1. Epoll是何方神圣?
Epoll可是当前在Linux下开发大规模并发网络程序的热门人选,Epoll 在Linux2.6内核中正式引入,和select相似,其实都I/O多路复用技术而已,并没有什么神秘的。
其实在Linux下设计并发网络程序,向来不缺少方法,比如典型的Apache模型(Process Per Connection,简称PPC),TPC(Thread PerConnection)模型,以及select模型和poll模型,那为何还要再引入Epoll这个东东呢?那还是有得说说的…
2. 常用模型的缺点
如果不摆出来其他模型的缺点,怎么能对比出Epoll的优点呢。
2.1 PPC/TPC模型
这两种模型思想类似,就是让每一个到来的连接一边自己做事去,别再来烦我。只是PPC是为它开了一个进程,而TPC开了一个线程。可是别烦我是有代价的,它要时间和空间啊,连接多了之后,那么多的进程/线程切换,这开销就上来了;因此这类模型能接受的最大连接数都不会高,一般在几百个左右。
2.2 select模型
1. 最大并发数限制,因为一个进程所打开的FD(文件描述符)是有限制的,由FD_SETSIZE设置,默认值是1024/2048,因此Select模型的最大并发数就被相应限制了。自己改改这个FD_SETSIZE?想法虽好,可是先看看下面吧…
2. 效率问题,select每次调用都会线性扫描全部的FD集合,这样效率就会呈现线性下降,把FD_SETSIZE改大的后果就是,大家都慢慢来,什么?都超时了??!!
3. 内核/用户空间 内存拷贝问题,如何让内核把FD消息通知给用户空间呢?在这个问题上select采取了内存拷贝方法。
2.3 poll模型
基本上效率和select是相同的,select缺点的2和3它都没有改掉。
3. Epoll的提升
把其他模型逐个批判了一下,再来看看Epoll的改进之处吧,其实把select的缺点反过来那就是Epoll的优点了。
3.1. Epoll没有最大并发连接的限制,上限是最大可以打开文件的数目,这个数字一般远大于2048, 一般来说这个数目和系统内存关系很大,具体数目可以cat /proc/sys/fs/file-max察看。
3.2. 效率提升,Epoll最大的优点就在于它只管你“活跃”的连接,而跟连接总数无关,因此在实际的网络环境中,Epoll的效率就会远远高于select和poll。
3.3. 内存拷贝,Epoll在这点上使用了“共享内存”,这个内存拷贝也省略了。
4. Epoll为什么高效
Epoll的高效和其数据结构的设计是密不可分的,这个下面就会提到。
首先回忆一下select模型,当有I/O事件到来时,select通知应用程序有事件到了快去处理,而应用程序必须轮询所有的FD集合,测试每个FD是否有事件发生,并处理事件;代码像下面这样:
int res = select(maxfd+1, &readfds, NULL, NULL, 120);
if(res > 0)
{
for(int i = 0; i < MAX_CONNECTION; i++)
{
if(FD_ISSET(allConnection[i],&readfds))
{
handleEvent(allConnection[i]);
}
}
}
// if(res == 0) handle timeout, res < 0 handle error
Epoll不仅会告诉应用程序有I/0事件到来,还会告诉应用程序相关的信息,这些信息是应用程序填充的,因此根据这些信息应用程序就能直接定位到事件,而不必遍历整个FD集合。
intres = epoll_wait(epfd, events, 20, 120);
for(int i = 0; i < res;i++)
{
handleEvent(events[n]);
}
5. Epoll关键数据结构
前面提到Epoll速度快和其数据结构密不可分,其关键数据结构就是:
structepoll_event {
__uint32_t events; // Epoll events
epoll_data_t data; // User datavariable
};
typedef union epoll_data {
void *ptr;
int fd;
__uint32_t u32;
__uint64_t u64;
} epoll_data_t;
可见epoll_data是一个union结构体,借助于它应用程序可以保存很多类型的信息:fd、指针等等。有了它,应用程序就可以直接定位目标了。
6. 使用Epoll
既然Epoll相比select这么好,那么用起来如何呢?会不会很繁琐啊…先看看下面的三个函数吧,就知道Epoll的易用了。
intepoll_create(int size);
生成一个Epoll专用的文件描述符,其实是申请一个内核空间,用来存放你想关注的socket fd上是否发生以及发生了什么事件。size就是你在这个Epoll fd上能关注的最大socket fd数,大小自定,只要内存足够。
intepoll_ctl(int epfd, intop, int fd, structepoll_event *event);
控制某个Epoll文件描述符上的事件:注册、修改、删除。其中参数epfd是epoll_create()创建Epoll专用的文件描述符。相对于select模型中的FD_SET和FD_CLR宏。
intepoll_wait(int epfd,structepoll_event * events,int maxevents,int timeout);
等待I/O事件的发生;参数说明:
epfd:由epoll_create() 生成的Epoll专用的文件描述符;
epoll_event:用于回传代处理事件的数组;
maxevents:每次能处理的事件数;
timeout:等待I/O事件发生的超时值;
返回发生事件数。
相对于select模型中的select函数。
7. 例子程序
下面是一个简单Echo Server的例子程序,麻雀虽小,五脏俱全,还包含了一个简单的超时检查机制,简洁起见没有做错误处理。
- //
- // a simple echo server using epoll in linux
- //
- // 2009-11-05
- // 2013-03-22:修改了几个问题,1是/n格式问题,2是去掉了原代码不小心加上的ET模式;
- // 本来只是简单的示意程序,决定还是加上 recv/send时的buffer偏移
- // by sparkling
- //
- #include <sys/socket.h>
- #include <sys/epoll.h>
- #include <netinet/in.h>
- #include <arpa/inet.h>
- #include <fcntl.h>
- #include <unistd.h>
- #include <stdio.h>
- #include <errno.h>
