Home  >  Article  >  PHP Framework  >  Introducing Process in Swoole

Introducing Process in Swoole

coldplay.xixi
coldplay.xixiforward
2021-03-01 10:13:232256browse

Introducing Process in Swoole

Recommended (free): swoole

Originally planned to develop the Process module in the swoft framework, so You need to have a deeper understanding of swoole's Process module. However, according to the practice process of swoole's official wiki, there are always parts that are not understood. Although I have done Multi-process programming many times before, but When you really need to develop a framework, you will find that the knowledge you have learned before is not comprehensive enough to guide the overall design. Fortunately, you have persisted and presented the current level of understanding.

Content overview:

  • Basic operations related to the process: fork/exit/kill/wait
  • Advanced operations related to the process: The main process exits and the child process also exits after finishing its work; the main process automatically restarts if the child process exits abnormally
  • Inter-process communication (IPC) - pipe (pipe)
  • Inter-process communication (IPC) - message queue (message queue)
  • More functions provided by the swoole process module

Basic operations related to processes

What is a process: A process is the runner’s program

Let’s look at it first Look at the simplest example:

<?phpecho posix_getpid(); // 获取当前进程的 pidswoole_set_process_name(&#39;swoole process master&#39;); // 修改所在进程的进程名sleep(100); // 模拟一个持续运行 100s 的程序, 这样就可以在进程中查看到它, 而不是运行完了就结束

View the process through ps aux:

The process name is not set

The process name is set

Let’s take a look at the basic operations of using subprocesses in swoole:

use Swoole\Process;

$process = new Process(function (Process $worker) {    if (Process::kill($worker->pid, 0)) { // kill操作常用来杀死进程, 传入 0 可以用来检测进程是否存在
        $worker->exit(); // 退出子进程
    }
});
$process->start(); // 启动子进程Process::wait(); // 回收退出的子进程
  • new Process(): Use the callback function to set the logic that the subprocess will execute

  • $process->start(): Call the fork() system call to generate a child process

  • Process::kill(): The kill operation sends a signal to the process, which is often used to kill the process. Passing in 0 can be used to detect whether the process exists

  • Process::wait(): Call the wait() system call to recycle the child process. If not recycled, the child process will be programmed zombie process, waste System resources

  • $worker->exit(): The child process actively exited

I have one here Question:

What is the life cycle of the main process? What is the life cycle of the child process?

This question also comes from my previous Thinking inertia: When understanding a thing, understand it from the life cycle of the thing. Combined with A process is a running program to understand together:

  • new Process(): Only the logic of the callback function will be executed in the process
  • Other code will be executed in the main process

Process Related advanced operations

  • The main process exits and the child process also exits after finishing its work
  • The child process exits abnormally and the main process automatically restarts
<?phpuse Swoole\Process;class MyProcess1{    public $mpid = 0; // master pid, 即当前程序的进程ID
    public $works = []; // 记录子进程的 pid
    public $maxProcessNum = 1;    public $newIndex = 0;    public function __construct()
    {        try {
            swoole_set_process_name(__CLASS__. &#39; : master&#39;);            $this->mpid = posix_getpid();            $this->run();            $this->processWait();
        } catch (\Exception $e) {            die(&#39;Error: &#39;. $e->getMessage());
        }
    }    public function run()
    {        for ($i=0; $i<$this->maxProcessNum; $i++) {            $this->createProcess();
        }
    }    public function createProcess($index = null)
    {        if (is_null($index)) {
            $index = $this->newIndex;            $this->newIndex++;
        }
        $process = new Process(function (Process $worker) use($index) { // 子进程创建后需要执行的函数
            swoole_set_process_name(__CLASS__. ": worker $index");            for ($j=0; $j<3; $j++) { // 模拟子进程执行耗时任务
                $this->checkMpid($worker);                echo "msg: {$j}\n";
                sleep(1);
            }
        }, false, false); // 不重定向输入输出; 不使用管道
        $pid = $process->start();        $this->works[$index] = $pid;        return $pid;
    }    // 主进程异常退出, 子进程工作完后退出
    public function checkMpid(Process $worker) // demo中使用的引用, 引用表示传的参数可以被改变, 由于传入 $worker 是 \Swoole\Process 对象, 所以不用使用 &    {        if (!Process::kill($this->mpid, 0)) { // 0 可以用来检测进程是否存在
            $worker->exit();
            $msg = "master process exited, worker {$worker->pid} also quit\n"; // 需要写入到日志中
            file_put_contents(&#39;process.log&#39;, $msg, FILE_APPEND); // todo: 这句话没有执行
        }
    }    // 重启子进程
    public function rebootProcess($pid)
    {
        $index = array_search($pid, $this->works);        if ($index !== false) {
            $newPid = $this->createProcess($index);            echo "rebootProcess: {$index}={$pid}->{$newPid} Done\n";            return;
        }        throw new \Exception("rebootProcess error: no pid {$pid}");
    }    // 自动重启子进程
    public function processWait()
    {        while (1) {            if (count($this->works)) {
                $ret = Process::wait(); // 子进程退出
                if ($ret) {                    $this->rebootProcess($ret[&#39;pid&#39;]);
                }
            } else {                break;
            }
        }
    }
}new MyProcess1();

