Gio*_*kis 5 java nonblocking completable-future
我试图理解 CompletableFutures 和返回已完成的 future 的调用链，我创建了下面的示例，它模拟了对数据库的两次调用。
第一个方法应该提供一个包含 userId 列表的完整 future，然后我需要调用另一个提供 userId 的方法来获取用户（在本例中为字符串）。
总结一下：
我创建了简单的方法来模拟睡眠线程的响应。请检查下面的代码
public class PipelineOfTasksExample {
private Map<Long, String> db = new HashMap<>();
PipelineOfTasksExample() {
db.put(1L, "user1");
db.put(2L, "user2");
db.put(3L, "user3");
db.put(4L, "user4");
}
private CompletableFuture<List<Long>> returnUserIdsFromDb() {
try {
Thread.sleep(500);
} catch (InterruptedException e) {
e.printStackTrace();
}
System.out.println("building the list of Ids" + " - thread: " + Thread.currentThread().getName());
return CompletableFuture.supplyAsync(() -> Arrays.asList(1L, 2L, 3L, 4L));
}
private CompletableFuture<String> fetchById(Long id) {
CompletableFuture<String> cfId = CompletableFuture.supplyAsync(() -> db.get(id));
try {
Thread.sleep(500);
} catch (InterruptedException e) {
e.printStackTrace();
}
System.out.println("fetching id: " + id + " -> " + db.get(id) + " thread: " + Thread.currentThread().getName());
return cfId;
}
public static void main(String[] args) {
PipelineOfTasksExample example = new PipelineOfTasksExample();
CompletableFuture<List<String>> result = example.returnUserIdsFromDb()
.thenCompose(listOfIds ->
CompletableFuture.supplyAsync(
() -> listOfIds.parallelStream()
.map(id -> example.fetchById(id).join())
.collect(Collectors.toList()
)
)
);
System.out.println(result.join());
}
}
我的问题是， join 调用 ( example.fetchById(id).join()) 是否会破坏进程的非阻塞性质？如果答案是肯定的，我该如何解决这个问题？
先感谢您
Hol*_*ger 10
您的示例有点奇怪，因为您returnUserIdsFromDb()在任何操作开始之前都减慢了主线程的速度，同样，fetchById减慢了调用者而不是异步操作的速度，这违背了异步操作的整个目的。
此外，.thenCompose(listOfIds -> CompletableFuture.supplyAsync(() -> \xe2\x80\xa6))您可以简单地使用.thenApplyAsync(listOfIds -> \xe2\x80\xa6).
所以一个更好的例子可能是
\n\npublic class PipelineOfTasksExample {\n private final Map<Long, String> db = LongStream.rangeClosed(1, 4).boxed()\n .collect(Collectors.toMap(id -> id, id -> "user"+id));\n\n PipelineOfTasksExample() {}\n\n private static <T> T slowDown(String op, T result) {\n LockSupport.parkNanos(TimeUnit.MILLISECONDS.toNanos(500));\n System.out.println(op + " -> " + result + " thread: "\n + Thread.currentThread().getName()+ ", "\n + POOL.getPoolSize() + " threads");\n return result;\n }\n private CompletableFuture<List<Long>> returnUserIdsFromDb() {\n System.out.println("trigger building the list of Ids - thread: "\n + Thread.currentThread().getName());\n return CompletableFuture.supplyAsync(\n () -> slowDown("building the list of Ids", Arrays.asList(1L, 2L, 3L, 4L)),\n POOL);\n }\n private CompletableFuture<String> fetchById(Long id) {\n System.out.println("trigger fetching id: " + id + " thread: "\n + Thread.currentThread().getName());\n return CompletableFuture.supplyAsync(\n () -> slowDown("fetching id: " + id , db.get(id)), POOL);\n }\n\n static ForkJoinPool POOL = new ForkJoinPool(2);\n\n public static void main(String[] args) {\n PipelineOfTasksExample example = new PipelineOfTasksExample();\n CompletableFuture<List<String>> result = example.returnUserIdsFromDb()\n .thenApplyAsync(listOfIds ->\n listOfIds.parallelStream()\n .map(id -> example.fetchById(id).join())\n .collect(Collectors.toList()\n ),\n POOL\n );\n System.out.println(result.join());\n }\n}\n打印类似的东西
