Chinaunix首页 | 论坛 | 博客

kainankainan.blog.chinaunix.net

首页　| 　博文目录　| 　关于我

gsnumen

  • 博客访问： 93496
  • 博文数量： 9
  • 博客积分： 330
  • 博客等级： 一等列兵
  • 技术积分： 117
  • 用 户 组： 普通用户
  • 注册时间： 2010-08-05 18:36
文章分类

全部博文（9）

文章存档

2012年（4）

2011年（1）

2010年（4）

我的朋友
最近访客
推荐博文
相关博文
nginx的进程模式

分类： C/C++

2012-09-14 16:04:50

nginx的进程模式

1 进程间的通信方式
    使用 UNIX域套接字 socketpair() 异步通讯机制: nginx在创建worker前将先调用 socketpair(int channel[2]) 然后将 channel[0-1]设置为异步通知方式,并注册evnet事件,父进程使用channel[0],子进程使用channel[1]实现双方的通讯.

2 都有哪些通信?
2.1 创建worker进程时
2.2 执行 nginx -s reload 时
2.3 执行 nginx -s stop 时
2.4 执行 nginx -s quit 时
2.5 执行 nginx -s reopen 时
2.6 某worker异常退出时
2.7 概要

点击(此处)折叠或打开

  1. SIGHUP NGX_CMD_QUIT|SIGQUIT
  2.     ./nginx -s reload ---------> master (ngx_reload=1) -------------------------> worker|cache_manager|cache_loader (ngx_quit=1, 处理完待读socket再退出)


  5.                       SIGTERM|SIGINT NGX_CMD_TERMINATE|SIGTERM
  6.     ./nginx -s stop -----------------> master (ngx_terminate=1,延迟退出时间) ----------------------------> worker|cache_manager|cache_loader (ngx_terminate=1,直接退出)


  9.                       SIGQUIT NGX_CMD_QUIT|SIGQUIT
  10.     ./nginx -s quit ----------> master (ngx_quit=1) ------------------------> worker|cache_manager|cache_loader (ngx_quit=1, 处理完待读socket再退出)

  12.                         SIGUSER1 NGX_CMD_REOPEN|SIGUSER1
  13.     ./nginx -s reopen -----------> master (ngx_reopen=1) -------------------------> worker|cache_manager|cache_loader (ngx_reopen=1)


3 通信的内容是什么?

点击(此处)折叠或打开

  1. src/os/unix/ngx_process_cycle.h:
  2.     #define NGX_CMD_OPEN_CHANNEL 1
  3.     #define NGX_CMD_CLOSE_CHANNEL 2
  4.     #define NGX_CMD_QUIT 3
  5.     #define NGX_CMD_TERMINATE 4
  6.     #define NGX_CMD_REOPEN 5


4 创建worker进程时 是如何通信的?
master(父进程):
    1> 在进程池 ngx_processes 中为 新worker进程 分配一个位置,如果进程池已满,也就是创建的进程数超过了最大值1024 nginx报错但不退出;
    2> 调用 socketpair(AF_UNIX, SOCK_STREAM, 0, ngx_processes[s].channel) 创建UNIX域套接字对.并设置为异步模式且execve时关闭;
    3> 将 新worker进程 的{NGX_CMD_OPEN_CHANNEL, channel[0], 在进程池中的位置, PID} 通过 UNIX域套接字 告知已存在的子进程.

点击(此处)折叠或打开

  1. {
  2.         ngx_start_worker_processes()
  3.         {
  4.             ch.command=NGX_CMD_OPEN_CHANNEL;
  5.             for (i=0; i < n; i++) {
  6.                 //...
  7.                 ch.pid=ngx_processes[ngx_process_slot].pid;
  8.                 ch.slot=ngx_process_slot;
  9.                 ch.fd=ngx_processes[ngx_process_slot].channel[0];
  10.                 ngx_pass_open_channel(cycle, &ch);
  11.             }
  12.         }
  13. }


新worker进程:
    1> 关闭进程池中其他子进程的 channel[1];
    2> 关闭自身的 channel[0];
    3> 将自身的 channel[1] 注册event读事件: ngx_channel_handler().