Explain the following points:

  • The child process will exit after running. Through Process::wait(), if the child process exit signal is detected and an automatic restart is performed, the child process will Keep executing
  • About passing function parametersreference/pointer, a good way to understand is: The parameters can be modified

Run and simulate the main process to exit abnormally:

Simulate the main process to exit abnormally

Output

Inter-process Communication (IPC) - Pipe(pipe)

Several keywords for pipelines:

  • Half-duplex: One-way data flow, one end is read-only, one end is write-only.
  • Synchronous vs asynchronous: Default is Synchronous blocking mode, you can use swoole_event_add() to add a pipe to the swoole event loop to implement asynchronous IO
  • Pipe type (data format): SOCK_STREAM, streaming , users need to process the packetization/unpacking of data by themselves; SOCK_DGRAM, datagram, each sending and receiving is a complete data packet (DGRAM/STREAM)

Note, swoole wiki - process->write() mentioned that SOCK_DGRAM will not lose packets out of order

Let’s look at a simple example first, php pipe from shell Reading data from:

// get pip data$fp = fopen(&#39;php://stdin&#39;, &#39;r&#39;);if ($fp) {    while ($line = fgets($fp, 4096)) {        echo "php get pip data: ". $line;
    }
    fclose($fp);
}

Reading data from shell pipe

The pipe in swoole process is very powerful and supports sub-process writing, main process reading and The main process writes, the child process reads:

use Swoole\Process;// 子进程写, 父进程读$process = new Process(function (Process $worker) {
    $worker->write("worker");
});
$process->start();
$msg = $process->read();echo "from process: $msg", "\n";// 父进程写, 子进程读$process = new Process(function (Process $worker) {
    $msg = $worker->read();    echo "from master: $msg", "\n";
});
$process->start();
$process->write(&#39;master&#39;);

Use the pipe to read and write multiple times

Note the distinction between$worker->write() and $process->write(), I have mistakenly thought that these two are the same. In fact, I mistook $process for a child process, which is equivalent to $process- >write() means the sub-process writes the pipe--actually, this is the logic executed in the main process. The main process writes data to the pipe for the sub-process to read

Other pipe-related operations in swoole :

  • Asynchronous IO
use Swoole\Process;use Swoole\Event;// 异步IO$process = new Process(function (Process $worker) {
    $GLOBALS[&#39;worker&#39;] = $worker;
    Event::add($worker->pipe, function (int $pipe) { // 使用 swoole_event_add 添加管道到异步IO
        /** @var Process $worker */
        $worker = $GLOBALS[&#39;worker&#39;];
        $msg = $worker->read();        echo "from master: $msg \n";
        $worker->write("hello master");
        sleep(2);
        $worker->exit(0);
    });
});
$process->start();
$process->write("master msg 1");
$msg = $process->read();echo "from process: $msg \n";

Asynchronous IO

  • Set timeout
use Swoole\Process;// 设置管道超时$process = new Process(function (Process $worker) {
    sleep(5);
});
$process->start();
$process->setTimeout(0.5);
$ret = $process->read();
var_dump($ret);
var_dump(swoole_errno());

Pipeline timeout

插播一个趣事, @thinkpc 看完 2017北京PHP开发者年会, 就知道为啥会点赞了

  • 关闭管道
// 关闭管道: 默认值0->关闭读写 1->关闭写 2->关闭读$process->close();

进程间通信(IPC) - 消息队列(message queue)

消息队列:

  • 一系列保存在内核中的消息链表
  • 有一个 msgKey, 可以通过此访问不同的消息队列
  • 有数据大小限制, 默认 8192, 可以通过内核修改
  • 阻塞 vs 非阻塞: 阻塞模式下 pop()空消息队列/push()满消息队列会阻塞, 非阻塞模式可以直接返回

swoole 中使用消息队列:

  • 通信模式: 默认为争抢模式, 无法将消息投递给指定子进程
  • 新建消息队列后, 主进程就可以使用
  • 消息队列不可和管道一起使用, 也无法使用 swoole event loop
  • 主进程中要调用 wait(), 否则子进程中调用 pop()/push() 会报错
use Swoole\Process;
$process = new Process(function (Process $worker) {    // $worker->push(&#39;worker&#39;);
    echo "from master: ". $worker->pop(). "\n";
    sleep(2);    // $worker->exit();}, false, false); // 关闭管道// 参数一为 msgKey, 这里是默认值// 参数二为 通信模式, 默认值 2 表示争抢模式, 这里还加上了 非阻塞$process->useQueue(ftok(__FILE__, 1), 2| Process::IPC_NOWAIT);
$process->push(&#39;hello1&#39;); // 使用 useQueue 后, 主进程就可以读写消息队列了$process->push(&#39;hello2&#39;);echo "from woker: ". $process->pop(). "\n";// echo "from woker: ". $process->pop(). "\n";$process->start(); // 启动子进程// 消息队列状态var_dump($process->statQueue());// 删除队列, 如果不调用则不会在程序结束时清楚数据, 下次使用相同 msgKey 时还可以访问数据$process->freeQueue();
var_dump(Process::wait()); // 要调用 wait(), 否则子进程中 push()/pop() 会报错

