| The child is killed after three iterations, so that’s when the “Child is still alive” messages end. One should probably point about that this is not a real parent/child relationship, because the child can continue running even after the parent finished. Or the child could kill the parent. One could modify the scheduler to have a more hierarchic task structure, but I won’t implement that in this article. There are many more process management calls one could implement, for examplewait(which waits until a task has finished running),exec(which replaces the current task) andfork(which creates a clone of the current task). Forking is pretty cool and you can actually implement it with PHP’s coroutines, because they support cloning. But I’ll leave these for the interested reader. Instead lets get to the next topic! | 译者信息 经过三次迭代以后子任务将被杀死，因此这就是"Child is still alive"消息结束的时候。可能应当指出的是这不是真正的父子关系。 因为甚至在父任务结束后子任务仍然可以运行。或者子任务可以杀死父任务。可以修改调度器使它具有更层级化的任务结构，不过 在这篇文章里我没有这么做。 你可以实现许多进程管理调用。例如 wait（它一直等待到任务结束运行时）,exec（它替代当前任务）和fork（它创建一个 当前任务的克隆）。fork非常酷，而且你可以使用PHP的协程真正地实现它，因为它们都支持克隆。 然而让我们把这些留给有兴趣的读者吧，我们去看下一个议题。 |
| Non-Blocking IO A really cool application of our task management system obviously is … a web server. There could be one task listening a socket for new connections and whenever a new connection is made it would create a new task handling that connection. The hard part about this is that normally socket operations like reading data are blocking, i.e. PHP will wait until the client has finished sending. For a web-server that’s obviously not good at all: It would mean that it can only handle a single connection at a time. The solution is to make sure that the socket is “ready” before actually reading/writing to it. To find out which sockets are ready to read from or write to the stream_select function can be used. | 译者信息 非阻塞IO 很明显，我们的任务管理系统的真正很酷的应用是web服务器。它有一个任务是在套接字上侦听是否有新连接，当有新连接要建立的时候 ，它创建一个新任务来处理新连接。
web服务器最难的部分通常是像读数据这样的套接字操作是阻塞的。例如PHP将等待到客户端完成发送为止。对一个WEB服务器来说，这 根本不行；这就意味着服务器在一个时间点上只能处理一个连接。
解决方案是确保在真正对套接字读写之前该套接字已经“准备就绪”。为了查找哪个套接字已经准备好读或者写了，可以使用 流选择函数。 |
| First, lets add two new syscalls, which will cause a task to wait until a certain socket is ready: <?php
function waitForRead($socket) {
return new SystemCall(
function(Task $task, Scheduler $scheduler) use ($socket) {
$scheduler->waitForRead($socket, $task);
}
);
}
function waitForWrite($socket) {
return new SystemCall(
function(Task $task, Scheduler $scheduler) use ($socket) {
$scheduler->waitForWrite($socket, $task);
}
);
}
These syscalls are just proxies to the respective methods in the scheduler: <?php
// resourceID => [socket, tasks]
protected $waitingForRead = [];
protected $waitingForWrite = [];
public function waitForRead($socket, Task $task) {
if (isset($this->waitingForRead[(int) $socket])) {
$this->waitingForRead[(int) $socket][1][] = $task;
} else {
$this->waitingForRead[(int) $socket] = [$socket, [$task]];
}
}
public function waitForWrite($socket, Task $task) {
if (isset($this->waitingForWrite[(int) $socket])) {
$this->waitingForWrite[(int) $socket][1][] = $task;
} else {
$this->waitingForWrite[(int) $socket] = [$socket, [$task]];
}
}
The waitingForRead and waitingForWriteproperties are just arrays containing the sockets to wait for and the tasks that are waiting for them. The interesting part is the following method, which actually checks whether the sockets are ready and reschedules the respective tasks: <?php
protected function ioPoll($timeout) {
$rSocks = [];
foreach ($this->waitingForRead as list($socket)) {
$rSocks[] = $socket;
}
$wSocks = [];
foreach ($this->waitingForWrite as list($socket)) {
$wSocks[] = $socket;
}
$eSocks = []; // dummy
if (!stream_select($rSocks, $wSocks, $eSocks, $timeout)) {
return;
}
foreach ($rSocks as $socket) {
list(, $tasks) = $this->waitingForRead[(int) $socket];
unset($this->waitingForRead[(int) $socket]);
foreach ($tasks as $task) {
$this->schedule($task);
}
}
foreach ($wSocks as $socket) {
