JAVA语言异步非阻塞设计模式（应用篇）

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/55/55c6af122f7b2ec148b24c32b3cc3360.gif","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"1.概述","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"本系列文章共2篇。在上一篇《原理篇》中，我们看到了异步非阻塞模型，它能够有效降低线程IO状态的耗时，提升资源利用率和系统吞吐量。异步API可以表现为listener或Promise形式；其中Promise API提供了更强的灵活性，支持同步返回和异步回调，也允许注册任意数目的回调。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在本文《应用篇》中，我们将进一步探索异步模式和Promise的应用：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"第2章：Promise与线程池。","attrs":{}},{"type":"text","text":" 在异步执行耗时请求时，ExecutorService+Future是一个备选方案；但是相比于Future，Promise支持纯异步获取响应数据，能够消除更多阻塞。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"第3章：异常处理。","attrs":{}},{"type":"text","text":" Java程序并不总能成功执行请求，有时会遇到网络问题等不可抗力。对于无法避免的异常情况，异步API必须提供异常处理机制，以提升程序的容错性。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"第4章：请求调度。","attrs":{}},{"type":"text","text":" Java程序有时需要提交多条请求，这些请求之间可能存在一定的关联关系，包括顺序执行、并行执行、批量执行。异步API需要对这些约束提供支持。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"本文不限定Promise的具体实现，读者在生产环境可以选择一个Promise工具类（如netty DefaultPromise[A]、jdk CompletableFuture[B]等）；此外，由于Promise的原理并不复杂，读者也可以自行实现所需功能。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"2.Promise与线程池","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Java程序有时需要执行耗时的IO操作，如数据库访问；在此期间，相比于纯内存计算，IO操作的持续时间明显更长。为了减少IO阻塞、提高资源利用率，我们应该使用异步模型，将请求提交到其他线程中执行，从而连续提交多条请求，而不必等待之前的请求返回。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"本章对几种IO模型进行对比（见2.1节），考察调用者线程的阻塞情况。其中，Promise支持纯异步的请求提交及响应数据处理，能够最大程度地消除不必要的阻塞。在实际项目中，如果底层API不支持纯异步，那么我们也可以进行适当重构，使其和Promise兼容（见2.2节）。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"2.1 对比：同步、Future、Promise","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"本节对几种 IO 模型进行对比，包括同步 IO、基于线程池（ExecutorService）的异步 IO、基于 Promise 的异步IO，考察调用者线程的阻塞情况。假设我们要执行数据库访问请求。由于需要跨越网络，单条请求需要进行耗时的 IO操作，才能最终收到响应数据；但是请求之间没有约束，允许随时提交新的请求，而不需要收到之前的响应数据。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先我们来看看","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"几种模型的样例代码","attrs":{}},{"type":"text","text":"：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"1.同步IO","attrs":{}},{"type":"text","text":"。db.writeSync()方法是同步阻塞的。函数阻塞，直至收到响应数据。因此，调用者一次只能提交一个请求，必须等待该请求返回，才能再提交下一个请求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"/* 提交请求并阻塞，直至收到响应数据*/\nString result = db.writeSync(\"data\");\nprocess(result);","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"2.基于线程池（ExecutorService）的异步IO","attrs":{}},{"type":"text","text":"。db.writeSync()方法不变；但是将其提交到线程池中来执行，使得调用者线程不会阻塞，从而可以连续提交多条请求data1-3。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"提交请求后，线程池返回 Future 对象，调用者调用 Future.get() 以获取响应数据。Future.get() 方法却是阻塞的，因此调用者在获得响应数据之前无法再提交后续请求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"/* 提交请求*/\n// executor: ExecutorService\nFuture resultFuture1 = executor.submit(() -> db.writeSync(\"data1\"));\nFuture resultFuture2 = executor.submit(() -> db.writeSync(\"data2\"));\nFuture resultFuture3 = executor.submit(() -> db.writeSync(\"data3\"));\n\n/* 获取响应：同步*/\nString result1 = resultFuture1.get();\nString result2 = resultFuture2.get();\nString result3 = resultFuture3.get();\nprocess(result1);\nprocess(result2);\nprocess(result3);","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"3.基于Promise的异步IO","attrs":{}},{"type":"text","text":"。db.writeAsync()方法是纯异步的，提交请求后返回 Promise 对象；调用者调用Promise.await()注册回调，当收到响应数据后触发回调。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在《原理篇》中，我们看到了 Promise API 可以基于线程池或响应式模型实现；不论哪种方式，回调函数可以在接收响应的线程中执行，而不需要调用者线程阻塞地等待响应数据。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"/* 提交请求*/\nPromise resultPromise1 = db.writeAsync(\"data1\");\nPromise resultPromise2 = db.writeAsync(\"data2\");\nPromise resultPromise3 = db.writeAsync(\"data3\");\n\n/* 获取响应：异步*/\nresultPromise1.await(result1 -> process(result1));\nresultPromise2.await(result2 -> process(result2));\nresultPromise3.await(result3 -> process(result3));","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"接下来我们看看以上几种模型中，调用者线程状态随时间变化的过程，如图2-1所示。