点击(此处)折叠或打开

  1. {
  2.     ngx_worker_process_cycle()
  3.     {
  4.         //...
  5.         ngx_worker_process_init()
  6.         {
  7.             //...
  8.             for (n=0; n < ngx_last_process; n++) {

  10.                 if (ngx_processes[n].pid == -1) {
  11.                     continue;
  12.                 }

  14.                 if (n == ngx_process_slot) {
  15.                     continue;
  16.                 }

  18.                 if (ngx_processes[n].channel[1] == -1) {
  19.                     continue;
  20.                 }

  22.                 if (close(ngx_processes[n].channel[1]) == -1) {
  23.                     ngx_log_error(NGX_LOG_ALERT, cycle->log, ngx_errno,
  24.                             "close() channel failed");
  25.                 }
  26.             }

  28.             if (close(ngx_processes[ngx_process_slot].channel[0]) == -1) {
  29.             }
  30.             if (ngx_add_channel_event(cycle, ngx_channel, NGX_READ_EVENT,
  31.                         ngx_channel_handler)
  32.                     == NGX_ERROR)
  33.             {
  34.                 /* fatal */
  35.                 exit(2);
  36.             }
  37.         }
  38.     }
  39. }


    自身可以通过 进程池中其他子进程的 channel[0] 收发 其他子进程 的信息.
    自身可以通过 自身的 channel[1] 收发 master进程 的信息.
    
其他子进程:
    此时 其他子进程的 UNIX域套接字 收到 master进程的 {NGX_CMD_OPEN_CHANNEL, channel[0], 在进程池中的位置, PID}; 读事件处理函数 ngx_channel_handler()被调用;
    1> 获得 新子进程 在进程池中的位置;
    2> 保存 新子进程 的 channel[0];
    2> 保存 新子进程 的 PID.

点击(此处)折叠或打开

  1. {
  2.     ngx_channel_handler()
  3.     {
  4.         for ( ;; ) {
  5.             n=ngx_read_channel(c->fd, &ch, sizeof(ngx_channel_t), ev->log);
  6.             switch (ch.command) {
  7.                 case NGX_CMD_QUIT:
  8.                     ngx_quit=1;
  9.                     break;

  11.                 case NGX_CMD_TERMINATE:
  12.                     ngx_terminate=1;
  13.                     break;

  15.                 case NGX_CMD_REOPEN:
  16.                     ngx_reopen=1;
  17.                     break;

  19.                 case NGX_CMD_OPEN_CHANNEL:
  20.                     ngx_processes[ch.slot].pid=ch.pid;
  21.                     ngx_processes[ch.slot].channel[0]=ch.fd;
  22.                     break;

  24.                 case NGX_CMD_CLOSE_CHANNEL:
  25.                     if (close(ngx_processes[ch.slot].channel[0]) == -1) {
  26.                     }
  27.                     ngx_processes[ch.slot].channel[0]=-1;
  28.                     break;
  29.             }
  30.         }
  31.     }
  32. }


5 执行 nginx -s reload 时
执行 nginx -s reload 的进程:
    从main函数开始一直执行到 

点击(此处)折叠或打开

  1. if (ngx_signal) {
  2.         return ngx_signal_process(cycle, ngx_signal);
  3.     }

    从nginx.pid文件中解析出当前的 master 进程PID, 向其发送 reload 信号

master:
    1> 接收到 reload 信号, 信号处理函数 ngx_signal_handler 中将 ngx_reconfigure=1;
    2> 进入 ngx_reconfigure 状态; 调用 ngx_init_cycle() 初始化环境,重新加载配置文件, 重新创建 worker进程, cache管理进程, cache加载进程(热代码忽略);
    3> 使用 UNIX域套接字 通知老的worker进程, cache管理进程(cache加载进程此时可能存在,也可能已经退出,因为在ngx_cache_loader_process_handler() {exit(0)}), NGX_CMD_QUIT;
        这里不通知的进程有: 刚刚创建的worker进程, cache管理进程, cache加载进程, 正在退出中且本次发送信号为 shutdown的进程, 热代码加载中的进程.
    4> 如果使用 UNIX域套接字发送 NGX_CMD_QUIT 失败 ,则使用 kill 发送 SIGQUIT, 将这些进程的属性 exiting=1; 如果发送信号时某进程已经退出,则将其属性 exited=1;
    5> 过N久之后, 这些被发送信号的进程会退出, master进程会收到内核发送来的 SIGCHLD,将 ngx_reap=1; 并调用 ngx_process_get_status()回收退出的子进程;
    6> 回收子进程 ngx_process_get_status();
    7> reload状态,也会进入 ngx_reap 状态, 这个状态是否会重新创建 那些老的可再生的子进程 呢?
        首先, 重新创建 老的可再生的子进程的条件是: 
            (ngx_processes[i].exited && ngx_processes[i].respawn && !ngx_processes[i].exiting && !ngx_terminate && !ngx_quit)
        reload状态,可以通过观察进程属性 respawn, exiting, exited 的值来判断是否会重新创建 那些老的可再生的子进程. 除了 cache loader 进程的respawn=0, 其他进程 respawn =1. 故只有 exited=1 && exiting=0 时才会重建.