消息队列

swoole process 模块提供的更多功能

  • swoole_set_process_name(): 修改进程名, 不兼容 mac

  • swoole_process->exec(string $execfile, array $args) 执行外部程序

参数 $execfile 需要使用可执行文件的绝对路径, 参数 args 为参数数组

// 比如 python test.py 123swoole_process->exec(&#39;/usr/bin/python&#39;, [&#39;test.py&#39;, 123]);// 更复杂的例子swoole_process->exec((&#39;/usr/local/bin/php&#39;, [&#39;/var/www/project/yii-best-practice/cli/yii&#39;, &#39;t/index&#39;, &#39;-m=123&#39;, &#39;abc&#39;, &#39;xyz&#39;]);// 父进程 exec 进程进行管道通信use Swoole\Process;
$process = new Process(function (Process $worker) {
    $worker->exec(&#39;/bin/echo&#39;, [&#39;hello&#39;]);
    $worker->write(&#39;hello&#39;);
}, true); // 需要启用标准输入输出重定向$process->start();echo "from exec: ". $process->read(). "\n";

父进程与exec进程通过管道通信

  • \Swoole\Process::kill($pid, $signo = SIGTERM): 向指定进程发送信号, 默认是终止进程, 传 0 可检测进程是否存在

  • \Swoole\Process::wait(): 回收子进程, 如果主进程不调用此方法, 子进程会变成 僵尸进程, 浪费系统资源

  • \Swoole\Process::signal(): 异步信号监听

use Swoole\Process;// 异步信号监听 + waitProcess::signal(SIGCHLD, function ($signal) { // 监听子进程退出信号
    // 可能同时有多个子进程退出, 所以要while循环
    while ($ret = Process::wait(false)) { // false 表示不阻塞
        var_dump($ret);
    }
});

\Swoole\Process::daemon(): 将当前进程变为一个守护进程

use Swoole\Process;// daemonProcess::daemon();
swoole_set_process_name(&#39;test daemon process&#39;);
sleep(100);

daemon-守护进程

  • \Swoole\Process::alarm(): 高精度定时器(微秒级), 对 setitimer 系统调用的封装, 可以配合 \Swoole\Process::signal() / pcntl_signal 使用

注意不可和 \Swoole\Timer 同时使用

// signal + alarm// 第一个参数表示时间, 单位 us, -1 表示清除定时器// 第二个参数表示类型 0->真实时间->SIGALAM 1->cpu时间->SIGVTALAM 2->用户态+内核态时间->SIGPROFProcess::alarm(100*1000); // 100msProcess::signal(SIGALRM, function ($signal) {    static $i = 0;    echo "#$i \t alarm \n";
    $i++;    if ($i>20) {
        Process::alarm(-1); // -1 表示清除
    }
});

alarm

  • \Swoole\Process::setaffinity(): 设置CPU亲和, 即将进程绑定到指定CPU核上

传值范围: [0, swoole_cpu_num())
CPU亲和: CPU的速度远远高于IO的速度, 所以CPU有多级缓存来解决IO等待的问题, 绑定指定CPU, 更容易命中CPU缓存

写在最后

资源推荐:

  • 图灵社区 - 理解UNIX进程 + 「理解Unix进程」读书笔记
  • blog - 「进程」编程

todo:

  • 使用输入输出重定向
  • 管道类型为 SOCK_STREAM 时的情况, 是否需要 封包/解包 处理, 即 swoole wiki - process->write() 中提到的 管道通信默认的方式是流式,write写入的数据在read可能会被底层合并
  • 多进程 + 异步IO 的注意事项

能理解 因为子进程会继承父进程的内存和IO句柄 这个会产生的影响, 但是给的示例并没有说明这个问题

use Swoole\Process;use Swoole\Event;// 多个子进程 + 异步IO$workers = [];
$workerNum = 3;for ($i=0; $i<$workerNum; $i++) {
    $process = new Process(function (Process $worker) {
        $worker->write($worker->pid);        echo "worker: {$worker->pid} \n";
    });
    $pid = $process->start();
    $workers[$pid] = $process;    // Event::add($process->pipe, function (int $pipe) use ($process) {
    //  $data = $process->read();
    //  echo "recv: $data \n";
    // });}foreach ($workers as $worker) {
    Event::add($worker->pipe, function (int $pipe) use ($worker) {
        $data = $worker->read();        echo "recv: $data \n";
    });
}

多进程异步IO

The above is the detailed content of Introducing Process in Swoole. For more information, please follow other related articles on the PHP Chinese website!

Statement:
This article is reproduced at:csdn.net. If there is any infringement, please contact admin@php.cn delete