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"a.","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"同步 IO","attrs":{}},{"type":"text","text":"。调用者一次只能提交一个请求，在收到响应之前不能提交下一个请求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"b.","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"基于线程池的异步 IO","attrs":{}},{"type":"text","text":"。同一组请求（请求1-3，以及请求4-6）可以连续提交，而不需要等待前一条请求返回。然而，一旦调用者使用 Future.get() 获取响应数据（result1-3），就会阻塞而无法再提交下一组请求（请求4-6），直至实际收到响应数据。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"c.","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"基于 Promise 的异步 IO。","attrs":{}},{"type":"text","text":" 调用者随时可以提交请求，并向Promise注册对响应数据的回调函数；稍后接收线程向Promise通知响应数据，以触发回调函数。上述过程中，调用者线程不需要等待响应数据，始终不会阻塞。","attrs":{}}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/a4/a49ddbfa2a9bff6dfd83aa0ef0a2cc63.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图2-1a 线程时间线：同步IO","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/30/306334c7f3a16ff1a2d29dc7647d4664.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图2-1b 线程时间线：基于线程池的异步IO","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/2d/2df938bc26820ca9480e364ac055ac01.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图2-1c 线程时间线：基于 Promise 的异步IO","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"2.2 Promise结合线程池","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"和 ExecutorService+Future相比，Promise 具有纯异步的优点；然而在某些场景下也需要把 Promise 和线程池结合使用。例如：1.底层 API 只支持同步阻塞模型，不支持纯异步；此时只能在线程池中调用 API，才能做到非阻塞。2.需要重构一段遗留代码，将其线程模型从线程池模型改为响应式模型；可以先将对外接口改为 Promise API，而底层实现暂时使用线程池。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"下面的代码片段展示了 ","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"Promise 和线程池结合的用法：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"创建 Promise 对象作为返回值。注意这里使用了 PromiseOrException，以防期间遇到异常；其可以通知响应数据，也可以在失败时通知抛出的 Exception。详见3.1小节。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"在线程池中执行请求(2a)，并在收到响应数据后向 Promise 通知(2b)","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"处理线程池满异常。线程池底层关联一个 BlockingQueue 来存储待执行的任务，一般设置为有界队列以防无限占用内存，当队列满时会丢弃某个任务。为了向调用者通知该异常，线程池的拒绝策略须设置为 AbortPolicy，当队列满时丢弃所提交的任务，并抛出 RejectedExecutionException；一旦捕获该异常，就要向 Promise 通知请求失败。","attrs":{}}]}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"public PromiseOrException writeAsync() {\n// 1.创建Promise对象\n PromiseOrException resultPromise = new PromiseOrException<>();\n try {\n executor.execute(() -> {\n String result = db.writeSync(\"data\"); // 2a.执行请求。只支持同步阻塞\n resultPromise.signalAllWithResult(result); // 2b.通知Promise\n });\n\n }catch (RejectedExecutionException e){ // 3.异常：线程池满\n resultPromise.signalAllWithException(e); \n }\n\n return resultPromise;\n}","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"3.异常处理：PromiseOrException","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Java 程序有时会遇到不可避免的异常情况，如网络连接断开；因此，程序员需要设计适当的异常处理机制，以提升程序的容错性。本章介绍异步 API 的异常处理，首先介绍 Java 语言异常处理规范；然后介绍 Promise 的变体 PromiseOrException，使得 Promise API 支持规范的异常处理。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.1异常处理规范","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"个人认为，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"Java 代码的异常处理","attrs":{}},{"type":"text","text":"应当符合下列规范：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"显式区分正常出口和异常出口。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"支持编译时刻检查，强制调用者处理不可避免的异常。","attrs":{}}]}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}},{"type":"strong","attrs":{}}],"text":"区分正常出口和异常出口","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"异常是 Java 语言的重要特性，是一种基本的控制流。Java 语言中，一个函数允许有一个返回值，以及抛出多个不同类型的异常。函数的返回值是正常出口，函数返回说明函数能够正常工作，并计算出正确的结果；相反，一旦函数遇到异常情况无法继续工作，如网络连接断开、请求非法等，就要抛出相应的异常。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"虽然 if-else 和异常都是控制流，但是程序员必","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"须辨析二者的使用场景","attrs":{}},{"type":"text","text":"。if-else 的各个分支一般是对等的，都用于处理正常情况；而函数的返回值和异常是不对等的，抛出异常表示函数遇到无法处理的故障，已经无法正常计算结果，其与函数正常工作所产生的返回值有本质区别。