        初始值 exiting=0, exited=0;
        reload状态, 通知老的子进程退出有三种情况:

点击(此处)折叠或打开

  1. {
  2.     /* 接受到 reload 信号: 1 "SIGHUP" ngx_signal_handler "reload" */
  3.     ngx_signal_handler()
  4.     {
  5.         switch (ngx_process) {
  6.             case NGX_PROCESS_MASTER:
  7.             case NGX_PROCESS_SINGLE:
  8.                 switch (signo) {
  9.                     //...
  10.                     case ngx_signal_value(NGX_RECONFIGURE_SIGNAL):
  11.                         ngx_reconfigure=1;
  12.                         action=", reconfiguring";
  13.                         break;
  14.                     case SIGCHLD:
  15.                         ngx_reap=1;
  16.                         break;

  18.                     //...
  19.                 }
  20.                     //...
  21.         }
  22.     }
  23.     ngx_master_process_cycle()
  24.     {
  25.         //...
  26.         for (;;)
  27.         {
  28.             if (ngx_reap)
  29.             {
  30.                 ngx_reap=0;
  31.                 ngx_reap_children(cycle);
  32.             }
  33.         if (ngx_reconfigure) {
  34.             ngx_reconfigure=0;

  36.             /* 热代码加载 */
  37.             if (ngx_new_binary) {
  38.                 ngx_start_worker_processes(cycle, ccf->worker_processes,
  39.                                            NGX_PROCESS_RESPAWN);
  40.                 ngx_start_cache_manager_processes(cycle, 0);
  41.                 ngx_noaccepting=0;

  43.                 continue;
  44.             }

  46.             /* 重新初始化,加载配置 */
  47.             cycle=ngx_init_cycle(cycle);

  49.             ngx_cycle=cycle;
  50.             ccf=(ngx_core_conf_t *) ngx_get_conf(cycle->conf_ctx,
  51.                                                    ngx_core_module);
  52.             /* 重新创建新的 worker进程 */
  53.             ngx_start_worker_processes(cycle, ccf->worker_processes,
  54.                                        NGX_PROCESS_JUST_RESPAWN);
  55.             /* 重新创建新的 缓存池管理进程 和 缓存池加载进程 */
  56.             ngx_start_cache_manager_processes(cycle, 1);
  57.             /* 通知 */
  58.             ngx_signal_worker_processes(cycle, ngx_signal_value(NGX_SHUTDOWN_SIGNAL));
  59.             {
  60.                 switch (signo) {

  62.                     case ngx_signal_value(NGX_SHUTDOWN_SIGNAL):
  63.                         ch.command=NGX_CMD_QUIT;
  64.                         break;

  66.                     default:
  67.                         ch.command=0;
  68.                 }
  69.                 for (i=0; i < ngx_last_process; i++) {

  71.                     /* 如果是 新创建的进程(NGX_PROCESS_JUST_RESPAWN), 正在退出的进程(ngx_processes[i]->exiting), 热代码加载的进程, 则不通知他们;
  72.                        否则通过 UNIX域套接字 告知.
  73.                      */
  74.                     if (ch.command) {
  75.                         if (ngx_write_channel(ngx_processes[i].channel[0],
  76.                                     &ch, sizeof(ngx_channel_t), cycle->log)
  77.                                 == NGX_OK)
  78.                         {
  79.                             if (signo != ngx_signal_value(NGX_REOPEN_SIGNAL)) {
  80.                                 ngx_processes[i].exiting=1; /* 将该进程的状态 exiting=1 */
  81.                             }

  83.                             continue;
  84.                         }
  85.                     }