- #include <iostream>
- using namespace std;
- #define MAX_EVENTS 500
- struct myevent_s
- {
- int fd;
- void (*call_back)(int fd, int events, void *arg);
- int events;
- void *arg;
- int status; // 1: in epoll wait list, 0 not in
- char buff[128]; // recv data buffer
- int len, s_offset;
- long last_active; // last active time
- };
- // set event
- void EventSet(myevent_s *ev, int fd, void (*call_back)(int, int, void*), void *arg)
- {
- ev->fd = fd;
- ev->call_back = call_back;
- ev->events = 0;
- ev->arg = arg;
- ev->status = 0;
- bzero(ev->buff, sizeof(ev->buff));
- ev->s_offset = 0;
- ev->len = 0;
- ev->last_active = time(NULL);
- }
- // add/mod an event to epoll
- void EventAdd(int epollFd, int events, myevent_s *ev)
- {
- struct epoll_event epv = {0, {0}};
- int op;
- epv.data.ptr = ev;
- epv.events = ev->events = events;
- if(ev->status == 1){
- op = EPOLL_CTL_MOD;
- }
- else{
- op = EPOLL_CTL_ADD;
- ev->status = 1;
- }
- if(epoll_ctl(epollFd, op, ev->fd, &epv) < 0)
- printf("Event Add failed[fd=%d], evnets[%d]\n", ev->fd, events);
- else
- printf("Event Add OK[fd=%d], op=%d, evnets[%0X]\n", ev->fd, op, events);
- }
- // delete an event from epoll
- void EventDel(int epollFd, myevent_s *ev)
- {
- struct epoll_event epv = {0, {0}};
- if(ev->status != 1) return;
- epv.data.ptr = ev;
- ev->status = 0;
- epoll_ctl(epollFd, EPOLL_CTL_DEL, ev->fd, &epv);
- }
- int g_epollFd;
- myevent_s g_Events[MAX_EVENTS+1]; // g_Events[MAX_EVENTS] is used by listen fd
- void RecvData(int fd, int events, void *arg);
- void SendData(int fd, int events, void *arg);
- // accept new connections from clients
- void AcceptConn(int fd, int events, void *arg)
- {
- struct sockaddr_in sin;
- socklen_t len = sizeof(struct sockaddr_in);
- int nfd, i;
- // accept
- if((nfd = accept(fd, (struct sockaddr*)&sin, &len)) == -1)
- {
- if(errno != EAGAIN && errno != EINTR)
- {
- }
- printf("%s: accept, %d", __func__, errno);
- return;
- }
- do
- {
- for(i = 0; i < MAX_EVENTS; i++)
- {
- if(g_Events[i].status == 0)
- {
- break;
- }
- }
- if(i == MAX_EVENTS)
- {
- printf("%s:max connection limit[%d].", __func__, MAX_EVENTS);
- break;
- }
- // set nonblocking
- int iret = 0;
- if((iret = fcntl(nfd, F_SETFL, O_NONBLOCK)) < 0)
- {
- printf("%s: fcntl nonblocking failed:%d", __func__, iret);
- break;
- }
- // add a read event for receive data
- EventSet(&g_Events[i], nfd, RecvData, &g_Events[i]);
- EventAdd(g_epollFd, EPOLLIN, &g_Events[i]);
- }while(0);
- printf("new conn[%s:%d][time:%d], pos[%d]\n", inet_ntoa(sin.sin_addr),
- ntohs(sin.sin_port), g_Events[i].last_active, i);
- }
- // receive data
- void RecvData(int fd, int events, void *arg)
- {
- struct myevent_s *ev = (struct myevent_s*)arg;
- int len;
- // receive data
- len = recv(fd, ev->buff+ev->len, sizeof(ev->buff)-1-ev->len, 0);
- EventDel(g_epollFd, ev);
- if(len > 0)
- {
- ev->len += len;
- ev->buff[len] = '\0';
- printf("C[%d]:%s\n", fd, ev->buff);
- // change to send event
- EventSet(ev, fd, SendData, ev);
- EventAdd(g_epollFd, EPOLLOUT, ev);
