ConcurrentHashMap通常在并发环境中用于聚合键下的某些事件 - 比如计算某些字符串值的命中。如果我们事先不知道密钥，我们需要有一个很好的方法来初始化密钥的需要，它应该是快速和安全的并发性。 这个问题的最佳模式（就效率而言）是什么？
我将使用<String, AtomicInteger>声明的模型地图：
ConcurrentHashMap<String, AtomicInteger> map = new ConcurrentHashMap<>();
但它可能是一个包含任何键值对的映射，如果键中的键尚未存在，则需要初始化键值对，并将值状态改为记录事件。
有两种流行的方法：
第一个使用ConcurrentHashMap.putIfAbsent：
AtomicInteger count = map.get(s);
if (count == null) {
count = new AtomicInteger(0);
AtomicInteger prevCount = map.putIfAbsent(s, count);
if (prevCount != null) {
count = prevCount;
}
}
count.incrementAndGet();
第二个使用ConcurrentHashMap.computeIfAbsent：
AtomicInteger count = map.computeIfAbsent(s, (k) -> new AtomicInteger(0));
count.incrementAndGet();
哪一个更适合这项任务？还有其他方法吗？
答案 0 :(得分：4)
不幸的是，直到jdk1.8.0_131，computeIfAbsent总是进入synchronized区块，如果密钥已经存在，则会显示，这使得它比putIfAbsent更慢
此基准确认了这一点，似乎取决于争用级别putIfAbsent比computeIfAbsent快2至50倍。
import java.util.Random;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.atomic.AtomicLong;
public class ConcurrentHashMapTest {
private final static int numberOfRuns = 1000000;
private final static int numberOfThreads = Runtime.getRuntime().availableProcessors();
private final static int keysSize = 10;
private final static String[] strings = new String[keysSize];
static {
for (int n = 0; n < keysSize; n++) {
strings[n] = "" + (char) ('A' + n);
}
}
public static void main(String[] args) throws InterruptedException {
for (int n = 0; n < 20; n++) {
testPutIfAbsent();
testComputeIfAbsentLambda();
}
}
private static void testPutIfAbsent() throws InterruptedException {
final AtomicLong totalTime = new AtomicLong();
final ConcurrentHashMap<String, AtomicInteger> map = new ConcurrentHashMap<String, AtomicInteger>();
final Random random = new Random();
ExecutorService executorService = Executors.newFixedThreadPool(numberOfThreads);
for (int i = 0; i < numberOfThreads; i++) {
executorService.execute(new Runnable() {
@Override
public void run() {
long start, end;
for (int n = 0; n < numberOfRuns; n++) {
String s = strings[random.nextInt(strings.length)];
start = System.nanoTime();
AtomicInteger count = map.get(s);
if (count == null) {
count = new AtomicInteger(0);
AtomicInteger prevCount = map.putIfAbsent(s, count);
if (prevCount != null) {
count = prevCount;
}
}
count.incrementAndGet();
end = System.nanoTime();
totalTime.addAndGet(end - start);
}
}
});
}
executorService.shutdown();
executorService.awaitTermination(Long.MAX_VALUE, TimeUnit.DAYS);
System.out.println("Test " + Thread.currentThread().getStackTrace()[1].getMethodName()
+ " average time per run: " + (double) totalTime.get() / numberOfThreads / numberOfRuns + " ns");
}
private static void testComputeIfAbsentLambda() throws InterruptedException {
final AtomicLong totalTime = new AtomicLong();
final ConcurrentHashMap<String, AtomicInteger> map = new ConcurrentHashMap<String, AtomicInteger>();
final Random random = new Random();
ExecutorService executorService = Executors.newFixedThreadPool(numberOfThreads);
for (int i = 0; i < numberOfThreads; i++) {
executorService.execute(new Runnable() {
@Override
public void run() {
long start, end;
for (int n = 0; n < numberOfRuns; n++) {
String s = strings[random.nextInt(strings.length)];
start = System.nanoTime();
AtomicInteger count = map.computeIfAbsent(s, (k) -> new AtomicInteger(0));
count.incrementAndGet();
end = System.nanoTime();
totalTime.addAndGet(end - start);
}
}
});
}
executorService.shutdown();
executorService.awaitTermination(Long.MAX_VALUE, TimeUnit.DAYS);
System.out.println("Test " + Thread.currentThread().getStackTrace()[1].getMethodName()
+ " average time per run: " + (double) totalTime.get() / numberOfThreads / numberOfRuns + " ns");
}
}
输出：
Test testPutIfAbsent average time per run: 115.756501 ns
Test testComputeIfAbsentLambda average time per run: 276.9667055 ns
Test testPutIfAbsent average time per run: 134.2332435 ns