  87.                     if (kill(ngx_processes[i].pid, signo) == -1) {
  88.                         if (err == NGX_ESRCH) { /* 该子进程进程为僵死进程 */
  89.                             ngx_processes[i].exited=1;
  90.                             ngx_processes[i].exiting=0;
  91.                             ngx_reap=1;
  92.                         }

  94.                         continue;
  95.                     }
  96.                     /* 除了 reopen 信号外, 其他信号都会使子进程退出, 这里将那些子进程的状态改为正在退出的状态.*/
  97.                     if (signo != ngx_signal_value(NGX_REOPEN_SIGNAL)) {
  98.                         ngx_processes[i].exiting=1;
  99.                     }
  100.                 }
  101.             }
  102.         }
  103.         }
  104.     }
  105.     ngx_process_get_status()
  106.     {
  107.         /* 循环回收 */
  108.         for ( ;; ) {

  110.             pid=waitpid(-1, &status, WNOHANG);

  112.             /* 如果子进程持有 accept 锁, 则释放. */
  113.             if (ngx_accept_mutex_ptr) {
  114.                 ngx_atomic_cmp_set(ngx_accept_mutex_ptr, pid, 0);
  115.             }

  117.             /* 将该进程在master进程池中的属性 status=其退出码, exited=1, */
  118.             for (i=0; i < ngx_last_process; i++) {
  119.                 if (ngx_processes[i].pid == pid) {
  120.                     ngx_processes[i].status=status;
  121.                     ngx_processes[i].exited=1;
  122.                     process=ngx_processes[i].name;
  123.                     break;
  124.                 }
  125.             }

  127.             /* 打印日志: 如果子进程被信号中断,则打印出信号编号, 否则打印出其退出码*/
  128.             //...

  130.             /* nginx子进程有个 可再生(respawn, NGX_PROCESS_RESPAWN)属性: 如果该属性有效 master检测其退出后会再次创建它.
  131.                其中 worker进程, cache管理进程创建后,对应的 ngx_processes[i]->respawn=1; cache加载进程 ngx_processes[i]->respawn=0;意思推出后不需要再次创建它.
  132.                子进程的退出方式有:
  133.                被信号中断
  134.                自身退出,退出码 0 对应 cache 加载进程
  135.                自身退出,退出码 2 对应reload时,被通知的哪些旧的 worker进程, cache管理进程, cache加载进程(如果它reload时它还存在)退出时的退出码
  136.                这里就是使用这中机制来确定 master回收这些进程时,是否重新创建它.

  138.                退出码为2且拥有可再生权限,说明在重新加载配置文件, 这些退出的进程无须再创建他们, 将其 respawn=0;
  139.              */
  140.             if (WEXITSTATUS(status) == 2 && ngx_processes[i].respawn) {
  141.                 ngx_processes[i].respawn=0;
  142.             }
  143.         }
  144.     }
  145.     ngx_reap_children()
  146.     {
  147.         ch.command=NGX_CMD_CLOSE_CHANNEL;
  148.         for (i=0; i < ngx_last_process; i++) {
  149.             if (ngx_processes[i].exited) {
  150.                 if (!ngx_processes[i].detached) {
  151.                     /* 清理子进程的 UNIX域套接字 资源, 通过 UNIX域套接字告知其他存活子进程 NGX_CMD_CLOSE_CHANNEL */
  152.                 }
  153.                 //...
  154.             } else if (ngx_processes[i].exiting || !ngx_processes[i].detached) {
  155.                 live=1;
  156.             }
  157.         }

  159.         return live;
  160.     }

  162. }


点击(此处)折叠或打开

  1. {
  2.     ngx_channel_handler()
  3.     {
  4.         for ( ;; ) {
  5.             n=ngx_read_channel(c->fd, &ch, sizeof(ngx_channel_t), ev->log);
  6.             switch (ch.command) {