在 API 设计中，混淆正常出口（返回值）与异常出口（抛出异常），或者在无法继续工作时不抛异常，都是严重的设计缺陷。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"以数据库访问为例，下面的代码对比了API进行异常处理的两种形式。数据库访问过程中，如果网络连接顺畅，并且服务端能够正确处理请求，那么db.write()应该返回服务端的响应数据，如服务端为所写数据生成的自增id、条件更新实际影响的数据条数等；如果网络连接断开，或者客户端和服务端版本不匹配导致请求无法解析，从而无法正常工作，那么db.write()应该抛出异常以说明具体原因。从“是否正常工作”的角度看，上述两种情况的性质是截然不同的，显然应该选用异常作为控制流，而不是 if-else。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"/* 正确*/\ntry {\n String result = db.write(\"data\");\n process(result); // 正常出口\n} catch (Exception e) {\n log.error(\"write fails\", e); // 异常出口\n}\n\n/* 错误*/\nString resultOrError = db.write(\"data\");\nif (resultOrError.equals(\"OK\")) {\n process(resultOrError); // 正常出口\n} else {\n log.error(\"write fails, error: \" + resultOrError); // 异常出口\n} ","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}},{"type":"strong","attrs":{}}],"text":"强制处理不可避免的异常","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Java 语言的异常处理体系中，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"异常主要分为以下几类","attrs":{}},{"type":"text","text":"：Exception、RuntimeException、Error；三者都是Throwable 的子类，即可以被函数抛出。注意，由于 RuntimeException 是 Exception 的子类，本文为避免混淆，“Exception”特指“是 Exception 但不是 RuntimeException ”的那些异常。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"个人认为，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"几种异常类型分别用于下列场景","attrs":{}},{"type":"text","text":"：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"1. Exception","attrs":{}},{"type":"text","text":"：程序外部的不可抗力造成的异常情况，如网络连接断开。即使 Java 代码完美无瑕，也绝对不可能避免这类异常（拔掉网线试试！）。既然无法避免，这种异常就应当强制处理，以提升系统的容错能力。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"2. RuntimeException","attrs":{}},{"type":"text","text":"：编程错误造成的异常情况，如数组下标越界ArrayOutOfBoundException、参数不符合取值范围 IllegalArgumentException 等。如果程序员对 API 的输入约束了如指掌，并在调用 API 之前对函数参数进行适当校验，那么 RuntimeException 是可以绝对避免的（除非被调 API 在应当抛 Exception 处，实际抛出了RuntimeException）。既然可以避免，这种异常就没有必要强制处理。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当然，人无完人。假设程序员真的违背了某些约束，函数抛出 RuntimeException 且未被处理，那么作为惩罚，线程或进程会退出，从而提醒程序员改正错误代码。如果线程或进程必须常驻，就要对 RuntimeException 进行兜底，如下面的代码所示。这里将代码缺陷视为无法避免的异常情况，捕获异常后可以记录日志、触发告警，提醒稍后来修正缺陷。","attrs":{}}]},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"new Thread(()->{\n while (true){\n try{\n doSomething();\n }catch (RuntimeException e){ // 对RuntimeException进行兜底，以防线程中断\n log.error(\"error occurs\", e);\n }\n }\n});","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"3. Error：jvm内部定义的异常","attrs":{}},{"type":"text","text":"，如 OutOfMemoryError。业务逻辑一般不抛出 Error，而是抛出某种Exception或 RuntimeException。","attrs":{}}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"上述几类异常中，只有 Exception 是强制处理的，称为 checked exception[C]。如下所示是一个 checked exception的例子。数据库访问DB.write()抛出Exception，表示遇到网络断开、消息解析失败等不可抗情况。异常类型为Exception而不是RuntimeException，以强制调用者添加catch子句处理上述情况；如调用者遗漏了 catch子句，则编译器会报错，从而提示调用者“这里一定会遇到异常情况，必须进行处理”，以完善程序容错能力。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"/**\n * 抛出异常，如果：\n * 1.网络连接断开\n * 2.消息无法解析\n * 3.业务逻辑相关，如服务端扣款时发现余额不足\n * 4.…… // 任何无法避免的情况，都应该抛出Exception！\n */\npublic String write(Object data) throws Exception {\n return \"\";\n}\n\n/**\n * 处理异常\n */\ntry {\n String result = db.write(\"data\");\n process(result); \n} catch (Exception e) { // 如遗漏catch子句，则编译不通过\n log.error(\"write fails, db: ..., data: ...\", e);\n}","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"3.2 Promise API的异常处理","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"上一小节讨论了异常处理的规范：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"显式区分正常出口和异常出口；","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"不可抗的异常，要在编译时刻强制处理。下面的代码展示了 Promise API 要如何设计异常处理机制，以符合上述规范。","attrs":{}}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"使用PromiseOrException来通知响应数据和异常","attrs":{}},{"type":"text","text":"。PromiseOrException是Promise的子类，泛型模版X为数据对象ResultOrException，其含有2个字段result和e：e==null表示正常，此时字段result有效；e!=null表示异常，此时不要使用字段result。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"在“重载1”中，调用者从回调函数中获得ResultOrException对象","attrs":{}},{"type":"text","text":"。