Epoll horizontal departure, which edge triggers

LT and ET Pattern

epoll Operations on file descriptors are 2 Patterns : LT and ET

LT    Level Trigger, Level trigger      Default      Only unprocessed read / write events on the file description symbol will be notified , As long as there are events, they will continue to trigger , Until it's done
ET    Edge Trigger, edge-triggered      adopt EPOLLET To set up      Notify when and only when a read-write event arrives , The same event is triggered only once, or only when two state transitions from non triggering to triggering are triggered

LT(level triggered) It's the default way of working , And at the same time Support block and no-block socket. In this way , The kernel tells you if a file descriptor is ready , Then you can be ready for this fd Conduct IO operation . If you don't do anything , The kernel will continue to inform you , therefore , This mode is less likely to make errors in programming . Conventional select/poll They are all representatives of this model . This model is equivalent to an efficient poll

To adopt LT File descriptor for mode operation , When epoll_wait When an event is detected on it and the application is notified of this event , The application does not have to deal with this event immediately . such ,　 When the application next calls epoll_wait when , epoll_wait This event will also be notified to the application again , Until the event is handled .

ET (edge-triggered) It's a high-speed way of working , Only support no-block socket. In this mode , When the descriptor is never ready to be ready , The kernel passes through epoll Tell you . Then it assumes that you know the file descriptor is ready , And I won't describe it for that file Sign to send more ready notifications , Until you do something that makes that file descriptor no longer ready ( such as , You're sending , To receive or receive requests , Or when the transmitted and received data is less than a certain amount One. EWOULDBLOCK error ）. But notice , If it's not right all the time fd do IO operation ( So it becomes not ready again ), The kernel will not send any more notifications (only once), But in the TCP Agreement ,ET The acceleration effect of the model still needs more benchmark confirm .

And when going epoll Register a file descriptor in the kernel event table EPOLLET When an event is , epoll Will be with ET Mode to operate the file descriptor , ET The pattern is epoll The efficient working mode of . To adopt ET File descriptor for mode operation , When epoll_wait When an event is detected on it and the application is notified of this event , The application must process the event immediately , Because of the following epoll_wait The call will no longer notify the application of this event .

so , ET The mode reduces the same in a large way epoll The number of times the event was triggered repeatedly , So it's more efficient than LT Mode high .