  8.                 case NGX_CMD_QUIT:
  9.                         ngx_quit=1;
  10.                             break;
  11.             }
  12.         }
  13.     }
  14.     ngx_signal_handler()
  15.     {
  16.         switch (ngx_process) {
  17.             //...
  18.             case NGX_PROCESS_WORKER:
  19.             case NGX_PROCESS_HELPER:
  20.                 switch (signo) {
  21.                     //...
  22.                     case ngx_signal_value(NGX_SHUTDOWN_SIGNAL):
  23.                         ngx_quit=1;
  24.                         action=", shutting down";
  25.                         break;
  26.                         //...
  27.                 }
  28.                 //...
  29.                 break;
  30.         }
  31.     }
  32.     ngx_worker_process_cycle()
  33.     {
  34.         if (ngx_exiting) {
  35.             c=cycle->connections;
  36.             for (cycle->connection) {
  37.                 if (c[i].fd != -1 && c[i].idle) {
  38.                     c[i].close=1;
  39.                     c[i].read->handler(c[i].read);
  40.                 }
  41.             }
  42.             if (ngx_event_timer_rbtree.root == ngx_event_timer_rbtree.sentinel)
  43.             {
  44.                 ngx_worker_process_exit(cycle);
  45.             }
  46.         }

  48.         if (ngx_quit) {
  49.             ngx_quit=0;
  50.             ngx_setproctitle("worker process is shutting down");

  52.             if (!ngx_exiting) {
  53.                 ngx_close_listening_sockets(cycle);
  54.                 ngx_exiting=1; /* 将 ngx_exiting 置为有效. */
  55.             }
  56.         }

  58.     }

  60. }





点击(此处)折叠或打开

  1. {
  2.     /* 接收信号 15 "SIGTERM" ngx_signal_handler "stop" */
  3.     ngx_signal_handler()
  4.     {
  5.         switch (ngx_process) {
  6.             case NGX_PROCESS_MASTER:
  7.             case NGX_PROCESS_SINGLE:
  8.                 switch (signo) {
  9.                     //...
  10.                     case ngx_signal_value(NGX_TERMINATE_SIGNAL):
  11.                     case SIGINT:
  12.                         ngx_terminate=1;
  13.                         break;
  14.                     case SIGALRM:
  15.                         ngx_sigalrm=1;
  16.                         break;

  18.                     //...
  19.                 }
  20.                     //...
  21.         }
  22.     }
  23.     ngx_master_process_cycle()
  24.     {
  25.         for (;;)
  26.         {
  27.             if (delay) {
  28.                 if (ngx_sigalrm) {
  29.                     sigio=0;
  30.                     delay *= 2;
  31.                     ngx_sigalrm=0;
  32.                 }
  33.                 itv.it_interval.tv_sec=0;
  34.                 itv.it_interval.tv_usec=0;
  35.                 itv.it_value.tv_sec=delay / 1000;
  36.                 itv.it_value.tv_usec=(delay % 1000 ) * 1000;

  38.                 if (setitimer(ITIMER_REAL, &itv, NULL) == -1) {
  39.                 }
  40.             }

  42.             sigsuspend(&set);

  44.             ngx_time_update();
  45.             if (ngx_reap) {
  46.                 ngx_reap = 0;
  47.                 ngx_log_debug0(NGX_LOG_DEBUG_EVENT, cycle->log, 0, "reap children");

  49.                 live = ngx_reap_children(cycle);
  50.             }

  52.             if (!live && (ngx_terminate || ngx_quit)) {
  53.                 ngx_master_process_exit(cycle);
  54.             }


  57.         /* 退出机制: nginx -s stop
  58.            master 收到 SIGTERM 或者 SIGINT 时, 进入 ngx_terminate 状态, 然后给所有子进程通过 UNIX域套接字发送 发送 NGX_CMD_QUIT, 通过kill() SIGTERM 信号
  59.            给每个子进程预留 50 毫秒的退出时间, 使用setitmer()延迟 n*50 毫秒, 当 n*50 还有子进程未退出, 则延迟时间增加为 n*2*50 以此类推. 如果子进程在 1000 毫秒内还没有全部退出,则对还在运行的子进程发送 SIGKILL.
  60.            SIGKILL 是不能屏蔽也不能注册信号处理函数的信号, 动作为是 终止.
  61.          */
  62.             if (ngx_terminate) {
  63.                 if (delay == 0) {
  64.                     delay=50;
  65.                 }

  67.                 if (sigio) {
  68.                     sigio--;
  69.                     continue;
  70.                 }

  72.                 sigio=ccf->worker_processes + 2 /* cache processes */;