list(, $tasks) = $this->waitingForWrite[(int) $socket];
unset($this->waitingForWrite[(int) $socket]);
foreach ($tasks as $task) {
$this->schedule($task);
}
}
}
Thestream_selectfunction takes arrays of read, write and except sockets to check (we’ll ignore that last category). The arrays are passed by reference and the function will only leave those elements in the arrays that changed state. We can then walk over those arrays and reschedule all tasks associated with them. | 译者信息 首先，让我们添加两个新的 syscall，它们将等待直到指定 socket 准备好： <?php
function waitForRead($socket) {
return new SystemCall(
function(Task $task, Scheduler $scheduler) use ($socket) {
$scheduler->waitForRead($socket, $task);
}
);
}
function waitForWrite($socket) {
return new SystemCall(
function(Task $task, Scheduler $scheduler) use ($socket) {
$scheduler->waitForWrite($socket, $task);
}
);
}
这些 syscall 只是在调度器中代理其各自的方法： <?php
// resourceID => [socket, tasks]
protected $waitingForRead = [];
protected $waitingForWrite = [];
public function waitForRead($socket, Task $task) {
if (isset($this->waitingForRead[(int) $socket])) {
$this->waitingForRead[(int) $socket][1][] = $task;
} else {
$this->waitingForRead[(int) $socket] = [$socket, [$task]];
}
}
public function waitForWrite($socket, Task $task) {
if (isset($this->waitingForWrite[(int) $socket])) {
$this->waitingForWrite[(int) $socket][1][] = $task;
} else {
$this->waitingForWrite[(int) $socket] = [$socket, [$task]];
}
}
waitingForRead 及 waitingForWrite 属性是两个承载等待的socket 及等待它们的任务的数组。有趣的部分在于下面的方法，它将检查 socket 是否可用，并重新安排各自任务： <?php
protected function ioPoll($timeout) {
$rSocks = [];
foreach ($this->waitingForRead as list($socket)) {
$rSocks[] = $socket;
}
$wSocks = [];
foreach ($this->waitingForWrite as list($socket)) {
$wSocks[] = $socket;
}
$eSocks = []; // dummy
if (!stream_select($rSocks, $wSocks, $eSocks, $timeout)) {
return;
}
foreach ($rSocks as $socket) {
list(, $tasks) = $this->waitingForRead[(int) $socket];
unset($this->waitingForRead[(int) $socket]);
foreach ($tasks as $task) {
$this->schedule($task);
}
}
foreach ($wSocks as $socket) {
list(, $tasks) = $this->waitingForWrite[(int) $socket];
unset($this->waitingForWrite[(int) $socket]);
foreach ($tasks as $task) {
$this->schedule($task);
}
}
}
stream_select 函数接受承载读取、写入以及待检查的socket的数组（我们无需考虑最后一类）。数组将按引用传递，函数只会保留那些状态改变了的数组元素。我们可以遍历这些数组，并重新安排与之相关的任务。 |
| In order to regularly perform the above polling action we’ll add a special task in the scheduler: <?php
protected function ioPollTask() {
while (true) {
if ($this->taskQueue->isEmpty()) {
$this->ioPoll(null);
} else {
$this->ioPoll(0);
}
yield;
}
}
This task needs to be registered at some point, e.g. one could add$this->newTask($this->ioPollTask())to the start of therun()method. Then it will work just like any other task, performing the polling operation once every full task cycle (this isn’t necessarily the best way to handle it). TheioPollTaskwill callioPollwith a0second timeout, which means thatstream_selectwill return right away (rather than waiting). Only if the task queue is empty we use anulltimeout, which means that it will wait until some socket becomes ready. If we wouldn’t do this the polling task would just run again and again and again until a new connection is made. This would result in 100% CPU usage. It’s much more efficient to let the operating system do the waiting instead. | 译者信息 为了正常地执行上面的轮询动作，我们将在调度器里增加一个特殊的任务： <?php
protected function ioPollTask() {
while (true) {
if ($this->taskQueue->isEmpty()) {
$this->ioPoll(null);
} else {
$this->ioPoll(0);
}
yield;
}
}
需要在某个地方注册这个任务，例如，你可以在run()方法的开始增加$this->newTask($this->ioPollTask())。然后就像其他 任务一样每执行完整任务循环一次就执行轮询操作一次（这么做一定不是最好的方法）。ioPollTask将使用0秒的超时来调用ioPoll， 这意味着stream_select将立即返回（而不是等待）。 只有任务队列为空时，我们才使用null超时，这意味着它一直等到某个套接口准备就绪。如果我们没有这么做，那么轮询任务将一而再， 再而三的循环运行，直到有新的连接建立。这将导致100%的CPU利用率。相反，让操作系统做这种等待会更有效。 |
| Writing the server is relatively easy now: <?php
function server($port) {
echo "Starting server at port $port...\n";