调用ResultOrException.get()获取响应数据result，或者get()方法抛出异常e。这种方式的代码结构和传统的异常处理一致，可以使用多个catch子句分别处理不同类型的异常。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"在“重载2”中，调用者从回调函数中直接获得result和e","attrs":{}},{"type":"text","text":"。含义同上。这种方式省去了ResultOrException.get()；但是如果需要处不同类型的异常，则需要用e instanceof MyException来判断异常类型。","attrs":{}}]}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"// extends Promise>\n PromiseOrException resultPromsie = db.writeAsync(\"data\");\n\n /* 重载1*/ \nresultPromsie.await(resultOrException -> { \n try {\n String result = resultOrException.get();\n process(result); // 正常出口\n } catch (Exception e) {\n log.error(\"write fails\", e); // 异常出口\n }\n});\n \n /* 重载2*/\n resultPromsie.await((result, e) -> {\n if (e == null) {\n process(result); // 正常出口\n } else {\n log.error(\"write fails\", e); // 异常出口\n }\n});","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"PromiseOrException符合上一小节提出的异常处理规范，","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"具有如下优点","attrs":{}},{"type":"text","text":"：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"区分正常出口和异常出口。响应数据和异常分别使用 result 和e两个变量来传递，可以靠e==null来判断是否正常。注意result==null不能作为判断条件，因为null有可能是响应数据的合法值。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"强制处理异常。不论使用哪一种回调，不存在一种代码结构能够只获得 result 而不获得e，因此语法上不会遗漏e的异常处理。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"允许定义异常类型。PromiseOrException 的泛型模版E填为 Excetion 不是必需的，也可以填为任意其他类型。注意，受限于 Java 语法，泛型模版处只允许填写一种异常类型，而不像函数抛异常那样允许抛出多种异常。为应对这种限制，我们只能为 API 定义一个异常父类，调用者用 catch 子句或 instanceof 进行向下转型。当然，这种“定义异常父类”的做法也是可以接受的，并在现有工具中广泛应用，因为可以将工具所抛异常区别于Java语言内置的异常类型。","attrs":{}}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最后，在异常处理结构方面个人提出","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"一个建议","attrs":{}},{"type":"text","text":"：全部异常通过 PromiseOrException 来通知，而 API 本身不要抛出异常。以数据库访问 API writeAsync()为例，面的代码对比了两种抛异常的方式。","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"正确的做法","attrs":{}},{"type":"text","text":"是PromiseOrException 作为唯一出口，如果 API 底层实现抛出异常（submit() throws Exception），则应该将异常封装于 PromiseOrException 对象，而不应该直接从API函数抛出（writeAsync() throws Exception）。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"/* 正确：唯一出口PromiseOrException*/\npublic PromiseOrException writeAsync(Object data) {\n try {\n submit(data); // throws exception\n } catch (Exception e) {\n return PromiseOrException.immediatelyException(e);\n }\n PromiseOrException resultPromise = new PromiseOrException<>();\n return resultPromise;\n}\n\n/* 错误：两个出口throws Exception和PromiseOrException*/\npublic PromiseOrException writeAsync(Object data) throws Exception {\n submit(data); // throws exception\n\n PromiseOrException resultPromise = new PromiseOrException<>();\n return resultPromise;\n}","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果错误地设计了含有两个异常出口的 API，调用者就不得不重复书写异常处理逻辑，如下面的代码所示。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"try {\n PromiseOrException resultPromise = db.writeAsync(\"data\");\n resultPromise.await((result, e) -> {\n if (e == null) {\n process(result); // 正常出口\n } else {\n log.error(\"write fails\", e); // 异常出口2\n }\n });\n} catch (Exception e) {\n log.error(\"write fails\", e); // 异常出口1\n}","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"4.请求调度","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Java 程序中有时需要提交多条异步请求，且这些请求之间存在一定的关联关系。在异步非阻塞场景下，这些关联关系都可以借助 Promise 来实现。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"1.顺序请求","attrs":{}},{"type":"text","text":"，如图4-1所示。后一条请求依赖前一条请求的响应数据；因此，必须等待前一条请求返回，才能构造并提交下一条请求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/2c/2c2c5cb4f4bdfc3dcc9b9e4a285a8e25.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图4-1 顺序请求","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"2.并行请求","attrs":{}},{"type":"text","text":"，如图4-2所示。一次提交多条请求，然后等待全部请求返回。所提交的请求之间没有依赖关系，因此可以同时执行；但是必须收到每条请求的响应数据（发生channelRead()事件，事件参数为响应数据），才能执行实际的处理process(result1,2,3)。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/01/0155cd28684a39bc1ad81b5fbdf9bda0.