  74.                 if (delay > 1000) {
  75.                     ngx_signal_worker_processes(cycle, SIGKILL);
  76.                 } else {
  77.                     ngx_signal_worker_processes(cycle,
  78.                             ngx_signal_value(NGX_TERMINATE_SIGNAL));
  79.                 }

  81.                 continue;
  82.             }
  83.         }
  84.     }
  85. }



点击(此处)折叠或打开

  1. {
  2.     ngx_channel_handler()
  3.     {
  4.         for ( ;; ) {
  5.             n=ngx_read_channel(c->fd, &ch, sizeof(ngx_channel_t), ev->log);
  6.             switch (ch.command) {

  8.                 case NGX_CMD_TERMINATE:
  9.                         ngx_terminate=1;
  10.                             break;
  11.             }
  12.         }
  13.     }
  14.     ngx_signal_handler()
  15.     {
  16.         switch (ngx_process) {
  17.             //...
  18.             case NGX_PROCESS_WORKER:
  19.             case NGX_PROCESS_HELPER:
  20.                 switch (signo) {
  21.                     //...
  22.                     case ngx_signal_value(NGX_TERMINATE_SIGNAL):
  23.                         ngx_terminate=1;
  24.                         action=", shutting down";
  25.                         break;
  26.                         //...
  27.                 }
  28.                 //...
  29.         }
  30.     }
  31.     ngx_worker_process_cycle()
  32.     {
  33.         //...
  34.         ngx_process_events_and_timers(cycle);

  36.         if (ngx_terminate) {
  37.             ngx_log_error(NGX_LOG_NOTICE, cycle->log, 0, "exiting");

  39.             ngx_worker_process_exit(cycle);
  40.         }
  41.     }
  42. }




点击(此处)折叠或打开

  1. {
  2.     // 接收信号 3 "SIGQUIT" ngx_signal_handler "quit"
  3.     ngx_signal_handler()
  4.     {
  5.         switch (ngx_process) {
  6.             case NGX_PROCESS_MASTER:
  7.             case NGX_PROCESS_SINGLE:
  8.                 switch (signo) {
  9.                     //...
  10.                     case ngx_signal_value(NGX_SHUTDOWN_SIGNAL):
  11.                         ngx_quit=1;
  12.                         break;
  13.                     //...
  14.                 }
  15.                     //...
  16.         }
  17.     }
  18.     ngx_master_process_cycle()
  19.     {
  20.         for (;;)
  21.         {
  22.             //...
  23.             sigsuspend(&set);

  25.             ngx_time_update();

  27.             if (ngx_reap) {
  28.                 ngx_reap = 0;
  29.                 ngx_log_debug0(NGX_LOG_DEBUG_EVENT, cycle->log, 0, "reap children");

  31.                 live = ngx_reap_children(cycle);
  32.             }

  34.             if (!live && (ngx_terminate || ngx_quit)) {
  35.                 ngx_master_process_exit(cycle);
  36.             }

  38.             if (ngx_quit) {
  39.                 ngx_signal_worker_processes(cycle,
  40.                         ngx_signal_value(NGX_SHUTDOWN_SIGNAL));

  42.                 ls = cycle->listening.elts;
  43.                 for (n = 0; n < cycle->listening.nelts; n++) {
  44.                     if (ngx_close_socket(ls[n].fd) == -1) {
  45.                         ngx_log_error(NGX_LOG_EMERG, cycle->log, ngx_socket_errno,
  46.                                 ngx_close_socket_n " %V failed",
  47.                                 &ls[n].addr_text);
  48.                     }
  49.                 }
  50.                 cycle->listening.nelts = 0;

  52.                 continue;
  53.             }
  54.         }
  55.     }
  56. }



点击(此处)折叠或打开

  1. {
  2.     ngx_channel_handler()
  3.     {
  4.         for ( ;; ) {
  5.             n=ngx_read_channel(c->fd, &ch, sizeof(ngx_channel_t), ev->log);
  6.             switch (ch.command) {

