異步編程的幾種方式，你知道幾種？

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"近期嘗試在搬磚專用語言 Java 上實現異步，起因和過程就不再詳述了，總而言之，心中一萬頭草泥馬奔過。但這個過程也沒有白白浪費，趁機回顧了一下各種異步編程的實現。","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":"這篇文章會涉及到回調、Promise、反應式、async/await、用戶態線程等異步編程的實現方案。如果你熟悉它們中的一兩種，那應該也能很快理解其他幾個。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"爲什麼需要異步？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"操作系統可以看作是個虛擬機（VM），進程生活在操作系統創造的虛擬世界裏。進程不用知道到底有多少 core 多少內存，只要進程不要索取的太過分，操作系統就假裝有無限多的資源可用。","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","text":"基於這個思想，線程（Thread）的個數並不受硬件限制：你的程序可以只有一個線程、也可以有成百上千個。操作系統會默默做好調度，讓諸多線程共享有限的 CPU 時間片。這個調度的過程對線程是完全透明 的。","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":"那麼，操作系統是怎樣做到在線程無感知的情況下調度呢？答案是上下文切換（Context Switch） ，簡單來說，操作系統利用軟中斷機制，把程序從任意位置打斷，然後保存當前所有寄存器——包括最重要的指令寄存器 PC 和棧頂指針 SP，還有一些線程控制信息（TCB），整個過程會產生數個微秒的 overhead。","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/1f/1f0cb4d2b782c9ff8eae6309feb45f2f.webp","alt":"圖片","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},"content":[{"type":"text","text":"然而作爲一位合格的程序員，你一定也聽說過，線程是昂貴的：","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"線程的上下文切換有不少的代價，佔用寶貴的 CPU 時間；","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"每個線程都會佔用一些（至少 1 頁）內存。","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":"這兩個原因驅使我們儘可能避免創建太多的線程 ，而異步編程的目的就是消除 IO wait 阻塞——絕大多數時候，這是我們創建一堆線程、甚至引入線程池的罪魁禍首。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"Continuation","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"回調函數知道的人很多，但瞭解 Continuation 的人不多。Continuation 有時被晦澀地翻譯成“計算續體”，咱們還是直接用單詞好了。","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":"把一個計算過程在中間打斷，剩下的部分用一個對象表示，這就是 Continuation 。操作系統暫停一個線程時保存的那些現場數據，也可以看作一個 Continuation。有了它，我們就能在這個點接着剛剛的斷點繼續執行。","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":"打斷一個計算過程聽起來很厲害吧！實際上它每時每刻都在發生——假設函數 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"f()","attrs":{}}],"attrs":{}},{"type":"text","text":" 中間調用了 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"g()","attrs":{}}],"attrs":{}},{"type":"text","text":"，那 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"g()","attrs":{}}],"attrs":{}},{"type":"text","text":" 運行完成時，要返回到 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"f()","attrs":{}}],"attrs":{}},{"type":"text","text":" 剛剛調用 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"g()","attrs":{}}],"attrs":{}},{"type":"text","text":" 的地方接着執行。這個過程再自然不過了，以至於所有編程語言（彙編除外）都把它掩藏起來，讓你在編程中感覺不到調用棧的存在。","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/03/03e900c4fcaa1dbc8038780bf4c1208b.webp","alt":"圖片","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},"content":[{"type":"text","text":"操作系統用昂貴的軟中斷機制實現了棧的保存和恢復。那有沒有別的方式實現 Continuation 呢？最樸素的想法就是，把所有用得到的信息包成一個函數對象，在調用 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"g()","attrs":{}}],"attrs":{}},{"type":"text","text":" 的時候一起傳進去，並約定：一旦 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"g()","attrs":{}}],"attrs":{}},{"type":"text","text":" 完成，就拿着結果去調用這個 Continuation。","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":"這種編程模式被稱爲 Continuation-passing style（CPS）：","attrs":{}}]},{"type":"numberedlist","attrs":{"start":null,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"把調用者 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"f()","attrs":{}}],"attrs":{}},{"type":"text","text":" 還未執行的部分包成一個函數對象 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"cont","attrs":{}}],"attrs":{}},{"type":"text","text":"，一同傳給被調用者 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"g()","attrs":{}}],"attrs":{}},{"type":"text","text":"；","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"正常運行 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"g()","attrs":{}}],"attrs":{}},{"type":"text","text":" 函數體；","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"codeinline","content":[{"type":"text","text":"g()","attrs":{}}],"attrs":{}},{"type":"text","text":" 完成後，連同它的結果一起回調 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"cont","attrs":{}}],"attrs":{}},{"type":"text","text":"，從而繼續執行 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"f()","attrs":{}}],"attrs":{}},{"type":"text","text":" 裏剩餘的代碼。","attrs":{}}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"再拿 Wikipedia 上的定義鞏固一下：","attrs":{}}]},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"A function written in continuation-passing style takes an extra argument: an explicit \"continuation\", i.e. a function of one argument. When the CPS function has computed its result value, it \"returns\" it by calling the continuation function with this value as the argument.","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"CPS 風格的函數帶一個額外的參數：一個顯式的 Continuation，具體來說就是個僅有一個參數的函數。