$socket = @stream_socket_server("tcp://localhost:$port", $errNo, $errStr);
if (!$socket) throw new Exception($errStr, $errNo);
stream_set_blocking($socket, 0);
while (true) {
yield waitForRead($socket);
$clientSocket = stream_socket_accept($socket, 0);
yield newTask(handleClient($clientSocket));
}
}
function handleClient($socket) {
yield waitForRead($socket);
$data = fread($socket, 8192);
$msg = "Received following request:\n\n$data";
$msgLength = strlen($msg);
$response = <<<RES
HTTP/1.1 200 OK\r
Content-Type: text/plain\r
Content-Length: $msgLength\r
Connection: close\r
\r
$msg
RES;
yield waitForWrite($socket);
fwrite($socket, $response);
fclose($socket);
}
$scheduler = new Scheduler;
$scheduler->newTask(server(8000));
$scheduler->run();
This will accept connections tolocalhost:8000and just send back a HTTP response with whatever it was sent. Doing anything “real” would be a lot more complicated (properly handling HTTP requests is way outside the scope of this article). The above snippet just demos the general concept. You can try the server out using something likeab -n 10000 -c 100 localhost:8000/. This will send 10000 requests to it with 100 of them arriving concurrently. Using these numbers I get a median response time of 10ms. But there is an issue with a few requests being handled really slowly (like 5 seconds), that’s why the total throughput is only 2000 reqs/s (with a 10ms response time it should be more like 10000 reqs/s). With higher concurrency count (e.g.-c 500) it mostly still works well, but some connections will throw a “Connection reset by peer” error. As I know very little about this low-level socket stuff I didn’t try to figure out what the issue is. | 译者信息现在编写服务器相对容易了：
<?php
function server($port) {
echo "Starting server at port $port...\n";
$socket = @stream_socket_server("tcp://localhost:$port", $errNo, $errStr);
if (!$socket) throw new Exception($errStr, $errNo);
stream_set_blocking($socket, 0);
while (true) {
yield waitForRead($socket);
$clientSocket = stream_socket_accept($socket, 0);
yield newTask(handleClient($clientSocket));
}
}
function handleClient($socket) {
yield waitForRead($socket);
$data = fread($socket, 8192);
$msg = "Received following request:\n\n$data";
$msgLength = strlen($msg);
$response = <<<RES
HTTP/1.1 200 OK\r
Content-Type: text/plain\r
Content-Length: $msgLength\r
Connection: close\r
\r
$msg
RES;
yield waitForWrite($socket);
fwrite($socket, $response);
fclose($socket);
}
$scheduler = new Scheduler;
$scheduler->newTask(server(8000));
$scheduler->run();
这段代码将接收到localhost:8000上的连接，然后仅仅返回发送来的内容作为HTTP响应。要做“实际”的事情的话就爱哪个非常复杂（处理 HTTP请求可能已经超出了这篇文章的范围）。上面的代码片段只是演示了一般性的概念。
你可以使用类似于ab -n 10000 -c 100 localhost:8000/这样命令来测试服务器。这条命令将向服务器发送10000个请求，并且其中100个请求将同时到达。使用这样的数目，我得到了处于中间的10毫秒的响应时间。不过还有一个问题：有少数几个请求真正处理的很慢（如5秒）， 这就是为什么总吞吐量只有2000请求/秒（如果是10毫秒的响应时间的话，总的吞吐量应该更像是10000请求/秒）。调高并发数（比如 -c 500),服务器大多数运行良好，不过某些连接将抛出“连接被对方重置”的错误。由于我对低级别的socket资料了解的非常少，所以 我不能指出问题出在哪儿。
|
| Stacked coroutines If you would try to build some larger system using our scheduling system you would soon run into a problem: We are used to breaking up code into smaller functions and calling them. But with coroutines this is no longer possible. E.g. consider the following code: <?php
function echoTimes($msg, $max) {
for ($i = 1; $i <= $max; ++$i) {
echo "$msg iteration $i\n";
yield;
}
}
function task() {
echoTimes('foo', 10); // print foo ten times
echo "---\n";
echoTimes('bar', 5); // print bar five times
yield; // force it to be a coroutine
}
$scheduler = new Scheduler;
$scheduler->newTask(task());
$scheduler->run();
This code tries to put the recurring “output n times” code into a separate coroutine and then invoke it from the main task. But this won’t work. As mentioned at the very beginning of this article calling a generator (or coroutine) will not actually do anything, it will only return an object. This also happens in the above case. TheechoTimescalls won’t do anything than return an (unused) coroutine object. | 译者信息 协程堆栈 如果你试图用我们的调度系统建立更大的系统的话，你将很快遇到问题：我们习惯了把代码分解为更小的函数，然后调用它们。然而， 如果使用了协程的话，就不能这么做了。例如，看下面代码： <?php