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图4-2 并行请求","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"3.批量请求","attrs":{}},{"type":"text","text":"，如图4-3所示。调用者连续提交多条请求，但是暂存在队列中（offer()），而不是立刻执行。一段时间后，从队列中取出若干条请求，组装为批量请求来提交（writeBulk()）；当收到批量请求的响应消息时，可以从中取出每条请求的响应数据。由于每次网络IO都带来额外开销，故实际应用中经常使用批量请求来减少网络IO频率，以提升总体吞吐量。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/d3/d3bd46c78eae3a616bd1165d06b7640f.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图4-3 批量请求","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"4.1.顺序请求：Promise.then()","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"假设一系列操作需要依次完成，即前一操作完成后，才能开始执行下一操作；如果这些操作均表现为 Promise API，我们可以对 Promise.await(listener)进行封装，使代码结构更加简洁。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如下所示是一个异步 Promise API。submit 方法提交请求 request 并返回 Promise 对象；当收到响应数据时， 该 Promise 对象被通知。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"/**\n * 异步Promise API\n */\n public static Promise submit(Object request) {\n Promise resultPromise = new Promise<>();\n // ……\n return resultPromise;\n}","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"现假设有5个请求称为“A”-“E”，这些请求必须依次提交。例如，由于请求B的参数依赖请求 A 的响应数据，故提交请求A后必须先处理其响应数据 resultA，然后才能再提交请求B。这种场景可以用如下所示的代码来实现。某次调用submit(\"X\")函数后，我们在其返回的 Promise 对象上注册回调；回调函数内处理响应数据 resultX，并调用submit(\"X+1\")来提交下一请求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这种方式虽然能实现功能需求，但是嵌套式的代码结构可读性非常差——每增加一个请求，代码就要多嵌套、缩进一个层级。当调用逻辑复杂、请求数较多时，代码会非常难以维护。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这种情况也称为 ","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"“回调地狱”","attrs":{}},{"type":"text","text":"[D]，在 JavaScript 语言中相关讨论颇多，可以作为参考。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"submit(\"A\").await(resultA -> {\n submit(\"B\").await(resultB -> {\n submit(\"C\").await(resultC -> {\n submit(\"D\").await(resultD -> {\n submit(\"E\").await(resultE -> {\n process(resultE);\n });\n });\n });\n });\n});","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为改进代码结构，我们对 Promise.await(Consumer) 方法进行封装，提供Promise.then(Function>)方法，如下所示。类似于await()，then()也可以注册一个回调函数resultX->submit(\"X+1\")，回调函数处理响应数据resultX，并提交下一请求submit(\"X+1\")；then()的返回值即submit(\"X+1\")的返回值，用于通知下一请求的响应数据resultX+1。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"Promise resultPromiseA = submit(\"A\");\nPromise resultPromiseB = resultPromiseA.then(resultA -> submit(\"B\"));\nPromise resultPromiseC = resultPromiseB.then(resultB -> submit(\"C\"));\nPromise resultPromiseD = resultPromiseC.then(resultC -> submit(\"D\"));\nPromise resultPromiseE = resultPromiseD.then(resultD -> submit(\"E\"));\nresultPromiseE.await(resultE -> process(resultE));","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"接下来，我们将中间变量 resultPromiseA-E 内联，即得到基于then()的链式调用结构。相比于await()，then()消除了套娃般的嵌套回调。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"submit(\"A\")\n .then(resultA -> submit(\"B\")) // 返回resultPromiseB\n .then(resultB -> submit(\"C\")) // 返回resultPromiseC\n .then(resultC -> submit(\"D\")) // 返回resultPromiseD\n .then(resultD -> submit(\"E\")) // 返回resultPromiseE\n .await(resultE -> process(resultE));","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最后，我们来看一下 ","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"Promise.then() 的一种简单实现","attrs":{}},{"type":"text","text":"，如下所示：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"then()方法提供一个泛型模版Next，以说明下一请求的响应数据类型。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"根据泛型模版Next，then()内部创建Promise作为返回值，用于通知下一请求的响应数据。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"对于当前请求，调用await()注册响应数据的回调result；当收到响应数据后，执行函数func，以提交下一请求：func.apply(result)。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"当收到下一请求的响应数据后，Promise被通知：nextPromise::signalAll。","attrs":{}}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"public Promise then(Function> func) {\n Promise nextPromise = new Promise<>();\n await(result -> {\n Promise resultPromiseNext = func.apply(result);\n resultPromiseNext.await(nextPromise::signalAll);\n });\n return nextPromise;\n}","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"注意，这里只展示了纯异步重载Promise.then(Function>)。