- }
- else if(len == 0)
- {
- close(ev->fd);
- printf("[fd=%d] pos[%d], closed gracefully.\n", fd, ev-g_Events);
- }
- else
- {
- close(ev->fd);
- printf("recv[fd=%d] error[%d]:%s\n", fd, errno, strerror(errno));
- }
- }
- // send data
- void SendData(int fd, int events, void *arg)
- {
- struct myevent_s *ev = (struct myevent_s*)arg;
- int len;
- // send data
- len = send(fd, ev->buff + ev->s_offset, ev->len - ev->s_offset, 0);
- if(len > 0)
- {
- printf("send[fd=%d], [%d<->%d]%s\n", fd, len, ev->len, ev->buff);
- ev->s_offset += len;
- if(ev->s_offset == ev->len)
- {
- // change to receive event
- EventDel(g_epollFd, ev);
- EventSet(ev, fd, RecvData, ev);
- EventAdd(g_epollFd, EPOLLIN, ev);
- }
- }
- else
- {
- close(ev->fd);
- EventDel(g_epollFd, ev);
- printf("send[fd=%d] error[%d]\n", fd, errno);
- }
- }
- void InitListenSocket(int epollFd, short port)
- {
- int listenFd = socket(AF_INET, SOCK_STREAM, 0);
- fcntl(listenFd, F_SETFL, O_NONBLOCK); // set non-blocking
- printf("server listen fd=%d\n", listenFd);
- EventSet(&g_Events[MAX_EVENTS], listenFd, AcceptConn, &g_Events[MAX_EVENTS]);
- // add listen socket
- EventAdd(epollFd, EPOLLIN, &g_Events[MAX_EVENTS]);
- // bind & listen
- sockaddr_in sin;
- bzero(&sin, sizeof(sin));
- sin.sin_family = AF_INET;
- sin.sin_addr.s_addr = INADDR_ANY;
- sin.sin_port = htons(port);
- bind(listenFd, (const sockaddr*)&sin, sizeof(sin));
- listen(listenFd, 5);
- }
- int main(int argc, char **argv)
- {
- unsigned short port = 12345; // default port
- if(argc == 2){
- port = atoi(argv[1]);
- }
- // create epoll
- g_epollFd = epoll_create(MAX_EVENTS);
- if(g_epollFd <= 0) printf("create epoll failed.%d\n", g_epollFd);
- // create & bind listen socket, and add to epoll, set non-blocking
- InitListenSocket(g_epollFd, port);
- // event loop
- struct epoll_event events[MAX_EVENTS];
- printf("server running:port[%d]\n", port);
- int checkPos = 0;
- while(1){
- // a simple timeout check here, every time 100, better to use a mini-heap, and add timer event
- long now = time(NULL);
- for(int i = 0; i < 100; i++, checkPos++) // doesn't check listen fd
- {
- if(checkPos == MAX_EVENTS) checkPos = 0; // recycle
- if(g_Events[checkPos].status != 1) continue;
- long duration = now - g_Events[checkPos].last_active;
- if(duration >= 60) // 60s timeout
- {
- close(g_Events[checkPos].fd);
- printf("[fd=%d] timeout[%d--%d].\n", g_Events[checkPos].fd, g_Events[checkPos].last_active, now);
- EventDel(g_epollFd, &g_Events[checkPos]);
- }
- }
- // wait for events to happen
- int fds = epoll_wait(g_epollFd, events, MAX_EVENTS, 1000);
- if(fds < 0){
- printf("epoll_wait error, exit\n");
- break;
- }
- for(int i = 0; i < fds; i++){
- myevent_s *ev = (struct myevent_s*)events[i].data.ptr;
- if((events[i].events&EPOLLIN)&&(ev->events&EPOLLIN)) // read event
- {
- ev->call_back(ev->fd, events[i].events, ev->arg);
- }
- if((events[i].events&EPOLLOUT)&&(ev->events&EPOLLOUT)) // write event
- {
- ev->call_back(ev->fd, events[i].events, ev->arg);