function echoTimes($msg, $max) {
for ($i = 1; $i <= $max; ++$i) {
echo "$msg iteration $i\n";
yield;
}
}
function task() {
echoTimes('foo', 10); // print foo ten times
echo "---\n";
echoTimes('bar', 5); // print bar five times
yield; // force it to be a coroutine
}
$scheduler = new Scheduler;
$scheduler->newTask(task());
$scheduler->run();
这段代码试图把重复循环“输出n次“的代码嵌入到一个独立的协程里，然后从主任务里调用它。然而它无法运行。正如在这篇文章的开始 所提到的，调用生成器（或者协程）将没有真正地做任何事情，它仅仅返回一个对象。这也出现在上面的例子里。echoTimes调用除了放回一个（无用的）协程对象外不做任何事情。 |
| In order to still allow this we need to write a small wrapper around the bare coroutines. I’ll call this a “stacked coroutine” because it will manage a stack of nested coroutine calls. It will be possible to call sub-coroutines by yielding them: $retval = (yield someCoroutine($foo, $bar));
The subcoroutines will also be able to return a value, again by usingyield: yield retval("I'm a return value!");
Theretvalfunction does nothing more than returning a wrapper around the value which will signal that it’s a return value: <?php
class CoroutineReturnValue {
protected $value;
public function __construct($value) {
$this->value = $value;
}
public function getValue() {
return $this->value;
}
}
function retval($value) {
return new CoroutineReturnValue($value);
}
In order to turn a coroutine into a stacked coroutine (which supports subcalls) we’ll have to write another function (which is obviously yet-another-coroutine): <?php
function stackedCoroutine(Generator $gen) {
$stack = new SplStack;
for (;;) {
$value = $gen->current();
if ($value instanceof Generator) {
$stack->push($gen);
$gen = $value;
continue;
}
$isReturnValue = $value instanceof CoroutineReturnValue;
if (!$gen->valid() || $isReturnValue) {
if ($stack->isEmpty()) {
return;
}
$gen = $stack->pop();
$gen->send($isReturnValue ? $value->getValue() : NULL);
continue;
}
$gen->send(yield $gen->key() => $value);
}
}
This function acts as a simple proxy between the caller and the currently running subcoroutine. This is handled in the$gen->send(yield $gen->key() => $value);line. Additionally it checks whether a return value is a generator, in which case it will start running it and pushes the previous coroutine on the stack. Once it gets aCoroutineReturnValueit will pop the stack again and continue executing the previous coroutine. | 译者信息为了仍然允许这么做，我们需要在这个裸协程上写一个小小的封装。我们将调用它：“协程堆栈”。因为它将管理嵌套的协程调用堆栈。 这将是通过生成协程来调用子协程成为可能：
$retval = (yield someCoroutine($foo, $bar));
使用yield，子协程也能再次返回值：
yield retval("I'm a return value!");
retval函数除了返回一个值的封装外没有做任何其他事情。这个封装将表示它是一个返回值。
<?php
class CoroutineReturnValue {
protected $value;
public function __construct($value) {
$this->value = $value;
}
public function getValue() {
return $this->value;
}
}
function retval($value) {
return new CoroutineReturnValue($value);
}
为了把协程转变为协程堆栈（它支持子调用），我们将不得不编写另外一个函数（很明显，它是另一个协程）： <?php
function stackedCoroutine(Generator $gen) {
$stack = new SplStack;
for (;;) {
$value = $gen->current();
if ($value instanceof Generator) {
$stack->push($gen);
$gen = $value;
continue;
}
$isReturnValue = $value instanceof CoroutineReturnValue;
if (!$gen->valid() || $isReturnValue) {
if ($stack->isEmpty()) {
return;
}
$gen = $stack->pop();
$gen->send($isReturnValue ? $value->getValue() : NULL);
continue;
}
$gen->send(yield $gen->key() => $value);
}
}
这个函数在调用者和当前正在运行的子协程之间扮演着简单代理的角色。在$gen->send(yield $gen->key()=>$value)；这行完成了代理功能。另外它检查返回值是否是生成器，万一是生成器的话，它将开始运行这个生成器，并把前一个协程压入堆栈里。一旦它获得了CoroutineReturnValue的话，它将再次请求堆栈弹出，然后继续执行前一个协程。 |
| In order to make the stacked coroutines usable in tasks the$this->coroutine = $coroutine;line in theTaskconstructor needs to be replaced with$this->coroutine = stackedCoroutine($coroutine);. Now we can improve the webserver example from above a bit by grouping the wait+read (and wait+write and wait+accept) actions into functions. To group the related functionality I’ll use a class: <?php
class CoSocket {
protected $socket;