\n\npublic class PipelineOfTasksExample {\n private final Map<Long, String> db = LongStream.rangeClosed(1, 4).boxed()\n .collect(Collectors.toMap(id -> id, id -> "user"+id));\n\n PipelineOfTasksExample() {}\n\n private static <T> T slowDown(String op, T result) {\n LockSupport.parkNanos(TimeUnit.MILLISECONDS.toNanos(500));\n System.out.println(op + " -> " + result + " thread: "\n + Thread.currentThread().getName()+ ", "\n + POOL.getPoolSize() + " threads");\n return result;\n }\n private CompletableFuture<List<Long>> returnUserIdsFromDb() {\n System.out.println("trigger building the list of Ids - thread: "\n + Thread.currentThread().getName());\n return CompletableFuture.supplyAsync(\n () -> slowDown("building the list of Ids", Arrays.asList(1L, 2L, 3L, 4L)),\n POOL);\n }\n private CompletableFuture<String> fetchById(Long id) {\n System.out.println("trigger fetching id: " + id + " thread: "\n + Thread.currentThread().getName());\n return CompletableFuture.supplyAsync(\n () -> slowDown("fetching id: " + id , db.get(id)), POOL);\n }\n\n static ForkJoinPool POOL = new ForkJoinPool(2);\n\n public static void main(String[] args) {\n PipelineOfTasksExample example = new PipelineOfTasksExample();\n CompletableFuture<List<String>> result = example.returnUserIdsFromDb()\n .thenApplyAsync(listOfIds ->\n listOfIds.parallelStream()\n .map(id -> example.fetchById(id).join())\n .collect(Collectors.toList()\n ),\n POOL\n );\n System.out.println(result.join());\n }\n}\n乍一看，线程数量可能令人惊讶。
\n\n答案是，join()可能会阻塞线程，但如果这种情况发生在 Fork/Join 池的工作线程内，这种情况将被检测到，并启动一个新的补偿线程，以确保配置的目标并行度。
作为一种特殊情况，当使用默认的 Fork/Join 池时，实现可能会在方法内选取新的挂起任务join()，以确保同一线程内的进度。
因此，代码总是会取得进展，join()如果替代方案要复杂得多，偶尔调用也没什么问题，但如果过度使用，则存在资源消耗过多的危险。毕竟，使用线程池的原因是为了限制线程数量。
另一种方法是在可能的情况下使用链式依赖操作。
\n\npublic class PipelineOfTasksExample {\n private final Map<Long, String> db = LongStream.rangeClosed(1, 4).boxed()\n .collect(Collectors.toMap(id -> id, id -> "user"+id));\n\n PipelineOfTasksExample() {}\n\n private static <T> T slowDown(String op, T result) {\n LockSupport.parkNanos(TimeUnit.MILLISECONDS.toNanos(500));\n System.out.println(op + " -> " + result + " thread: "\n + Thread.currentThread().getName()+ ", "\n + POOL.getPoolSize() + " threads");\n return result;\n }\n private CompletableFuture<List<Long>> returnUserIdsFromDb() {\n System.out.println("trigger building the list of Ids - thread: "\n + Thread.currentThread().getName());\n return CompletableFuture.supplyAsync(\n () -> slowDown("building the list of Ids", Arrays.asList(1L, 2L, 3L, 4L)),\n POOL);\n }\n private CompletableFuture<String> fetchById(Long id) {\n System.out.println("trigger fetching id: " + id + " thread: "\n + Thread.currentThread().getName());\n return CompletableFuture.supplyAsync(\n () -> slowDown("fetching id: " + id , db.get(id)), POOL);\n }\n\n static ForkJoinPool POOL = new ForkJoinPool(2);\n\n public static void main(String[] args) {\n PipelineOfTasksExample example = new PipelineOfTasksExample();\n\n CompletableFuture<List<String>> result = example.returnUserIdsFromDb()\n .thenComposeAsync(listOfIds -> {\n List<CompletableFuture<String>> jobs = listOfIds.parallelStream()\n .map(id -> example.fetchById(id))\n .collect(Collectors.toList());\n return CompletableFuture.allOf(jobs.toArray(new CompletableFuture<?>[0]))\n .thenApply(_void -> jobs.stream()\n .map(CompletableFuture::join).collect(Collectors.toList()));\n },\n POOL\n );\n\n System.out.println(result.join());\n System.out.println(ForkJoinPool.commonPool().getPoolSize());\n }\n}\n不同之处在于，首先提交所有异步作业，然后join调度调用它们的依赖操作，仅在所有作业完成时执行，因此这些join调用永远不会阻塞。join只有方法末尾的最终调用main可能会阻塞主线程。
所以这会打印出类似的东西
\n\ntrigger building the list of Ids - thread: main\nbuilding the list of Ids -> [1, 2, 3, 4] thread: ForkJoinPool-1-worker-1, 1 threads\ntrigger fetching id: 2 thread: ForkJoinPool-1-worker-0\ntrigger fetching id: 3 thread: ForkJoinPool-1-worker-1\ntrigger fetching id: 4 thread: ForkJoinPool-1-worker-2\nfetching id: 4 -> user4 thread: ForkJoinPool-1-worker-3, 4 threads\nfetching id: 2 -> user2 thread: ForkJoinPool-1-worker-3, 4 threads\nfetching id: 3 -> user3 thread: ForkJoinPool-1-worker-2, 4 threads\ntrigger fetching id: 1 thread: ForkJoinPool-1-worker-3\nfetching id: 1 -> user1 thread: ForkJoinPool-1-worker-2, 4 threads\n[user1, user2, user3, user4]\n显示无需创建补偿线程，因此线程数与配置的目标并行度匹配。
\n\n请注意，如果实际工作是在后台线程中完成的，而不是在fetchById方法本身内完成的，那么您现在不再需要并行流，因为没有阻塞调用join()。对于这样的场景，仅仅使用stream()通常会带来更高的性能。