- }
- }
- }
- // free resource
- return 0;
- }
// // a simple echo server using epoll in linux // // 2009-11-05 // 2013-03-22:修改了几个问题,1是/n格式问题,2是去掉了原代码不小心加上的ET模式;// 本来只是简单的示意程序,决定还是加上 recv/send时的buffer偏移// by sparkling // #include <sys/socket.h> #include <sys/epoll.h> #include <netinet/in.h> #include <arpa/inet.h> #include <fcntl.h> #include <unistd.h> #include <stdio.h> #include <errno.h> #include <iostream> using namespace std; #define MAX_EVENTS 500 struct myevent_s { int fd; void (*call_back)(int fd, int events, void *arg); int events; void *arg; int status; // 1: in epoll wait list, 0 not in char buff[128]; // recv data buffer int len, s_offset; long last_active; // last active time }; // set event void EventSet(myevent_s *ev, int fd, void (*call_back)(int, int, void*), void *arg) { ev->fd = fd; ev->call_back = call_back; ev->events = 0; ev->arg = arg; ev->status = 0; bzero(ev->buff, sizeof(ev->buff)); ev->s_offset = 0; ev->len = 0; ev->last_active = time(NULL); } // add/mod an event to epoll void EventAdd(int epollFd, int events, myevent_s *ev) { struct epoll_event epv = {0, {0}}; int op; epv.data.ptr = ev; epv.events = ev->events = events; if(ev->status == 1){ op = EPOLL_CTL_MOD; } else{ op = EPOLL_CTL_ADD; ev->status = 1; } if(epoll_ctl(epollFd, op, ev->fd, &epv) < 0) printf("Event Add failed[fd=%d], evnets[%d]\n", ev->fd, events); else printf("Event Add OK[fd=%d], op=%d, evnets[%0X]\n", ev->fd, op, events); } // delete an event from epoll void EventDel(int epollFd, myevent_s *ev) { struct epoll_event epv = {0, {0}}; if(ev->status != 1) return; epv.data.ptr = ev; ev->status = 0; epoll_ctl(epollFd, EPOLL_CTL_DEL, ev->fd, &epv); } int g_epollFd; myevent_s g_Events[MAX_EVENTS+1]; // g_Events[MAX_EVENTS] is used by listen fd void RecvData(int fd, int events, void *arg); void SendData(int fd, int events, void *arg); // accept new connections from clients void AcceptConn(int fd, int events, void *arg) { struct sockaddr_in sin; socklen_t len = sizeof(struct sockaddr_in); int nfd, i; // accept if((nfd = accept(fd, (struct sockaddr*)&sin, &len)) == -1) { if(errno != EAGAIN && errno != EINTR) { } printf("%s: accept, %d", __func__, errno); return; } do { for(i = 0; i < MAX_EVENTS; i++) { if(g_Events[i].status == 0) { break; } } if(i == MAX_EVENTS) { printf("%s:max connection limit[%d].", __func__, MAX_EVENTS); break; } // set nonblocking int iret = 0; if((iret = fcntl(nfd, F_SETFL, O_NONBLOCK)) < 0) { printf("%s: fcntl nonblocking failed:%d", __func__, iret); break; } // add a read event for receive data EventSet(&g_Events[i], nfd, RecvData, &g_Events[i]); EventAdd(g_epollFd, EPOLLIN, &g_Events[i]); }while(0); printf("new conn[%s:%d][time:%d], pos[%d]\n", inet_ntoa(sin.sin_addr), ntohs(sin.sin_port), g_Events[i].last_active, i); } // receive data void RecvData(int fd, int events, void *arg) { struct myevent_s *ev = (struct myevent_s*)arg; int len; // receive data len = recv(fd, ev->buff+ev->len, sizeof(ev->buff)-1-ev->len, 0); EventDel(g_epollFd, ev); if(len > 0) { ev->len += len; ev->buff[len] = '\0'; printf("C[%d]:%s\n", fd, ev->buff); // change to send event EventSet(ev, fd, SendData, ev); EventAdd(g_epollFd, EPOLLOUT, ev); } else if(len == 0) { close(ev->fd); printf("[fd=%d] pos[%d], closed gracefully.