在我们日常编程中，常见的例子比如 php think 需要不断执行的任务，比如 php arts… 和 php yii…，都会通过 nohup 挂载到后台，以保持长期运行状态。同样，在workerman中，使用类似php index.php start的命令来启动进程，但不同的是，它不需要使用nohup来挂载并在后台运行。有的朋友可能会好奇是怎么实现的呢？为了解决小伙伴们关心的问题，今天我们重点深入分析workerman守护进程的实现原理。
我们先来了解一些流程相关的知识：
父进程：父进程是生成其他进程的进程。当一个进程创建另一个进程时，创建者称为父进程，创建的进程成为子进程。父进程可以通过进程标识符（pid）来识别它创建的子进程。
子进程：子进程是父进程创建的新进程。子进程继承了父进程的一些属性，比如环境变量、文件描述符等。子进程独立于父进程运行，可以执行自己的代码，拥有自己的资源和内存空间。
进程组：进程组是关联进程的集合。每个进程组都有一个唯一的进程组 id (pgid)，用于标识该进程组。进程组通常由父进程创建，并包括与父进程具有相同会话 id (sid) 的所有子进程。
会话：会话是相关进程的集合，通常在用户登录系统时开始，在用户注销或关闭终端时结束。会话中的进程共享同一个控制终端。每个会话都有一个唯一的会话 id (sid)，用于标识该会话。一个会话通常包含一个或多个进程组，第一个进程组成为会话的主进程组。
这些概念俗称八足文，从来都不容易理解。让我们看一个例子。执行命令“php index.html”后php’，生成了进程61052。该进程的父进程是bash进程8243，所以这里无需担心。然后，通过fork创建了一个子进程61053，其父进程为61052。这两个进程有共同的进程组61052和会话8243。调用posix_setsid函数会为子进程打开一个新的进程组61053和一个新的会话61053进程61053，其中的会话可以理解为新的命令窗口终端。最后，子进程61053通过fork创建了子进程61054，进程61053升级为父进程。再次fork的原因是为了避免与终端控制的进程相关联。这个进程61052是在终端模式下创建的，从此以后，进程61054就形成了守护进程。
[ manongsen@root phpwork]$ php index.php [parent] process id: 61052, parent process id: 8243, process group id: 61052, session id: 8243 [parent1] process id: 61052, parent process id: 8243, process group id: 61052, session id: 8243 exited the process [child1] process id: 61053, parent process id: 61052, process group id: 61052, session id: 8243 [child1] process id: 61053, parent process id: 61052, process group id: 61053, session id: 61053 [parent2] process id: 61053, parent process id: 61052, process group id: 61053, session id: 61053 exited the process [child2] process id: 61054, parent process id: 61053, process group id: 61053, session id: 61053. this process is retained [ manongsen@root phpwork]$ ps aux | grep index.php root 66064 0.0 0.0 408105040 1472 s080 s+10:00 pm 00:00 grep index.php root 61054 0.0 438073488 280?? s 10:00 pm 00:00 php index.php
上面提到的进程信息就是通过这段代码的执行而生成的。如果你仔细阅读这段代码，你会发现为什么 posix_setsid 函数不是在第一个 fork 之前调用，而是在第二个 fork 之前调用。这样就不用fork两次了？原因是leader进程无法创建会话，而进程组id 61052与进程id 61052相同，即当前进程是leader进程。因此需要一个子进程来创建新的session，这一点需要特别注意。
0 ) {
//[parent1] process id: 61052, parent process id: 8243, process group id: 61052, session id: 8243 exited the process
echoes ("parent1", "exited the process");
exit;
}
//the child process id created is 61053, but the process group, session, and parent process are still the same
//[child1] process id: 61053, parent process id: 61052, process group id: 61052, session id: 8243
echomsg("child1");
//calling the posix_setsid function will create a new session and process group, and set the process group id and session id to that process id
if (-1 === \posix_setsid()) {
throw new exception("setsid fail");
}
//now you will find that both the process group id and session id have changed to 61053, which is equivalent to starting a session window similar to a linux terminal
//[child1] process id: 61053, parent process id: 61052, process group id: 61053, session id: 61053
echomsg("child1");
//second fork process
//the reason for requiring a secondary fork process here is to avoid being associated with terminal controlled processes. this process 61052 was created in terminal mode
//need to detach from process 61052 to ensure the stability of the daemon process
$pid = pcntl_fork();
if ( $pid < 0 ){
exit('fork error');
} else if( $pid > 0 ) {
//[parent2] process id: 61053, parent process id: 61052, process group id: 61053, session id: 61053 exited the process
echoes ("parent2", "exited the process");
exit;
}
//at this point, the process has broken free from the control of the terminal process and formed a daemon process
//[child2] process id: 61054, parent process id: 61053, process group id: 61053, session id: 61053. this process is retained
echoes ("child2", "this process is reserved");
sleep(100);
最好有时间的朋友自己执行代码分析一下，会有不同的奖励。这里假装你已经实践过了，我们看一下workerman的worker.php文件第554行runall方法中的static::daemon()函数。实现的流程逻辑与上面的例子几乎相同。不过这里也用到了umask函数，它的主要作用是给进程创建的文件或目录授予相应的权限，保证有对文件或目录进行操作的权限。
// workerman/Worker.php:554
/**
* Run all worker instances.
*Run process
* @return void
*/
public static function runAll()
{
static::checkSapiEnv();
static::init();
static::parseCommand();
static::lock();
//Create a process and form a daemon process
static::daemonize();
static::initWorkers();
static::installSignal();
static::saveMasterPid();
static::lock(\LOCK_UN);
static::displayUI();
static::forkWorkers();
static::resetStd();
static::monitorWorkers();
}
// workerman/Worker.php:1262
/**
* Run as daemon mode.
*Run in daemon mode
* @throws Exception
*/
protected static function daemonize()
{
//Determine whether it is already in a guarded state and whether the current system is in a Linux environment
if (! static::$daemonize || static::$_OS !== \OS_TYPE_LINUX) {
return;
}
//If umask is set to 0, the file permissions created by the current process will be 777, which has the highest permission
\umask(0);
//The first time creating a process
$pid = \pcntl_fork();
if (-1 === $pid) {
//Process creation failed
throw new Exception('Fork fail');
} elseif ($pid > 0) {
//Main process exits
exit(0);
}
//The child process continues to execute
//By calling the posix_setsid function, a process can detach from its parent process and transform into a daemon process
if (-1 === \posix_setsid()) {
throw new Exception("Setsid fail");
}
//The second creation process, in a System V based system, exits by forking the parent process again
//Ensure that the formed daemon process does not become the first session process and does not have control terminals
$pid = \pcntl_fork();
if (-1 === $pid) {
//Process creation failed
throw new Exception("Fork fail");
} elseif (0 !== $pid) {
//Main process exits
exit(0);
}
//The child process continues to execute
}
守护进程也是workerman的重要组成部分，保证了workerman进程的稳定性。与我们通过nohup启动的命令不同，这些命令有时会在后台挂起而没有人注意到，朋友们可能已经经历过这一点。当然，市面上也有一些开源的守护进程管理软件，比如supervisors，也有人使用screen、tmux等会话终端来实现。事实上，守护进程的实现方式有很多种。这里仅介绍一个php中实现守护进程模式的例子来分析workerman中守护进程的实现原理。希望本内容能够对您有所帮助。
如果您想体验新版本php或者切换php版本，欢迎使用servbay部署php开发环境。这是我正在开发的产品，我会长期维护这个工具。
立即学习“PHP免费学习笔记（深入）”；