public function __construct($socket) {
$this->socket = $socket;
}
public function accept() {
yield waitForRead($this->socket);
yield retval(new CoSocket(stream_socket_accept($this->socket, 0)));
}
public function read($size) {
yield waitForRead($this->socket);
yield retval(fread($this->socket, $size));
}
public function write($string) {
yield waitForWrite($this->socket);
fwrite($this->socket, $string);
}
public function close() {
@fclose($this->socket);
}
}
Now the server can be rewritten a bit cleaner: <?php
function server($port) {
echo "Starting server at port $port...\n";
$socket = @stream_socket_server("tcp://localhost:$port", $errNo, $errStr);
if (!$socket) throw new Exception($errStr, $errNo);
stream_set_blocking($socket, 0);
$socket = new CoSocket($socket);
while (true) {
yield newTask(
handleClient(yield $socket->accept())
);
}
}
function handleClient($socket) {
$data = (yield $socket->read(8192));
$msg = "Received following request:\n\n$data";
$msgLength = strlen($msg);
$response = <<<RES
HTTP/1.1 200 OK\r
Content-Type: text/plain\r
Content-Length: $msgLength\r
Connection: close\r
\r
$msg
RES;
yield $socket->write($response);
yield $socket->close();
}
| 译者信息 为了使协程堆栈在任务里可用，任务构造器里的$this-coroutine =$coroutine;这行需要替代为$this->coroutine = StackedCoroutine($coroutine);。 现在我们可以稍微改进上面web服务器例子：把wait+read(和wait+write和warit+accept)这样的动作分组为函数。为了分组相关的 功能，我将使用下面类： <?php
class CoSocket {
protected $socket;
public function __construct($socket) {
$this->socket = $socket;
}
public function accept() {
yield waitForRead($this->socket);
yield retval(new CoSocket(stream_socket_accept($this->socket, 0)));
}
public function read($size) {
yield waitForRead($this->socket);
yield retval(fread($this->socket, $size));
}
public function write($string) {
yield waitForWrite($this->socket);
fwrite($this->socket, $string);
}
public function close() {
@fclose($this->socket);
}
}
现在服务器可以编写的稍微简洁点了： <?php
function server($port) {
echo "Starting server at port $port...\n";
$socket = @stream_socket_server("tcp://localhost:$port", $errNo, $errStr);
if (!$socket) throw new Exception($errStr, $errNo);
stream_set_blocking($socket, 0);
$socket = new CoSocket($socket);
while (true) {
yield newTask(
handleClient(yield $socket->accept())
);
}
}
function handleClient($socket) {
$data = (yield $socket->read(8192));
$msg = "Received following request:\n\n$data";
$msgLength = strlen($msg);
$response = <<<RES
HTTP/1.1 200 OK\r
Content-Type: text/plain\r
Content-Length: $msgLength\r
Connection: close\r
\r
$msg
RES;
yield $socket->write($response);
yield $socket->close();
}
|
| Error handling As a good programmer you obviously noticed that the above examples all lack error handling. Pretty much every socket operation is fallible and can produce errors. I obviously did this because error handling is really tedious (especially for sockets!) and would easily blow up the code size by a few factors. But still I’d like to cover how error handling for coroutines works in general: Coroutines provide the ability to throw exceptions inside them using thethrow()method. As of this writing this method does not yet exist in PHP’s implementation, but I will commit it later today. Thethrow()method takes an exception and throws it at the current suspension point in the coroutine. Consider this code: <?php
function gen() {
echo "Foo\n";
try {
yield;
} catch (Exception $e) {
echo "Exception: {$e->getMessage()}\n";
}
echo "Bar\n";
}
$gen = gen();
$gen->rewind(); // echos "Foo"
$gen->throw(new Exception('Test')); // echos "Exception: Test"
// and "Bar"
This is really awesome for our purposes, because we can make system calls and subcoroutine calls throw exceptions. For the system calls theScheduler::run()method needs a small adjustment: <?php
if ($retval instanceof SystemCall) {
try {
$retval($task, $this);
} catch (Exception $e) {
$task->setException($e);
$this->schedule($task);
}
continue;
}
And theTaskclass needs to handlethrowcalls too: <?php
class Task {
// ...
