mirror of
https://gitee.com/dromara/hutool.git
synced 2025-06-28 13:34:09 +08:00
!1343 可召回批处理线程池执行器，主线程、线程池混合执行批处理任务，主线程空闲时会尝试召回线程池队列中的任务执行
Merge pull request !1343 from lk/v5-dev
This commit is contained in:
Looly
Gitee
commit
0c19f0b9a4
@ -0,0 +1,303 @@
|
||||
package cn.hutool.core.thread;
|
||||
|
||||
import java.util.*;
|
||||
import java.util.concurrent.*;
|
||||
import java.util.concurrent.atomic.AtomicBoolean;
|
||||
import java.util.concurrent.atomic.AtomicInteger;
|
||||
import java.util.function.Function;
|
||||
import java.util.function.Supplier;
|
||||
import java.util.stream.Collectors;
|
||||
import java.util.stream.Stream;
|
||||
|
||||
/**
|
||||
* 可召回批处理线程池执行器
|
||||
* <pre>
|
||||
* 1.数据分批并行处理
|
||||
* 2.主线程、线程池混合执行批处理任务，主线程空闲时会尝试召回线程池队列中的任务执行
|
||||
* 3.线程安全，可用同时执行多个任务，线程池满载时，效率与单线程模式相当，无阻塞风险，无脑提交任务即可
|
||||
* </pre>
|
||||
*
|
||||
* 适用场景：
|
||||
* <pre>
|
||||
* 1.批量处理数据且需要同步结束的场景，能一定程度上提高吞吐量、防止任务堆积 {@link #process(List, int, Function)}
|
||||
* 2.普通查询接口加速 {@link #processByWarp(Warp[])}
|
||||
* </pre>
|
||||
*
|
||||
* @author likuan
|
||||
*/
|
||||
public class RecyclableBatchThreadPoolExecutor {
|
||||
|
||||
private final ExecutorService executor;
|
||||
|
||||
public RecyclableBatchThreadPoolExecutor(int poolSize){
|
||||
this(poolSize,"recyclable-batch-pool-");
|
||||
}
|
||||
|
||||
/**
|
||||
* 建议的构造方法
|
||||
* <pre>
|
||||
* 1.使用无界队列，主线程会召回队列中的任务执行，不会有任务堆积，无需考虑拒绝策略
|
||||
* 2.假如在web场景中请求量过大导致oom，不使用此工具也会有同样的结果，甚至更严重，应该对请求做限制或做其他优化
|
||||
* </pre>
|
||||
*
|
||||
* @param poolSize 线程池大小
|
||||
* @param threadPoolPrefix 线程名前缀
|
||||
*/
|
||||
public RecyclableBatchThreadPoolExecutor(int poolSize, String threadPoolPrefix){
|
||||
AtomicInteger threadNumber = new AtomicInteger(1);
|
||||
ThreadFactory threadFactory = r -> {
|
||||
Thread t = new Thread(r, threadPoolPrefix + threadNumber.getAndIncrement());
|
||||
if (t.isDaemon()) {
|
||||
t.setDaemon(false);
|
||||
}
|
||||
if (t.getPriority() != Thread.NORM_PRIORITY) {
|
||||
t.setPriority(Thread.NORM_PRIORITY);
|
||||
}
|
||||
return t;
|
||||
};
|
||||
this.executor = new ThreadPoolExecutor(poolSize, poolSize, 0L, TimeUnit.MILLISECONDS, new LinkedBlockingQueue<>(),threadFactory);
|
||||
}
|
||||
|
||||
/**
|
||||
* 自定义线程池，一般不需要使用
|
||||
* @param executor 线程池
|
||||
*/
|
||||
public RecyclableBatchThreadPoolExecutor(ExecutorService executor){
|
||||
this.executor = executor;
|
||||
}
|
||||
|
||||
/**
|
||||
* 关闭线程池
|
||||
*/
|
||||
public void shutdown(){
|
||||
executor.shutdown();
|
||||
}
|
||||
|
||||
/**
|
||||
* 获取线程池
|
||||
* @return ExecutorService
|
||||
*/
|
||||
public ExecutorService getExecutor(){
|
||||
return executor;
|
||||
}
|
||||
|
||||
/**
|
||||
* 分批次处理数据
|
||||
* <pre>
|
||||
* 1.所有批次执行完成后会过滤null并返回合并结果，保持输入数据顺序，不需要结果{@link Function}返回null即可
|
||||
* 2.{@link Function}需自行处理异常、保证线程安全
|
||||
* 3.原始数据在分片后可能被外部修改，导致批次数据不一致，如有必要，传参之前进行数据拷贝
|
||||
* 4.主线程会参与处理批次数据，如果要异步执行任务请使用普通线程池
|
||||
* </pre>
|
||||
*
|
||||
* @param <T> 输入数据类型
|
||||
* @param <R> 输出数据类型
|
||||
* @param data 待处理数据集合
|
||||
* @param batchSize 每批次数据量
|
||||
* @param processor 单条数据处理函数
|
||||
* @return 处理结果集合
|
||||
*/
|
||||
public <T,R> List<R> process(List<T> data, int batchSize, Function<T,R> processor) {