Test testComputeIfAbsentLambda average time per run: 223.222063625 ns
Test testPutIfAbsent average time per run: 119.968893625 ns
Test testComputeIfAbsentLambda average time per run: 216.707419875 ns
Test testPutIfAbsent average time per run: 116.173902375 ns
Test testComputeIfAbsentLambda average time per run: 215.632467375 ns
Test testPutIfAbsent average time per run: 112.21422775 ns
Test testComputeIfAbsentLambda average time per run: 210.29563725 ns
Test testPutIfAbsent average time per run: 120.50643475 ns
Test testComputeIfAbsentLambda average time per run: 200.79536475 ns
我们可以使用putIfAbsent方法更快地创建computeIfAbsent。
唯一的区别是，在同时初始化相同密钥的情况下，这个新computeIfAbsent可以多次调用initialization function。基准测试结果与'putIfAbsent'相同，因为它是相同的代码，这不是一个大惊喜，但如果有人想测试这个，这里是基准：
import java.util.Random;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.atomic.AtomicLong;
import java.util.function.Function;
public class CocnurrentHashMap2Benchmark {
private final static int numberOfRuns = 1000000;
private final static int numberOfThreads = Runtime.getRuntime().availableProcessors();
private final static int keysSize = 10;
private final static String[] strings = new String[keysSize];
static {
for (int n = 0; n < keysSize; n++) {
strings[n] = "" + (char) ('A' + n);
}
}
public static void main(String[] args) throws InterruptedException {
for (int n = 0; n < 20; n++) {
testPutIfAbsent();
testComputeIfAbsent2Lambda();
}
}
private static void testPutIfAbsent() throws InterruptedException {
final AtomicLong totalTime = new AtomicLong();
final ConcurrentHashMap<String, AtomicInteger> map = new ConcurrentHashMap<String, AtomicInteger>();
final Random random = new Random();
ExecutorService executorService = Executors.newFixedThreadPool(numberOfThreads);
for (int i = 0; i < numberOfThreads; i++) {
executorService.execute(new Runnable() {
@Override
public void run() {
long start, end;
for (int n = 0; n < numberOfRuns; n++) {
String s = strings[random.nextInt(strings.length)];
start = System.nanoTime();
AtomicInteger count = map.get(s);
if (count == null) {
count = new AtomicInteger(0);
AtomicInteger prevCount = map.putIfAbsent(s, count);
if (prevCount != null) {
count = prevCount;
}
}
count.incrementAndGet();
end = System.nanoTime();
totalTime.addAndGet(end - start);
}
}
});
}
executorService.shutdown();
executorService.awaitTermination(Long.MAX_VALUE, TimeUnit.DAYS);
System.out.println("Test " + Thread.currentThread().getStackTrace()[1].getMethodName()
+ " average time per run: " + (double) totalTime.get() / numberOfThreads / numberOfRuns + " ns");
}
private static void testComputeIfAbsent2Lambda() throws InterruptedException {
final AtomicLong totalTime = new AtomicLong();
final ConcurrentHashMap2<String, AtomicInteger> map = new ConcurrentHashMap2<String, AtomicInteger>();
final Random random = new Random();
ExecutorService executorService = Executors.newFixedThreadPool(numberOfThreads);
for (int i = 0; i < numberOfThreads; i++) {
executorService.execute(new Runnable() {
@Override
public void run() {
long start, end;
for (int n = 0; n < numberOfRuns; n++) {
String s = strings[random.nextInt(strings.length)];
start = System.nanoTime();
AtomicInteger count = map.computeIfAbsent2(s, (k) -> new AtomicInteger(0));
count.incrementAndGet();
end = System.nanoTime();
totalTime.addAndGet(end - start);
}
}
});
}
executorService.shutdown();
executorService.awaitTermination(Long.MAX_VALUE, TimeUnit.DAYS);
System.out.println("Test " + Thread.currentThread().getStackTrace()[1].getMethodName()
+ " average time per run: " + (double) totalTime.get() / numberOfThreads / numberOfRuns + " ns");
}
public static class ConcurrentHashMap2<K,V> extends ConcurrentHashMap<K,V> {
/**
* If there is no mapping for the key then computes and puts the mapping,
* otherwise it simply return the value for that key.