當 CPS 函數計算完返回值時，它“返回”的方式就是拿着返回值調用那個 Continuation。","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":"你應該已經發現了，這也就是回調函數，我只是換了個名字而已。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"異步的樸素實現：Callback","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"光有回調函數其實並沒有卵用。對於純粹的計算工作，Call Stack 就很好，爲何要費時費力用回調來做 Continuation 呢？你說的對，但僅限於沒有 IO 的情況。我們知道 IO 通常要比 CPU 慢上好幾個數量級，在 BIO 中，線程發起 IO 之後只能暫停，然後等待 IO 完成再由操作系統喚醒。","attrs":{}}]},{"type":"codeblock","attrs":{"lang":null},"content":[{"type":"text","text":"var input = recv_from_socket()  // Block at syscall recv()\n        var result = calculator.calculate(input)\n        send_to_socket(result) // Block at syscall send()\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":"而異步 IO 中，進程發起 IO 操作時也會一併輸入回調（也就是 Continuation），這大大解放了生產力——現場無需等待，可以立即返回去做其他事情。一旦 IO 成功後，AIO 的 Event Loop 會調用剛剛設置的回調函數，把剩下的工作完成。這種模式有時也被稱爲 Fire and Forget。","attrs":{}}]},{"type":"codeblock","attrs":{"lang":null},"content":[{"type":"text","text":"recv_from_socket((input) -> {\n        var result = calculator.calculate(input)\n        send_to_socket(result) // ignore result\n        })\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"就這麼簡單，通過我們自己實現的 Continuation，線程不再受 IO 阻塞，可以自由自在地跑滿 CPU。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"一顆語法糖：Promise","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"回調函數哪裏都好，就是不大好用，以及太醜了。","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":"第一個問題是可讀性大大下降，由於我們繞開操作系統自制 Continuation，所有函數調用都要傳入一個 lambda 表達式，你的代碼看起來就像要起飛一樣，縮進止不住地往右挪（the \"Callback Hell\"）。","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":"第二個問題是各種細節處理起來很麻煩，比如，考慮下異常處理，看來傳一個 Continuation 還不夠，最好再傳個異常處理的 callback。","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":"Promise 是對異步調用結果的一個封裝 ，在 Java 中它叫作 CompletableFuture (JDK8) 或者 ListenableFuture (Guava)。Promise 有兩層含義：","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":"第一層含義是：我現在還不是真正的結果，但是承諾以後會拿到這個結果 。這很容易理解，異步的任務遲早會完成，調用者如果比較蠢萌，他也可以用 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"Promise.get()","attrs":{}}],"attrs":{}},{"type":"text","text":" 強行要拿到結果，順便阻塞了當前線程，異步變成了同步。","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":"第二層含義是：如果你（調用者）有什麼吩咐，就告訴我好了 。這就有趣了，換句話說，回調函數不再是傳給 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"g()","attrs":{}}],"attrs":{}},{"type":"text","text":"，而是 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"g()","attrs":{}}],"attrs":{}},{"type":"text","text":" 返回的 Promise，比如之前那段代碼，我們用 Promise 來書寫，看起來順眼了不少。","attrs":{}}]},{"type":"codeblock","attrs":{"lang":null},"content":[{"type":"text","text":"var promise_input = recv_from_socket()\n        promise_input.then((input) -> {\n        var result = calculator.calculate(input)\n        send_to_socket(result) // ignore result\n        })\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Promise 改善了 Callback 的可讀性，也讓異常處理稍稍優雅了些，但終究是顆語法糖。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"反應式編程","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"反應式（Reactive）最早源於函數式編程中的一種模式，隨着微軟發起 ReactiveX 項目並一步步壯大，被移植到各種語言和平臺上。Reactive 最初在 GUI 編程中有廣泛的應用，由於異步調用的高性能，很快也在服務器後端領域遍地開花。","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":"Reactive 可以看作是對 Promise 的極大增強，相比 Promise，反應式引入了流（Flow）的概念。","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":"ReactiveX 中的事件流從一個 Observable 對象流出，這個對象可以是一個按鈕，也可以是 Restful API，總之，它能被外界觸發。與 Promise 不同的是，事件可能被觸發多次，所以處理代碼也會被多次調用。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一旦允許調用多次，從數據流動的角度看，事實上模型已經是 Push 而非 Pull 。那麼問題來了，如果調用頻率非常高，以至於我們處理速度跟不上了怎麼辦？