I/o multiplexing & event driven yyds dry inventory

Event driven

Definition

A driver architecture , Decoupling the generation and consumption of events . You can refer to   Release / Subscription model

The core

I/O Operations are abstracted into events , Each event corresponds to Callback function To deal with it . Distribute by event , These events can be detected one by one , And call the corresponding callback function to handle .

Lord - from reactor+worker threads Pattern

Definition

The main reactor thread is only responsible for distributing Acceptor Connection is established , Connected... On socket I/O The incident was handed over to from -reactor Responsible for distribution . take CPU Intensive work is entrusted to worker Thread pool to handle , And these tasks are divided into one sub task .encode After that, the completed results are determined by sub-reactor Of I/O Thread send out .

solve the problem

It's solved I/O The efficiency of distribution , Business logic and I/O Coupling between distributions .

---

I/O Multiplexing

select & poll

technological process

1. User thread call select, take fd_set Copy from user space to kernel space

2. The kernel is in kernel space for fd_set Go through it , Check to see if there are any ready socket The descriptor , If not , It goes into sleep , Until there's something ready socket The descriptor

3. The kernel returns select To the user thread , The number of ready file descriptors

4. After the user gets the number of ready file descriptors , Again fd_set Traversal , Find the file descriptor in place

5. The user thread reads and writes the ready file descriptor

shortcoming

1、 Every time you call select function , All of them fd_set from User mode copy to Kernel mode , influence performance .

2、 have O(n) The undifferentiated polling time complexity of , Poll all fd_set, Long polling time .

3、select maximum connection （ Maximum number of file descriptors supported ） Limited , It's usually 1024

difference

poll Use pollfd Structure, not select Of fd_set structure ,pollfd Structures use linked lists instead of arrays ,pollfd There is no limit to the length of

epoll（ Synchronous nonblocking 、 subscribe 、 Callback 、 Red and black trees 、 Linked list ）

Implement and use ：

1、 call  epoll_create  Build a epoll Handle object

2、 call  epoll_ctl Add a new socket , Add monitoring events and register callback functions

3、 When a file descriptor is ready , The kernel will use something like callback The callback mechanism of , Quickly activate the file descriptor , The event triggered will be   Callback function   Join in eventpoll Of Linked list rdlist in .

4、 call  epoll_wait  Collect connections for events that occur （ Check eventpoll Object rdlist Whether there are elements in the linked list ）

Level trigger 、 edge-triggered

Level trigger ：epoll_wait Monitoring , Data that has not been processed will always be returned

edge-triggered ： Pay attention to whether there are new events in the file descriptor , What has been returned will not be returned

Red and black tree root node rbr： The root node of the red black tree , This tree stores everything added to epoll Events to be monitored in

Linked list rdlist： In the linked list are stored the information to be passed through epoll_wait Events that meet the conditions returned to the user

shortcoming

Above 3 Both are synchronous for the caller