  8.                 case NGX_CMD_QUIT:
  9.                         ngx_quit=1;
  10.                             break;
  11.             }
  12.         }
  13.     }
  14.     ngx_signal_handler()
  15.     {
  16.         switch (ngx_process) {
  17.             //...
  18.             case NGX_PROCESS_WORKER:
  19.             case NGX_PROCESS_HELPER:
  20.                 switch (signo) {
  21.                     //...
  22.                     case ngx_signal_value(NGX_SHUTDOWN_SIGNAL):
  23.                         ngx_quit=1;
  24.                         action=", shutting down";
  25.                         break;
  26.                         //...
  27.                 }
  28.                 //...
  29.                 break;
  30.         }
  31.     }
  32.     ngx_worker_process_cycle()
  33.     {
  34.         if (ngx_exiting) {
  35.             c=cycle->connections;
  36.             for (cycle->connection) {
  37.                 if (c[i].fd != -1 && c[i].idle) {
  38.                     c[i].close=1;
  39.                     c[i].read->handler(c[i].read);
  40.                 }
  41.             }
  42.             if (ngx_event_timer_rbtree.root == ngx_event_timer_rbtree.sentinel)
  43.             {
  44.                 ngx_worker_process_exit(cycle);
  45.             }
  46.         }

  48.         if (ngx_quit) {
  49.             ngx_quit=0;
  50.             ngx_setproctitle("worker process is shutting down");

  52.             if (!ngx_exiting) {
  53.                 ngx_close_listening_sockets(cycle);
  54.                 ngx_exiting=1; /* 将 ngx_exiting 置为有效. */
  55.             }
  56.         }

  58.     }
  59. }





点击(此处)折叠或打开

  1. {
  2.     // 接收信号 10 "SIGUSR1" ngx_signal_handler "reopen"
  3.     ngx_signal_handler()
  4.     {
  5.         switch (ngx_process) {
  6.             case NGX_PROCESS_MASTER:
  7.             case NGX_PROCESS_SINGLE:
  8.                 switch (signo) {
  9.                     //...
  10.                     case ngx_signal_value(NGX_REOPEN_SIGNAL):
  11.                         ngx_reopen=1;
  12.                         break;
  13.                     //...
  14.                 }
  15.                     //...
  16.         }
  17.     }
  18.     ngx_master_process_cycle()
  19.     {
  20.         for (;;)
  21.         {
  22.             //...
  23.             sigsuspend(&set);

  25.             ngx_time_update();

  27.             if (ngx_reopen) {
  28.                 ngx_reopen = 0;
  29.                 ngx_log_error(NGX_LOG_NOTICE, cycle->log, 0, "reopening logs");
  30.                 ngx_reopen_files(cycle, ccf->user);
  31.                 ngx_signal_worker_processes(cycle,
  32.                         ngx_signal_value(NGX_REOPEN_SIGNAL));
  33.             }
  34.         }
  35.     }
  36. }


点击(此处)折叠或打开

  1. {
  2.     ngx_channel_handler()
  3.     {
  4.         for ( ;; ) {
  5.             n=ngx_read_channel(c->fd, &ch, sizeof(ngx_channel_t), ev->log);
  6.             switch (ch.command) {

  8.                 case NGX_CMD_REOPEN:
  9.                         ngx_reopen=1;
  10.                             break;
  11.             }
  12.         }
  13.     }
  14.     ngx_signal_handler()
  15.     {
  16.         switch (ngx_process) {
  17.             //...
  18.             case NGX_PROCESS_WORKER:
  19.             case NGX_PROCESS_HELPER:
  20.                 switch (signo) {
  21.                     //...
  22.                     case ngx_signal_value(NGX_REOPEN_SIGNAL):
  23.                         ngx_reopen=1;
  24.                         action=", shutting down";
  25.                         break;
  26.                         //...
  27.                 }
  28.                 //...
  29.                 break;
  30.         }
  31.     }
  32.     ngx_worker_process_cycle()
  33.     {
  34.         if (ngx_reopen) {
  35.             ngx_reopen = 0;
  36.             ngx_log_error(NGX_LOG_NOTICE, cycle->log, 0, "reopening logs");
  37.             ngx_reopen_files(cycle, -1);
  38.         }
  39.     }
  40. }



点击(此处)折叠或打开

  1. {
  2.     ngx_signal_handler()
  3.     {
  4.         switch (ngx_process) {
  5.             case NGX_PROCESS_MASTER:
  6.             case NGX_PROCESS_SINGLE:
  7.                 switch (signo) {
  8.                     case SIGCHLD:
  9.                         ngx_reap=1;
  10.                         break;
  11.                 }
  12.         }