|
||||
if (batchSize < 1) {
|
||||
throw new IllegalArgumentException("batchSize >= 1");
|
||||
}
|
||||
List<List<T>> batches = splitData(data, batchSize);
|
||||
int batchCount = batches.size();
|
||||
int minusOne = batchCount - 1;
|
||||
ArrayDeque<IdempotentTask<R>> taskQueue = new ArrayDeque<>(minusOne);
|
||||
Map<Integer,Future<TaskResult<R>>> futuresMap = new HashMap<>();
|
||||
// 提交前 batchCount-1 批任务
|
||||
for (int i = 0 ; i < minusOne ; i++) {
|
||||
final int index = i;
|
||||
IdempotentTask<R> task = new IdempotentTask<>(i,() -> processBatch(batches.get(index), processor));
|
||||
taskQueue.add(task);
|
||||
futuresMap.put(i,executor.submit(task));
|
||||
}
|
||||
@SuppressWarnings("unchecked")
|
||||
List<R>[] resultArr = new ArrayList[batchCount];
|
||||
// 处理最后一批
|
||||
resultArr[minusOne] = processBatch(batches.get(minusOne), processor);
|
||||
// 处理剩余任务
|
||||
processRemainingTasks(taskQueue, futuresMap,resultArr);
|
||||
//排序、过滤null
|
||||
return Stream.of(resultArr)
|
||||
.filter(Objects::nonNull)
|
||||
.flatMap(List::stream)
|
||||
.collect(Collectors.toList());
|
||||
}
|
||||
|
||||
/**
|
||||
* 处理剩余任务并收集结果
|
||||
* @param taskQueue 任务队列
|
||||
* @param futuresMap 异步任务映射
|
||||
* @param resultArr 结果存储数组
|
||||
*/
|
||||
private <R> void processRemainingTasks(Queue<IdempotentTask<R>> taskQueue, Map<Integer,Future<TaskResult<R>>> futuresMap, List<R>[] resultArr) {
|
||||
// 主消费未执行任务
|
||||
IdempotentTask<R> task;
|
||||
while ((task = taskQueue.poll()) != null) {
|
||||
try {
|
||||
TaskResult<R> call = task.call();
|
||||
if (call.effective) {
|
||||
// 取消被主线程执行任务
|
||||
Future<TaskResult<R>> future = futuresMap.remove(task.index);
|
||||
future.cancel(false);
|
||||
//加入结果集
|
||||
resultArr[task.index] = call.result;
|
||||
}
|
||||
} catch (Exception e) {
|
||||
// 不处理异常
|
||||
throw new RuntimeException(e);
|
||||
}
|
||||
}
|
||||
futuresMap.forEach((index,future)->{
|
||||
try {
|
||||
TaskResult<R> taskResult = future.get();
|
||||
if(taskResult.effective){
|
||||
resultArr[index] = taskResult.result;
|
||||
}
|
||||
} catch (InterruptedException | ExecutionException e) {
|
||||
throw new RuntimeException(e);
|
||||
}
|
||||
});
|
||||
}
|
||||
|
||||
/**
|
||||
* 幂等任务包装类，确保任务只执行一次
|
||||
*/
|
||||
private static class IdempotentTask<R> implements Callable<TaskResult<R>> {
|
||||
|
||||
private final int index;
|
||||
private final Callable<List<R>> delegate;
|
||||
private final AtomicBoolean executed = new AtomicBoolean(false);
|
||||
|
||||
IdempotentTask(int index,Callable<List<R>> delegate) {
|
||||
this.index = index;
|
||||
this.delegate = delegate;
|
||||
}
|
||||
|
||||
@Override
|
||||
public TaskResult<R> call() throws Exception {
|
||||
if (executed.compareAndSet(false, true)) {
|
||||
return new TaskResult<>(delegate.call(), true);
|
||||
}
|
||||
return new TaskResult<>(null, false);
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* 结果包装类，标记结果有效性
|
||||
*/
|
||||
private static class TaskResult<R>{
|
||||
private final List<R> result;
|
||||
private final boolean effective;
|
||||
TaskResult(List<R> result, boolean effective){
|
||||
this.result = result;
|
||||