根据回调函数是否有返回值、同步执行还是异步执行，Promise可以提供then()的更多种重载；受限于Java语法，如编译器无法辨析各个重载，则可以使用函数名称进行显式区别，如：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"thenRun(Runnable)","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"thenAccept(Consumer)","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"thenApply(Function)","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"thenApplyAsync(Function>)","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"4.2.并行请求：LatchPromise","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"上一小节介绍了“顺序请求”的场景，即多条请求需要依次执行；而“并行请求”场景下，多条请求之间没有顺序约束，但是我们仍然需要等待全部请求返回，才能执行后续操作。例如，我们需要查询多张数据库表，这些查询语句可以同时执行；但是必须等待每条查询都返回，我们才能获得完整信息。jdk 提供CountDownLatch 来实现这一场景，但是其只支持同步等待；作为改进，我们采用 LatchPromise 实现相同的功能，并且支持纯异步API。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"以数据库访问为例，如下所示的代码","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"展示了 LatchPromise 的使用","attrs":{}},{"type":"text","text":"：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"提交3条请求，并获取每个请求所对应的 Promise 对象 resultPromise1-3，以获取响应数据。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"创建LatchPromise对象，并向其注册需要等待的 Promise 对象resultPromise1-3。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"LatchPromise.untilAllSignaled()返回一个 Promise 对象 allSignaled。当所注册的 resultPromise1-3均被通知后，allSignaled 会被通知。allSignaled 的类型为 VoidPromise，表示 allSignaled 被通知时没有需要处理的响应数据。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"在 allSignaled 上注册回调，在回调函数中调用 resultPromiseX.await() 获取实际的响应数据；此时由于请求已执行完毕，故 await() 立刻返回而不阻塞。","attrs":{}}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"/* 创建Promise对象*/\nPromise resultPromise1 = db.writeAsync(\"a\");\nPromise resultPromise2 = db.writeAsync(\"b\");\nPromise resultPromise3 = db.writeAsync(\"c\");\n\n/* 向LatchPromise注册要等待的Promise*/\nLatchPromise latch = new LatchPromise();\nlatch.add(resultPromise1);\nlatch.add(resultPromise2);\nlatch.add(resultPromise3);\n\n/* 等待全部Promise被通知*/\nVoidPromise allSignaled = latch.untilAllSignaled();\nallSignaled.await(() -> {\n String result1 = resultPromise1.await();\n String result2 = resultPromise2.await();\n String result3 = resultPromise3.await();\n process(result1, result2, result3);\n});","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"作为对比，下面的代码使用 CountDownLatch 实现相同功能，但是","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"存在以下缺陷","attrs":{}},{"type":"text","text":"：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"CountDownLatch.await() 只支持同步等待。在纯异步场景下是无法接受的。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"CountDownLatch 对业务逻辑有侵入性。程序员需要在业务逻辑中添加对 CountDownLatch.countDown()的调用，以控制CountDownLatch 的时序；相反，LatchPromise 依赖本来就已经存在的 resultPromise 对象，而不需要编写额外的时序控制代码。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"CountDownLatch 引入了冗余逻辑。创建 CountDownLatch 时，必须在构造参数中填写要等待的请求数；因此，一旦所提交的请求的数目改变，就必须相应地更新创建 CountDownLatch 的代码，修改构造参数。","attrs":{}}]}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"CountDownLatch latch = new CountDownLatch(3);\nresultPromise1.await(result1 -> latch.countDown());\nresultPromise2.await(result2 -> latch.countDown());\nresultPromise3.await(result3 -> latch.countDown());\n\nlatch.await();\nString result1 = resultPromise1.await();\nString result2 = resultPromise2.await();\nString result3 = resultPromise3.await();\nprocess(result1, result2, result3); ","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最后，我们来看一下 ","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"LatchPromise 的参考实现","attrs":{}},{"type":"text","text":"。代码原理如下所示：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"设立countUnfinished变量，记录还没有被通知的Promise对象的数目","attrs":{}},{"type":"text","text":"。每当注册一个Promise对象，countUnfinished递增；每当一个Promise被通知，countUnfinished递减。当countUnfinished减到0时，说明所注册全部Promise对象都被通知了，故通知allSignaled。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"设立noMore变量","attrs":{}},{"type":"text","text":"，记录是否还需要继续注册新的Promise对象，仅当调用了untilAllSignaled()才认为完成注册；在此之前，即使countUnfinished减至0，也不应该通知allSignaled。