID, final long recordID, boolean afterCommit) {
this.duplID = duplID;
AddDuplicateIDOperation(final ByteArrayHolder holder, final long recordID, boolean afterCommit) {
this.holder = holder;
this.recordID = recordID;
this.afterCommit = afterCommit;
}
private void process() {
if (!done) {
addToCacheInMemory(duplID, recordID);
addToCacheInMemory(holder, recordID);
done = true;
}
@ -372,32 +417,20 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
private static final class ByteArrayHolder {
private final byte[] bytes;
private int hash;
ByteArrayHolder(final byte[] bytes) {
this.bytes = bytes;
}
final byte[] bytes;
int hash;
int pos;
@Override
public boolean equals(final Object other) {
if (other instanceof ByteArrayHolder) {
ByteArrayHolder s = (ByteArrayHolder) other;
if (bytes.length != s.bytes.length) {
return false;
}
for (int i = 0; i < bytes.length; i++) {
if (bytes[i] != s.bytes[i]) {
return false;
}
}
return true;
return ByteUtil.equals(bytes, s.bytes);
} else {
return false;
}
@ -406,9 +439,7 @@ public class DuplicateIDCacheImpl implements DuplicateIDCache {
@Override
public int hashCode() {
if (hash == 0) {
for (byte b : bytes) {
hash = 31 * hash + b;
}
hash = ByteUtil.hashCode(bytes);
}
return hash;

View File

@ -0,0 +1,97 @@
/*
* 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.activemq.artemis.tests.performance.jmh;
import java.util.Arrays;
import java.util.SplittableRandom;
import java.util.concurrent.TimeUnit;
import org.apache.activemq.artemis.utils.ByteUtil;
import org.openjdk.jmh.annotations.Benchmark;
import org.openjdk.jmh.annotations.BenchmarkMode;
import org.openjdk.jmh.annotations.Fork;
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.Warmup;
@State(Scope.Benchmark)
@BenchmarkMode(Mode.AverageTime)
@OutputTimeUnit(TimeUnit.NANOSECONDS)
@Fork(2)
@Warmup(iterations = 5, time = 200, timeUnit = TimeUnit.MILLISECONDS)
@Measurement(iterations = 8, time = 200, timeUnit = TimeUnit.MILLISECONDS)
public class ByteArrayHashCodeBenchamark {
@Param({"7", "8", "36"})
private int length;
@Param("0")
private int seed;
/**
* The higher the less predictable for the CPU branch predictor.
*/
@Param({"4", "10"})
private int logPermutations;
@Param({"true"})
private boolean unsafe;
private long seq;
private int inputMask;
private byte[][] inputs;
@Setup
public void init() {
System.setProperty("io.netty.noUnsafe", Boolean.valueOf(!unsafe).toString());
int inputSize = 1 << logPermutations;
inputs = new byte[inputSize][];
inputMask = inputs.length - 1;
// splittable random can create repeatable sequence of inputs
SplittableRandom random = new SplittableRandom(seed);
for (int i = 0; i < inputs.length; i++) {
final byte[] bytes = new byte[length];
for (int b = 0; b < length; b++) {
bytes[b] = (byte) random.nextInt(Byte.MIN_VALUE, Byte.MAX_VALUE + 1);
}
inputs[i] = bytes;
}
}
private byte[] nextInput() {
final long seq = this.seq;
final int index = (int) (seq & inputMask);
this.seq = seq + 1;
return inputs[index];
}
@Benchmark
public int artemisHashCode() {
return ByteUtil.hashCode(nextInput());
}
@Benchmark
public int arraysHashCode() {
return Arrays.hashCode(nextInput());
}
}