Entering the nginx event-driven model

Recently I was reading the event module of nginx in "In-depth Understanding of nginx" written by senior Ali Tao Hui. The reason why I want to read this content is because nginx can handle huge concurrent connections, and I want to see how the event driver behind it is built

I don’t want to post a whole bunch of code to talk about nginx in this blog post. Event-driven, I may not understand it so thoroughly, and it will be more difficult for readers to quickly grasp the knowledge, so I will simply summarize a few that I think may be helpful in my future server program development. Knowledge points

Briefly discuss the epoll model of nginx

This article only takes epoll as an example

nginx uses epoll as the basis of event drive. To handle new connection events, in our usual server design, since connection events are relatively sensitive (high requirements for fast response), I will open a separate thread (process) to specifically handle the connection, obtain the connection, and then distribute it to Each I/O multiplexing thread, however, nginx processes connection events and handles other events under the same I/O multiplexing. So how does it ensure the response requirements of connection events? Niginx does not call its callbacks on the acquired events, but puts them into two post queues. The two queues are

<code>.处理新连接事件
.处理定时事件
.处理普通读写事件
.处理从磁盘读事件
</code>

. The first queue is used to save connection events, while the second queue uses To save ordinary read and write events, we can ensure that the events in ngx_posted_accept_events are processed first during execution, so as to ensure the sensitivity of the connection to response speed

(2)How to prevent crosstalk

The crosstalk problem can be said to be a problem that needs to be dealt with in server programs. The crosstalk problem means that a socket has just been closed and a new connection has come in. The new connection happens to be allocated by the system to the socket that was just closed. Then if there are still unprocessed events in the socket just now, , and then it sends data to the corresponding socket, which is likely to be sent to the newly created user. So how does nginx solve this problem? It's very simple. Every time nginx obtains a new connection, it will invert a flag in the connection, so that the instance of this connection and the previous connection will be different, and each event contains the connection, so each time the event is processed You only need to compare whether the instances in the event are the same and it will be OK
(3) How to deal with the "thundering herd problem"
The so-called thundering herd problem means that multiple processes are listening to the same port at the same time. When a connection arrives, the system will wake up multiple processes, but of course only one process can handle the new connection, so other processes cannot. It needs to be awakened, but it is awakened. This is the noted thundering herd problem. The method nginx solves is also very simple. You only need to ensure that only one process is listening on the port at the same time to avoid the thundering group problem. But the key to the problem is how to ensure that only one process is listening to the port at the same time. nginx adopts the method of trying to lock, and determines whether the process should handle new connection events based on the return value of the lock, thereby solving the "thundering herd problem"
(4)How to solve load balancing
In a small network library I wrote before, the load balancing I used is very simple. The main thread is used to accept new connections, and then distributes the new connections to each child thread in turn, while nginx solves the load balancing problem between processes. There is no balanced distribution, but when the number of connections handled by each process exceeds 7/8 of the maximum number of connections it is allowed to handle, the connection will not be processed this time, but the number of connections it handles will be -1. This is equivalent to giving the opportunity to other threads, thereby achieving load balancing

Summary

Let’s write this about the nginx driver module first, and then I won’t be able to read it when I watch it. I am just writing this blog for Supervise yourself to study

').addClass('pre-numbering').hide(); $(this).addClass('has-numbering').parent().append($numbering); for (i = 1; i

').text(i)); }; $numbering.fadeIn(1700); }); }); The above introduces the introduction to the nginx event-driven model, including aspects of the content. I hope it will be helpful to friends who are interested in PHP tutorials.