protected $exception = null;
public function setException($exception) {
$this->exception = $exception;
}
public function run() {
if ($this->beforeFirstYield) {
$this->beforeFirstYield = false;
return $this->coroutine->current();
} elseif ($this->exception) {
$retval = $this->coroutine->throw($this->exception);
$this->exception = null;
return $retval;
} else {
$retval = $this->coroutine->send($this->sendValue);
$this->sendValue = null;
return $retval;
}
}
// ...
}
Now we can start throwing exceptions from system calls! E.g. for thekillTaskcall, lets throw an exception if the passed task ID is invalid: <?php
function killTask($tid) {
return new SystemCall(
function(Task $task, Scheduler $scheduler) use ($tid) {
if ($scheduler->killTask($tid)) {
$scheduler->schedule($task);
} else {
throw new InvalidArgumentException('Invalid task ID!');
}
}
);
}
Try it out: <?php
function task() {
try {
yield killTask(500);
} catch (Exception $e) {
echo 'Tried to kill task 500 but failed: ', $e->getMessage(), "\n";
}
}
Sadly this won’t work properly yet, because thestackedCoroutinefunction doesn’t handle the exception correctly. To fix it the function needs some modifications: <?php
function stackedCoroutine(Generator $gen) {
$stack = new SplStack;
$exception = null;
for (;;) {
try {
if ($exception) {
$gen->throw($exception);
$exception = null;
continue;
}
$value = $gen->current();
if ($value instanceof Generator) {
$stack->push($gen);
$gen = $value;
continue;
}
$isReturnValue = $value instanceof CoroutineReturnValue;
if (!$gen->valid() || $isReturnValue) {
if ($stack->isEmpty()) {
return;
}
$gen = $stack->pop();
$gen->send($isReturnValue ? $value->getValue() : NULL);
continue;
}
try {
$sendValue = (yield $gen->key() => $value);
} catch (Exception $e) {
$gen->throw($e);
continue;
}
$gen->send($sendValue);
} catch (Exception $e) {
if ($stack->isEmpty()) {
throw $e;
}
$gen = $stack->pop();
$exception = $e;
}
}
}
| 译者信息 错误处理 作为一个优秀的程序员，相信你已经察觉到上面的例子缺少错误处理。几乎所有的 socket 都是易出错的。我这样做的原因一方面固然是因为错误处理的乏味（特别是 socket！），另一方面也在于它很容易使代码体积膨胀。 不过，我仍然了一讲一下常见的协程错误处理：协程允许使用 throw() 方法在其内部抛出一个错误。尽管此方法还未在 PHP 中实现，但我很快就会提交它，就在今天。 throw() 方法接受一个 Exception，并将其抛出到协程的当前悬挂点，看看下面代码： <?php
function gen() {
echo "Foo\n";
try {
yield;
} catch (Exception $e) {
echo "Exception: {$e->getMessage()}\n";
}
echo "Bar\n";
}
$gen = gen();
$gen->rewind(); // echos "Foo"
$gen->throw(new Exception('Test')); // echos "Exception: Test"
// and "Bar"
这非常棒，因为我们可以使用系统调用以及子协程调用异常抛出。对与系统调用，Scheduler::run() 方法需要一些小调整： <?php
if ($retval instanceof SystemCall) {
try {
$retval($task, $this);
} catch (Exception $e) {
$task->setException($e);
$this->schedule($task);
}
continue;
}
Task 类也许要添加 throw 调用处理： <?php
class Task {
// ...