\n", fd, ev-g_Events); } else { close(ev->fd); printf("recv[fd=%d] error[%d]:%s\n", fd, errno, strerror(errno)); } } // send data void SendData(int fd, int events, void *arg) { struct myevent_s *ev = (struct myevent_s*)arg; int len; // send data len = send(fd, ev->buff + ev->s_offset, ev->len - ev->s_offset, 0); if(len > 0) { printf("send[fd=%d], [%d<->%d]%s\n", fd, len, ev->len, ev->buff); ev->s_offset += len; if(ev->s_offset == ev->len) { // change to receive event EventDel(g_epollFd, ev); EventSet(ev, fd, RecvData, ev); EventAdd(g_epollFd, EPOLLIN, ev); } } else { close(ev->fd); EventDel(g_epollFd, ev); printf("send[fd=%d] error[%d]\n", fd, errno); } } void InitListenSocket(int epollFd, short port) { int listenFd = socket(AF_INET, SOCK_STREAM, 0); fcntl(listenFd, F_SETFL, O_NONBLOCK); // set non-blocking printf("server listen fd=%d\n", listenFd); EventSet(&g_Events[MAX_EVENTS], listenFd, AcceptConn, &g_Events[MAX_EVENTS]); // add listen socket EventAdd(epollFd, EPOLLIN, &g_Events[MAX_EVENTS]); // bind & listen sockaddr_in sin; bzero(&sin, sizeof(sin)); sin.sin_family = AF_INET; sin.sin_addr.s_addr = INADDR_ANY; sin.sin_port = htons(port); bind(listenFd, (const sockaddr*)&sin, sizeof(sin)); listen(listenFd, 5); } int main(int argc, char **argv) { unsigned short port = 12345; // default port if(argc == 2){ port = atoi(argv[1]); } // create epoll g_epollFd = epoll_create(MAX_EVENTS); if(g_epollFd <= 0) printf("create epoll failed.%d\n", g_epollFd); // create & bind listen socket, and add to epoll, set non-blocking InitListenSocket(g_epollFd, port); // event loop struct epoll_event events[MAX_EVENTS]; printf("server running:port[%d]\n", port); int checkPos = 0; while(1){ // a simple timeout check here, every time 100, better to use a mini-heap, and add timer event long now = time(NULL); for(int i = 0; i < 100; i++, checkPos++) // doesn't check listen fd { if(checkPos == MAX_EVENTS) checkPos = 0; // recycle if(g_Events[checkPos].status != 1) continue; long duration = now - g_Events[checkPos].last_active; if(duration >= 60) // 60s timeout { close(g_Events[checkPos].fd); printf("[fd=%d] timeout[%d--%d].\n", g_Events[checkPos].fd, g_Events[checkPos].last_active, now); EventDel(g_epollFd, &g_Events[checkPos]); } } // wait for events to happen int fds = epoll_wait(g_epollFd, events, MAX_EVENTS, 1000); if(fds < 0){ printf("epoll_wait error, exit\n"); break; } for(int i = 0; i < fds; i++){ myevent_s *ev = (struct myevent_s*)events[i].data.ptr; if((events[i].events&EPOLLIN)&&(ev->events&EPOLLIN)) // read event { ev->call_back(ev->fd, events[i].events, ev->arg); } if((events[i].events&EPOLLOUT)&&(ev->events&EPOLLOUT)) // write event { ev->call_back(ev->fd, events[i].events, ev->arg); } } } // free resource return 0; }