* In case of concurrent initialization of the same key the mappingFunction can be called more than once.
* @param key - the key to be initialized or retrieved
* @param mappingFunction - the function to be called for computation of initial value.
* @return computed value if the key wasn't already in the map otherwise return the actual value for provided key.
*/
public V computeIfAbsent2(K key, Function<K,V> mappingFunction) {
V value = get(key);
if (value == null) {
value = mappingFunction.apply(key);
V prevValue = putIfAbsent(key, value);
if (prevValue != null) {
value = prevValue;
}
}
return value;
}
}
}
结果：
Test testComputeIfAbsent2Lambda average time per run: 138.1053415 ns
Test testPutIfAbsent average time per run: 129.45236425 ns
Test testComputeIfAbsent2Lambda average time per run: 128.48006825 ns
Test testPutIfAbsent average time per run: 118.733798375 ns
Test testComputeIfAbsent2Lambda average time per run: 134.038046625 ns
Test testPutIfAbsent average time per run: 119.7947695 ns
Test testComputeIfAbsent2Lambda average time per run: 134.183876375 ns
Test testPutIfAbsent average time per run: 137.969932625 ns
Test testComputeIfAbsent2Lambda average time per run: 137.97531275 ns
Test testPutIfAbsent average time per run: 136.904379125 ns
Test testComputeIfAbsent2Lambda average time per run: 148.899750125 ns
Test testPutIfAbsent average time per run: 129.788293125 ns
Test testComputeIfAbsent2Lambda average time per run: 141.50586625 ns
Test testPutIfAbsent average time per run: 129.081558875 ns
Test testComputeIfAbsent2Lambda average time per run: 122.36628625 ns
Test testPutIfAbsent average time per run: 127.1215535 ns
Test testComputeIfAbsent2Lambda average time per run: 108.129917625 ns
Test testPutIfAbsent average time per run: 133.630786875 ns
Test testComputeIfAbsent2Lambda average time per run: 134.978805625 ns
Test testPutIfAbsent average time per run: 132.7747585 ns
Test testComputeIfAbsent2Lambda average time per run: 132.4352885 ns
Test testPutIfAbsent average time per run: 133.753792875 ns
Test testComputeIfAbsent2Lambda average time per run: 134.09569175 ns
Test testPutIfAbsent average time per run: 145.610141125 ns
Test testComputeIfAbsent2Lambda average time per run: 139.437622125 ns
如果我们比较当键不存在时将映射放入映射的速度，则看起来新的'computeIfAbsent2'也快得多。基准：
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicLong;
import java.util.function.Function;
public class CocnurrentHashMap2PutBenchmark {
private final static int numberOfRuns = 1000000;
private final static int numberOfThreads = Runtime.getRuntime().availableProcessors();
public static void main(String[] args) throws InterruptedException {
for (int n = 0; n < 20; n++) {
testComputeIfAbsent2();
testComputeIfAbsent();
}
}
private static void testComputeIfAbsent2() throws InterruptedException {
final AtomicLong totalTime = new AtomicLong();
final ConcurrentHashMap2<Integer, String> map = new ConcurrentHashMap2<Integer, String>();
ExecutorService executorService = Executors.newFixedThreadPool(numberOfThreads);
for (int i = 0; i < numberOfThreads; i++) {
executorService.execute(new Runnable() {
@Override
public void run() {
long start, end;
for (int n = 0; n < numberOfRuns; n++) {
Integer key = Integer.valueOf(n);
start = System.nanoTime();
String value = map.computeIfAbsent2(key, (k) -> "value");
end = System.nanoTime();
totalTime.addAndGet(end - start);
}
}
});
}
executorService.shutdown();
executorService.awaitTermination(Long.MAX_VALUE, TimeUnit.DAYS);
System.out.println("Test " + Thread.currentThread().getStackTrace()[1].getMethodName()
+ " average time per run: " + (double) totalTime.get() / numberOfThreads / numberOfRuns + " ns");
}
private static void testComputeIfAbsent() throws InterruptedException {
final AtomicLong totalTime = new AtomicLong();
final ConcurrentHashMap<Integer, String> map = new ConcurrentHashMap<Integer, String>();
ExecutorService executorService = Executors.newFixedThreadPool(numberOfThreads);
for (int i = 0; i < numberOfThreads; i++) {
executorService.execute(new Runnable() {
@Override
public void run() {
long start, end;
for (int n = 0; n < numberOfRuns; n++) {
Integer key = Integer.valueOf(n);
start = System.nanoTime();
String value = map.computeIfAbsent(key, (k) -> "value");
end = System.nanoTime();
totalTime.addAndGet(end - start);
}
}
});
}
executorService.shutdown();
executorService.awaitTermination(Long.MAX_VALUE, TimeUnit.DAYS);
System.out.println("Test " + Thread.currentThread().getStackTrace()[1].getMethodName()
+ " average time per run: " + (double) totalTime.get() / numberOfThreads / numberOfRuns + " ns");
}
public static class ConcurrentHashMap2<K, V> extends ConcurrentHashMap<K, V> {
/**
* If there is no mapping for the key then computes and puts the
* mapping, otherwise it simply return the value for that key. In case
* of concurrent initialization of the same key the mappingFunction can
* be called more than once.