考虑这样一种情况：需要注册并等待resultPromise1-3，其中resultPromise1、2在注册期间即已被通知，而resultPromise3未被通知。如果不判断noMore，那么注册完resultPromise1、2后，countUnfinished即已减至0，导致提前通知allSignaled；这是一个时序错误，因为实际上resultPromise3还没有完成。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"为保证线程安全，访问变量时须上锁","attrs":{}},{"type":"text","text":"，此处使用synchronized来实现。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"注意","attrs":{}},{"type":"text","text":"，调用untilAllSignaled()时，如果countUnfinished的初值已经为0，则应立刻通知allSignaled；因为countUnfinished已经不可能再递减，之后没有机会再通知allSignaled了。","attrs":{}}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"// private static class Lock。无成员，仅用于synchronized(lock)\nprivate final Lock lock = new Lock();\nprivate int countUnfinished = 0;\nprivate final VoidPromise allSignaled = new VoidPromise();\n\npublic void add(Promise> promise) {\n if (promise.isSignaled()) {\n return;\n }\n\n synchronized (lock) {\n countUnfinished++;\n }\n\n promise.await(unused -> {\n synchronized (lock) {\n countUnfinished--;\n if (countUnfinished == 0 && noMore) {\n allSignaled.signalAll();\n }\n }\n });\n}\n\npublic VoidPromise untilAllSignaled() {\n synchronized (lock) {\n if (countUnfinished == 0) {\n allSignaled.signalAll();\n } else {\n noMore = true;\n }\n }\n\n return allSignaled;\n}","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"4.3.批量请求：ExecutorAsync","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}},{"type":"strong","attrs":{}}],"text":"批量请求的特性","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“批量请求”（也称“bulk”、“batch”）是指发送一条消息即可携带多条请求，主要用于数据库访问和远程调用等场景。由于减少了网络 IO 次数、节约了构造和传输消息的开销，批量请求能有效提升吞吐量。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"很多数据库 API 都支持批量读写，如JDBC PreparedStatement[E]、elasticsearch bulk API[F]、mongo DB insertMany()[G]、influx DB BatchPoints[H]，读者可以查阅参考文献进一步了解。为了提升性能，部分API会牺牲易用性。其中，elasticsearch bulk API 对调用者的限制最少，允许混杂增删改等不同类型的请求，允许写入不同的数据库表（index）；mongo DB、influx DB 次之，一个批量请求只能写入同一个数据库表，但是可以自定义每条数据的字段；PreparedStatement 的灵活性最低，其定义了 SQL 语句的模版，调用者只能填写模版参数，而不能修改语句结构。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"虽然数据库 API 已经支持批量访问，但是很多原生 API 仍然需要调用者自己构造批量请求，需要调用者处理请求组装、批量大小、并发请求数等复杂的细节。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在此，我们设计出通用组件 ExecutorAsync，封装请求调度策略以提供更简洁的API。","attrs":{}},{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"ExecutorAsync 的使用流程","attrs":{}},{"type":"text","text":"如下面的代码片段所示：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"类似于线程池 ExecutorService.submit()，调用者可以调用ExecutorAsync.submit()来提交一个请求。其中，请求以数据对象 Request 表示，用于存储请求类型和请求参数。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"提交请求后，调用者获得 Promise 对象，以获取响应数据。由于使用了 Promise，ExecutorAsync 支持纯异步操作，提交请求和获取响应数据都不需要阻塞。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"ExecutorAsync 内部对请求进行调度，并非提交一条请求就立刻执行，而是每隔固定时间收集一批请求，将其组装为一个批量请求，再调用实际的数据库访问 API。如果数据库访问 API 允许，那么一批请求可以混杂不同的请求类型，或者操作不同的数据库表。","attrs":{}}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"ExecutorAsync executor = new ExecutorAsync();\nPromise<...> resultPromise1 = executor.submit(new Request(\"data1\"));\nPromise<...> resultPromise2 = executor.submit(new Request(\"data2\"));\nPromise<...> resultPromise3 = executor.submit(new Request(\"data3\"));","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"具体而言，ExecutorAsync 支持如下调度策略：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"1.排队","attrs":{}},{"type":"text","text":"，如图4-4a所示。调用者提交请求Request后不要立刻执行，而是将其缓存在队列queue中。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/c2/c275260ea662c626ef1d75971533626a.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图4-4a ExecutorAsync特性：排队","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"2.批量","attrs":{}},{"type":"text","text":"，如图4-4b所示。每隔固定时间间隔，ExecutorAsync从队列中取出若干条请求，将其组装为批量请求bulk，并调用底层数据库API提交给服务端。如果队列长度增长得很快，我们也可以定义一个批量大小bulk size，当队列长度到达该值时立刻组装一个批量请求并提交。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/aa/aaf8563140b3b3be19e41f70c877bb4f.