protected $exception = null;
public function setException($exception) {
$this->exception = $exception;
}
public function run() {
if ($this->beforeFirstYield) {
$this->beforeFirstYield = false;
return $this->coroutine->current();
} elseif ($this->exception) {
$retval = $this->coroutine->throw($this->exception);
$this->exception = null;
return $retval;
} else {
$retval = $this->coroutine->send($this->sendValue);
$this->sendValue = null;
return $retval;
}
}
// ...
}
现在，我们已经可以在系统调用中使用异常抛出了！例如，要调用 killTask，让我们在传递 ID 不可用时抛出一个异常： <?php
function killTask($tid) {
return new SystemCall(
function(Task $task, Scheduler $scheduler) use ($tid) {
if ($scheduler->killTask($tid)) {
$scheduler->schedule($task);
} else {
throw new InvalidArgumentException('Invalid task ID!');
}
}
);
}
试试看： <?php
function task() {
try {
yield killTask(500);
} catch (Exception $e) {
echo 'Tried to kill task 500 but failed: ', $e->getMessage(), "\n";
}
}
这些代码现在尚不能正常运作，因为 stackedCoroutine 函数无法正确处理异常。要修复需要做些调整： <?php
function stackedCoroutine(Generator $gen) {
$stack = new SplStack;
$exception = null;
for (;;) {
try {
if ($exception) {
$gen->throw($exception);
$exception = null;
continue;
}
$value = $gen->current();
if ($value instanceof Generator) {
$stack->push($gen);
$gen = $value;
continue;
}
$isReturnValue = $value instanceof CoroutineReturnValue;
if (!$gen->valid() || $isReturnValue) {
if ($stack->isEmpty()) {
return;
}
$gen = $stack->pop();
$gen->send($isReturnValue ? $value->getValue() : NULL);
continue;
}
try {
$sendValue = (yield $gen->key() => $value);
} catch (Exception $e) {
$gen->throw($e);
continue;
}
$gen->send($sendValue);
} catch (Exception $e) {
if ($stack->isEmpty()) {
throw $e;
}
$gen = $stack->pop();
$exception = $e;
}
}
}
|
| Wrapping up In this article we built a task scheduler using cooperative multitasking, including the ability to perform “system calls”, doing non-blocking IO operations and handling errors. The really cool thing about all this is that the resulting code for the tasks looks totally synchronous, even though it is performing a lot of asynchronous operations. If you want to read data from a socket you don’t have to pass some callback or register an event listener. Instead you writeyield $socket->read(). Which is basically what you would normally do too, just with ayieldin front of it. When I first heard about all this I found this concept totally awesome and that’s what motivated me to implement it in PHP. At the same time I find coroutines really scary. There is a thin line between awesome code and a total mess and I think coroutines sit exactly on that line. It’s hard for me to say whether writing async code in the way outlined above is really beneficial. In any case, I think it’s an interesting topic and I hope you found it interesting too. Comments welcome :) | 译者信息 结束语 在这篇文章里，我使用多任务协作构建了一个任务调度器，其中包括执行“系统调用”，做非阻塞操作和处理错误。所有这些里真正很酷的事情是任务的结果代码看起来完全同步，甚至任务正在执行大量的异步操作的时候也是这样。如果你打算从套接口读取数据的话，你将不需要传递某个回调函数或者注册一个事件侦听器。相反，你只要书写yield $socket->read()。这儿大部分都是你常常也要编写的，只在它的前面增加yield。 当我第一次听到所有这一切的时候，我发现这个概念完全令人折服，而且正是这个激励我在PHP中实现了它。同时我发现协程真正令人心慌。在令人敬畏的代码和很大一堆代码之间只有单薄的一行，我认为协程正好处在这一行上。讲讲使用上面所述的方法书写异步代码是否真的有益对我来说很难。 无论如何，我认为这是一个有趣的话题，而且我希望你也能找到它的乐趣。欢迎评论:) |
5109
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