  14.         if (signo == SIGCHLD) {
  15.             ngx_process_get_status();
  16.         }
  17.     }
  18.     ngx_master_process_cycle()
  19.     {
  20.         for (;;)
  21.         {
  22.             if (ngx_reap)
  23.             {
  24.                 ngx_reap=0;
  25.                 live=ngx_reap_children(cycle);
  26.             }
  27.         }
  28.     }
  29.     ngx_reap_children()
  30.     {
  31.         ch.command=NGX_CMD_CLOSE_CHANNEL;
  32.         for (i=0; i < ngx_last_process; i++) {
  33.             if (ngx_processes[i].exited) {

  35.                 if (!ngx_processes[i].detached) {

  37.                     ngx_close_channel(ngx_processes[i].channel, cycle->log);

  39.                     ngx_processes[i].channel[0]=-1;
  40.                     ngx_processes[i].channel[1]=-1;

  42.                     ch.pid=ngx_processes[i].pid;
  43.                     ch.slot=i;

  45.                     for (n=0; n < ngx_last_process; n++) {
  46.                         if (ngx_processes[n].exited
  47.                                 || ngx_processes[n].pid == -1
  48.                                 || ngx_processes[n].channel[0] == -1)
  49.                         {
  50.                             continue;
  51.                         }
  52.                         ngx_write_channel(ngx_processes[n].channel[0],
  53.                                 &ch, sizeof(ngx_channel_t), cycle->log);
  54.                     }
  55.                 }

  57.                 if (ngx_processes[i].respawn
  58.                         && !ngx_processes[i].exiting
  59.                         && !ngx_terminate
  60.                         && !ngx_quit)
  61.                 {
  62.                     if (ngx_spawn_process(cycle, ngx_processes[i].proc,
  63.                                 ngx_processes[i].data,
  64.                                 ngx_processes[i].name, i)
  65.                             == NGX_INVALID_PID)
  66.                     {
  67.                         ngx_log_error(NGX_LOG_ALERT, cycle->log, 0,
  68.                                 "could not respawn %s",
  69.                                 ngx_processes[i].name);
  70.                         continue;
  71.                     }


  74.                     ch.command=NGX_CMD_OPEN_CHANNEL;
  75.                     ch.pid=ngx_processes[ngx_process_slot].pid;
  76.                     ch.slot=ngx_process_slot;
  77.                     ch.fd=ngx_processes[ngx_process_slot].channel[0];

  79.                     ngx_pass_open_channel(cycle, &ch);

  81.                     live=1;

  83.                     continue;
  84.                 }

  86.                 if (ngx_processes[i].pid == ngx_new_binary) {

  88.                     ccf=(ngx_core_conf_t *) ngx_get_conf(cycle->conf_ctx,
  89.                             ngx_core_module);

  91.                     if (ngx_rename_file((char *) ccf->oldpid.data,
  92.                                 (char *) ccf->pid.data)
  93.                             != NGX_OK)
  94.                     {
  95.                         ngx_log_error(NGX_LOG_ALERT, cycle->log, ngx_errno,
  96.                                 ngx_rename_file_n " %s back to %s failed "
  97.                                 "after the new binary process \"%s\" exited",
  98.                                 ccf->oldpid.data, ccf->pid.data, ngx_argv[0]);
  99.                     }

  101.                     ngx_new_binary=0;
  102.                     if (ngx_noaccepting) {
  103.                         ngx_restart=1;
  104.                         ngx_noaccepting=0;
  105.                     }
  106.                 }

  108.                 if (i == ngx_last_process - 1) {
  109.                     ngx_last_process--;

  111.                 } else {
  112.                     ngx_processes[i].pid=-1;
  113.                 }

  115.             } else if (ngx_processes[i].exiting || !ngx_processes[i].detached) {
  116.                 live=1;
  117.             }
  118.         }

  120.         return live;
  121.     }
  122. }




阅读(4213) | 评论(0) | 转发(0) |
0

上一篇：shell 获取域名或IP地址的地理位置

下一篇：没有了

给主人留下些什么吧！~~
关于我们 | 关于IT168 | 联系方式 | 广告合作 | 法律声明 | 免费注册

Copyright 2001-2010 ChinaUnix.net All Rights Reserved 北京皓辰网域网络信息技术有限公司. 版权所有

感谢所有关心和支持过ChinaUnix的朋友们

16024965号-6