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图4-4b ExecutorAsync 特性：批量","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"3.并发","attrs":{}},{"type":"text","text":"，如图4-4c所示。如果底层数据库 API 支持异步提交请求，那么 ExecutorAsync 就可以充分利用这种特性，连续提交多个批量请求，而不需要等待之前的批量请求返回。为避免数据库服务器超载，我们可以定义并发度 parallelism，限制正在执行（in flight）的批量请求的数目；当达到限制时，如果调用者再提交新的请求，就暂存在队列 queue 中等待执行，而不会组装新的批量请求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/68/6821fb25a533eb75e1fc6c004768c0a1.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图4-4c ExecutorAsync 特性：并发","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"4.丢弃","attrs":{}},{"type":"text","text":"。如图4-4d所示。在上文提到的 bulk size 和 parallelism 的限制下，如果提交请求的速率远高于服务端响应的速率，那么大量请求就会堆积在队列中等待处理，最终导致超时失败。在这种情况下，将请求发送给服务端已经没有意义，因为调用者已经认定请求失败，而不再关心响应数据。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/e0/e04761d4dd1d2ad408cb98228b85b9ba.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图4-4d 请求超时","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因此，ExecutorAsync 应该及时从队列中移除无效请求，而剩余请求仍然“新鲜”。这种策略能够强制缩短队列长度，以降低后续请求在队列中的堆积时长、预防请求超时；同时，由于避免存储和发送无效请求，这种策略也能节约内存和 IO 开销。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/17/17c008ce182273eae10ec0b90ebcf66e.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图4-4e ExecutorAsync特性：丢弃","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}},{"type":"strong","attrs":{}}],"text":"批量请求的实现","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"上一小节我们看到了 ExecutorAsync 的调度策略，包括排队、批量、并发、丢弃。如下面的代码所示，ExecutorAsync 只需对外提供 submit(Request) 方法，用于提交单条请求。请求以数据对象 Request 表示，其字段 Request.resultPromise 是 Promise 对象，用于通知响应数据；在需要进行异常处理的场景下，我们使用 PromiseOrException作为 Promise 的实现，其中泛型模版T改为响应数据的实际类型。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"public class ExecutorAsync {\n\n public PromiseOrException submit(Request request) {\n return request.resultPromise;\n }\n}","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"接下来我们来看看 ExecutorAsync 的实现原理。由于源码细节较多、篇幅较长，故本节用流程图的形式，来讲解更高层的设计，如图4-5所示。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/cd/cd51a4373a87489cbbd5c2264d782528.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"图4-5 ExecutorAsync原理","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"1.提交请求","attrs":{}},{"type":"text","text":"。调用者调用 ExecutorAsync.submit(Request)，每次调用提交一条请求。该条请求存入队列 queue，等待后续调度执行。参数 Request 的结构如下面的代码所示，包括下列字段：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"predicate：函数，判断请求是否有效，无效请求（如超时的请求）将被丢弃。详见步骤2。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"resultPromise：通知响应数据。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"public class Request {\n public final PredicateE predicate; \n public final PromiseOrException resultPromise;\n}","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"2.每隔固定间隔，或者queue.size()达到bulk size，尝试组装批量请求","attrs":{}},{"type":"text","text":"。从队列queue中依次取出请求，每条请求执行函数 Request.predicate，以判断是否仍然要提交该请求；取出的有效请求的条数，不超过bulk size。predicate 是一个函数，类似于 jdk Predicate 接口，形式如下面的代码所示。接口函数test()可以正常返回，表示请求仍然有效；也可以抛出异常，说明请求无效的原因，如等待超时。如果抛出异常，则该条请求直接丢弃，并将发生的异常将通知给 Request.resultPromise，使得调用者执行异常处理逻辑。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"public interface PredicateE {\n void test() throws Exception;\n}","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"3.提交批量请求","attrs":{}},{"type":"text","text":"。第2步从队列 queue 中取出了至多 bulk size 条请求，将其作为参数调用RequestFunc.execute(requests)，以提交批量请求。接口 RequestFunc 的形式如下面的代码所示。接口方法execute(requests)以若干条请求为参数，将其组装为批量请求，调用底层的数据库 API 来提交。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"javascript"},"content":[{"type":"text","text":"public interface RequestFunc{\n void execute(List> requests);\n}","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"4.当收到响应后，对于每条请求，依次向 Request.resultPromise 通知响应数据","attrs":{}},{"type":"text","text":"。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"5.为防止服务端超载，ExecutorAsync 可限制并发请求数不超过 parallelism","attrs":{}},{"type":"text","text":"。我们设置计数变量 inFlight=0，以统计正在执行的批量请求数：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"a.当尝试组装批量请求（步骤2）时，首先判断inFlight