Nginx study notes (a): Nginx process model / event handling model

Nginx process model

​​​​

1. Multi-process model of
   inter-process independent of each other, no lock, and independently of each other;
   a process to exit the operation does not affect other processes, reduce risk;
   when the request comes, more worker to obtain a registration event by reading listenfd competition accrpt_mutex, reading events in the call accept to accept the connection.
    
2. Non-blocking asynchronous:
   processing a network event and I / O multiplexing the same, while monitoring a plurality of events, when an event is ready to go the read and write;
   request a process to handle only a constantly switch between the request;
   handover is asynchronous because the event is not ready to give up the initiative, similar to the loop handle multiple events prepared without any cost
    
3. Signal
   if Nginx in (when epoll_wait) wait event, program received signal, the signal handler after processing is complete, epoll_wait returns an error, then the program can be called again to enter epoll_wait
    
4. Timer
   Nginx With timer timeout epoll_wait implementation;
   Nginx timer event is placed on a red-black tree inside the maintenance timer each time before entering epoll_wait, starting with the red-black tree to get all the timer event the minimum time, in the calculation of the timeout epoll_wait into the epoll_wait;
   so, when there is no network events as they arrive, and no signal generation, epoll_wait time out, that is, to the timer event;
   then Nginx will check all of the timeout event, they state is set to timeout, and then to deal with network events.

Nginx event handling model (pseudo-code)

while (true) {
    for t in run_tasks:
        t.handler();
    update_time(&now);
    timeout = ETERNITY;  // 重置超时时间
    for t in wait_tasks: /* sorted already */
        if (t.time <= now) {
            t.timeout_handler(); // 超时
        } else {
            timeout = t.time - now; // 更新超时时间
            break;
        }
    nevents = poll_function(events, timeout); // I/O复用处理事件
    for i in nevents:
        task t;
        if (events[i].type == READ) {
            t.handler = read_handler;
        } else { /* events[i].type == WRITE */
            t.handler = write_handler;
        }
        run_tasks_add(t);
}