this.effective = effective;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* 数据分片方法
|
||||
* @param data 原始数据
|
||||
* @param batchSize 每批次数据量
|
||||
* @return 分片后的二维集合
|
||||
*/
|
||||
private static <T> List<List<T>> splitData(List<T> data, int batchSize) {
|
||||
int batchCount = (data.size() + batchSize - 1) / batchSize;
|
||||
return new AbstractList<List<T>>() {
|
||||
@Override
|
||||
public List<T> get(int index) {
|
||||
int from = index * batchSize;
|
||||
int to = Math.min((index + 1) * batchSize, data.size());
|
||||
return data.subList(from, to);
|
||||
}
|
||||
|
||||
@Override
|
||||
public int size() {
|
||||
return batchCount;
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
* 单批次数据处理
|
||||
* @param batch 单批次数据
|
||||
* @param processor 处理函数
|
||||
* @return 处理结果
|
||||
*/
|
||||
private static <T,R> List<R> processBatch(List<T> batch, Function<T,R> processor) {
|
||||
return batch.stream().map(processor).filter(Objects::nonNull).collect(Collectors.toList());
|
||||
}
|
||||
|
||||
/**
|
||||
* 处理Warp数组
|
||||
*
|
||||
* <pre>{@code
|
||||
* Warp<String> warp1 = Warp.of(this::select1);
|
||||
* Warp<List<String>> warp2 = Warp.of(this::select2);
|
||||
* executor.processByWarp(warp1, warp2);
|
||||
* String r1 = warp1.get();
|
||||
* List<String> r2 = warp2.get();
|
||||
* }</pre>
|
||||
*
|
||||
* @param warps Warp数组
|
||||
* @return Warp集合,此方法返回结果为空的不会被过滤
|
||||
*/
|
||||
public List<Warp<?>> processByWarp(Warp<?>... warps) {
|
||||
return processByWarp(Arrays.asList(warps));
|
||||
}
|
||||
|
||||
/**
|
||||
* 处理Warp集合
|
||||
* @param warps Warp集合
|
||||
* @return Warp集合,此方法返回结果为空的不会被过滤
|
||||
*/
|
||||
public List<Warp<?>> processByWarp(List<Warp<?>> warps) {
|
||||
return process(warps, 1, Warp::execute);
|
||||
}
|
||||
|
||||
/**
|
||||
* 处理逻辑包装类
|
||||
* @param <R> 结果类型
|
||||
*/
|
||||
public static class Warp<R>{
|
||||
|
||||
private Warp(Supplier<R> supplier){
|
||||
Objects.requireNonNull(supplier);
|
||||
this.supplier = supplier;
|
||||
}
|
||||
|
||||
/**
|
||||
* 创建Warp
|
||||
* @param supplier 执行逻辑
|
||||
* @return Warp
|
||||
* @param <R> 结果类型
|
||||
*/
|
||||
public static <R> Warp<R> of(Supplier<R> supplier){
|
||||
return new Warp<>(supplier);
|
||||
}
|
||||
|
||||
private final Supplier<R> supplier;
|
||||
private R result;
|
||||
|
||||
/**
|
||||
* 获取结果
|
||||
* @return 结果
|
||||
*/
|
||||
public R get() {
|
||||
return result;
|
||||
}
|
||||
|
||||
/**
|
||||
* 执行
|
||||
* @return this
|
||||
*/
|
||||
public Warp<R> execute() {
|
||||
result = supplier.get();
|
||||
return this;
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
}
|
||||
@ -0,0 +1,115 @@
|
||||
package cn.hutool.core.thread;
|
||||
|
||||
import cn.hutool.core.thread.RecyclableBatchThreadPoolExecutor.Warp;
|
||||
import org.junit.jupiter.api.Test;
|
||||
|
||||
import java.util.*;
|
||||
import java.util.concurrent.ExecutionException;
|
||||
import java.util.concurrent.ExecutorService;
|
||||
import java.util.concurrent.Executors;
|
||||
import java.util.concurrent.Future;
|
||||
import java.util.function.Function;
|
||||
|
||||
/**
|
||||
* {@link RecyclableBatchThreadPoolExecutor} 测试类
|
||||
*/
|
||||
public class RecyclableBatchThreadPoolExecutorTest {
|
||||
|