所以 RX 框架又引入了 Backpressure 機制來進行流控，最簡單的流控方式就是：一旦 buffer 滿，就丟棄掉之後的事件。","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":"ReactiveX 框架的另一個優點是內置了很多好用的算子，比如：","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"merge","attrs":{}}],"attrs":{}},{"type":"text","text":"（Flow 合併），","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"debounce","attrs":{}}],"attrs":{}},{"type":"text","text":"（開關除顫）等等，方便了業務開發。下面是一個 RxJava 的例子：","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/98/98e761b6892e63aae472b4f331350534.gif","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"CPS 變換：Coroutine 與 async/await","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"無論是反應式還是 Promise，說到底仍然沒有擺脫手工構造 Continuation：開發者要把業務邏輯寫成回調函數。對於線性的邏輯基本可以應付自如，但是如果邏輯複雜一點呢？（比如，考慮下包含循環的情況）","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/d1/d16ec549d9ee8e25b6a8b6b76e3c74e1.webp","alt":"圖片","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},"content":[{"type":"text","text":"有些語言例如 C#，JavaScript 和 Python 提供了 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"async/await","attrs":{}}],"attrs":{}},{"type":"text","text":" 關鍵字。與 Reactive 一樣，這同樣出自微軟 C# 語言。在這些語言中，你會感到前所未有的爽感：異步編程終於擺脫了回調函數！唯一要做的只是在異步函數調用時加上 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"await","attrs":{}}],"attrs":{}},{"type":"text","text":"，編譯器就會自動把它轉化爲協程（Coroutine），而非昂貴的線程。","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":"魔法的背後是 CPS 變換，CPS 變換把普通函數轉換成一個 CPS 的函數，即 Continuation 也能作爲一個調用參數。函數不僅能從頭運行，還能根據 Continuation 的指示繼續某個點（比如調用 IO 的地方）運行。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"例子可以參見我的下一篇文章。由於代碼太長，就不貼在這兒了。","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":"可以看到，函數已經不再是一個函數了，而是變成一個狀態機 。每次 call 它、或者它 call 其他異步函數時，狀態機都會做一些計算和狀態輪轉。說好的 Continuation 在哪呢？就是對象自己（","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"this","attrs":{}}],"attrs":{}},{"type":"text","text":"）啊。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"CPS 變換實現非常複雜，尤其是考慮到 try-catch 之後。但是沒關係，複雜性都在編譯器裏，用戶只要學兩個關鍵詞即可。這個特性非常優雅，比 Java 那個廢柴的 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"CompletableFuture","attrs":{}}],"attrs":{}},{"type":"text","text":"不知道高到哪去了。（更新：也沒有那麼廢柴啦）","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":"JVM 上也有一個實現：electronicarts/ea-async，原理和 C# 的 async/await 類似，在編譯期修改 Bytecode 實現 CPS 變換。","attrs":{}}]}],"attrs":{}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"終極方案：用戶態線程","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"有了 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"async/await","attrs":{}}],"attrs":{}},{"type":"text","text":"，代碼已經簡潔很多了，基本上和同步代碼無異。是否有可能讓異步代碼和同步代碼完全一樣呢？聽起來就像免費午餐，但是的確可以做到！","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":"用戶態線程的代表是 Golang。JVM 上也有些實現，比如 Quasar，不過因爲 JDBC、Spring 這些周邊生態（它們佔據了大部分 IO 操作）的缺失基本沒有什麼用。","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","text":"用戶態線程是把操作系統提供的線程機制完全拋棄 ，換句話說，不去用這個 VM 的虛擬化機制。比如硬件有 8 個核心，那就創建 8 個系統線程，然後把 N 個用戶線程調度到這 8 個系統線程上跑。N 個用戶線程的調度在用戶進程裏實現，由於一切都在進程內部，切換代價要遠遠小於操作系統 Context Switch。","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/9c/9cc5aaa0453cb16757e617d740e51052.webp","alt":"圖片","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},"content":[{"type":"text","text":"另一方面，所有可能阻塞系統級線程的事情，例如 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"sleep()","attrs":{}}],"attrs":{}},{"type":"text","text":"、","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"recv()","attrs":{}}],"attrs":{}},{"type":"text","text":" 等，用戶態線程一定不能碰，否則它一旦阻塞住也就帶着那 8 個系統線程中的一個阻塞了。Go Runtime 接管了所有這樣的系統調用，並用一個統一的 Event loop 來輪詢和分發。","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":"另外，由於用戶態線程很輕量，我們完全沒必要再用線程池，如果需要開線程就直接創建。比如 Java 中的 WebServer 幾乎一定有個線程池，而 Go 可以給每個請求開闢一個 goroutine 去處理。併發編程從未如此美好！","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"總結","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"以上方案中，Promise、Reactive 本質上還是回調函數，只是框架的存在一定程度上降低了開發者的心智負擔。而 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"async/await","attrs":{}}],"attrs":{}},{"type":"text","text":" 和用戶態線程的解決方案要優雅和徹底的多，前者通過編譯期的 CPS 變換幫用戶創造出 CPS 式的函數調用；後者則繞開操作系統、重新實現一套線程機制，一切調度工作由 Runtime 接管。","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":"不知道是不是因爲歷史包袱太重，Java 語言本身提供的異步編程支持弱得可憐，即便是 CompletableFuture 還是在 Java 8 才引入，其後果就是很多庫都沒有異步的支持。雖然 Quasar 在沒有語言級支持的情況下引入了 CPS 變換，但是由於缺少周邊生態的支持，實際很難用在項目中。","attrs":{}}]}]}