*
* @param key
* - the key to be initialized or retrieved
* @param mappingFunction
* - the function to be called for computation of initial
* value.
* @return computed value if the key wasn't already in the map otherwise
* return the actual value for provided key.
*/
public V computeIfAbsent2(K key, Function<K, V> mappingFunction) {
V value = get(key);
if (value == null) {
value = mappingFunction.apply(key);
V prevValue = putIfAbsent(key, value);
if (prevValue != null) {
value = prevValue;
}
}
return value;
}
}
}
结果：
Test testComputeIfAbsent2 average time per run: 445.077932375 ns
Test testComputeIfAbsent average time per run: 784.786391 ns
Test testComputeIfAbsent2 average time per run: 294.10136375 ns
Test testComputeIfAbsent average time per run: 314.8724765 ns
Test testComputeIfAbsent2 average time per run: 236.56533275 ns
Test testComputeIfAbsent average time per run: 350.863664625 ns
Test testComputeIfAbsent2 average time per run: 346.19498275 ns
Test testComputeIfAbsent average time per run: 641.995172625 ns
Test testComputeIfAbsent2 average time per run: 255.441646125 ns
Test testComputeIfAbsent average time per run: 326.399150125 ns
Test testComputeIfAbsent2 average time per run: 275.626666125 ns
Test testComputeIfAbsent average time per run: 201.207314125 ns
Test testComputeIfAbsent2 average time per run: 289.19059725 ns
Test testComputeIfAbsent average time per run: 318.448059 ns
Test testComputeIfAbsent2 average time per run: 225.19701825 ns
Test testComputeIfAbsent average time per run: 306.461814125 ns
Test testComputeIfAbsent2 average time per run: 213.460366 ns
Test testComputeIfAbsent average time per run: 334.325044625 ns
Test testComputeIfAbsent2 average time per run: 256.4048955 ns
Test testComputeIfAbsent average time per run: 256.366700625 ns
Test testComputeIfAbsent2 average time per run: 231.88875575 ns
Test testComputeIfAbsent average time per run: 246.076624 ns
Test testComputeIfAbsent2 average time per run: 222.4649485 ns
Test testComputeIfAbsent average time per run: 266.505719625 ns
Test testComputeIfAbsent2 average time per run: 228.708391375 ns
Test testComputeIfAbsent average time per run: 261.866442625 ns
Test testComputeIfAbsent2 average time per run: 198.614718875 ns
Test testComputeIfAbsent average time per run: 225.43031925 ns
Test testComputeIfAbsent2 average time per run: 300.478359 ns
Test testComputeIfAbsent average time per run: 306.03640225 ns
Test testComputeIfAbsent2 average time per run: 195.0444215 ns
Test testComputeIfAbsent average time per run: 271.461982625 ns
Test testComputeIfAbsent2 average time per run: 224.306529875 ns
Test testComputeIfAbsent average time per run: 334.52790425 ns
Test testComputeIfAbsent2 average time per run: 212.217131625 ns
Test testComputeIfAbsent average time per run: 184.541579125 ns
Test testComputeIfAbsent2 average time per run: 265.417909625 ns
Test testComputeIfAbsent average time per run: 213.9811425 ns
Test testComputeIfAbsent2 average time per run: 298.76602575 ns
Test testComputeIfAbsent average time per run: 347.883728125 ns