||||
|
||||
/**
|
||||
* 批量处理数据
|
||||
* @throws InterruptedException
|
||||
*/
|
||||
@Test
|
||||
public void test() throws InterruptedException {
|
||||
int corePoolSize = 10;// 线程池大小
|
||||
int batchSize = 100;// 每批次数据量
|
||||
int clientCount = 30;// 调用者数量
|
||||
test(corePoolSize,batchSize,clientCount);
|
||||
}
|
||||
|
||||
/**
|
||||
* 普通查询接口加速
|
||||
*/
|
||||
@Test
|
||||
public void test2() {
|
||||
RecyclableBatchThreadPoolExecutor executor = new RecyclableBatchThreadPoolExecutor(10);
|
||||
long s = System.nanoTime();
|
||||
Warp<String> warp1 = Warp.of(this::select1);
|
||||
Warp<List<String>> warp2 = Warp.of(this::select2);
|
||||
executor.processByWarp(warp1, warp2);
|
||||
Map<String, Object> map = new HashMap<>();
|
||||
map.put("key1",warp1.get());
|
||||
map.put("key2",warp2.get());
|
||||
long d = System.nanoTime() - s;
|
||||
System.out.printf("总耗时：%.2f秒%n",d/1e9);
|
||||
System.out.println(map);
|
||||
}
|
||||
|
||||
public void test(int corePoolSize,int batchSize,int clientCount ) throws InterruptedException{
|
||||
RecyclableBatchThreadPoolExecutor processor = new RecyclableBatchThreadPoolExecutor(corePoolSize);
|
||||
// 模拟多个调用者线程提交任务
|
||||
ExecutorService testExecutor = Executors.newFixedThreadPool(clientCount);
|
||||
Map<Integer, List<Integer>> map = new HashMap<>();
|
||||
for(int i = 0; i < clientCount; i++){
|
||||
map.put(i,testDate(1000));
|
||||
}
|
||||
long s = System.nanoTime();
|
||||
List<Future<?>> futures = new ArrayList<>();
|
||||
for (int j = 0; j < clientCount; j++) {
|
||||
final int clientId = j;
|
||||
Future<?> submit = testExecutor.submit(() -> {
|
||||
Function<Integer, String> function = p -> {
|
||||
try {
|
||||
Thread.sleep(10);
|
||||
} catch (InterruptedException e) {
|
||||
throw new RuntimeException(e);
|
||||
}
|
||||
return Thread.currentThread().getName() + "#" + p;
|
||||
};
|
||||
long start = System.nanoTime();
|
||||
List<String> process = processor.process(map.get(clientId), batchSize, function);
|
||||
long duration = System.nanoTime() - start;
|
||||
System.out.printf("【clientId：%s】处理结果：%s\n处理耗时：%.2f秒%n", clientId, process, duration / 1e9);
|
||||
});
|
||||
futures.add(submit);
|
||||
}
|
||||
futures.forEach(p-> {
|
||||
try {
|
||||
p.get();
|
||||
} catch (InterruptedException | ExecutionException e) {
|
||||
throw new RuntimeException(e);
|
||||
}
|
||||
});
|
||||
long d = System.nanoTime() - s;
|
||||
System.out.printf("总耗时：%.2f秒%n",d/1e9);
|
||||
testExecutor.shutdown();
|
||||
processor.shutdown();
|
||||
}
|
||||
public static List<Integer> testDate(int count){
|
||||
List<Integer> list = new ArrayList<>();
|
||||
for(int i = 1;i<=count;i++){
|
||||
list.add(i);
|
||||
}
|
||||
return list;
|
||||
}
|
||||
|
||||
private String select1() {
|
||||
try {
|
||||
Thread.sleep(3000);
|
||||
} catch (InterruptedException e) {
|
||||
throw new RuntimeException(e);
|
||||
}
|
||||
return "1";
|
||||
}
|
||||
|
||||
private List<String> select2() {
|
||||
try {
|
||||
Thread.sleep(5000);
|
||||
} catch (InterruptedException e) {
|
||||
throw new RuntimeException(e);
|
||||
}
|
||||
return Arrays.asList("1","2","3");
|
||||
}
|
||||
|
||||
}
|
||||
Loading…
Reference in New Issue
Block a user