Linux 惊群现象

        在linux中，惊群现象已经消失了的，我们可以看   http://simohayha.javaeye.com/blog/561424 ，但是当我们在开发服务器时候，需要使用epoll，发现一个问题，就是当一个请求过来的时候，发现有的时候被唤起的进程不止一个，看下面的程序:#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 <pthread.h> #include <errno.h> #include <unistd.h> #include <sys/ipc.h> #include <sys/shm.h> #include <errno.h> #define KEY 1234 #define SIZE 1024 #define PORT 9999 #define MAXFDS 5000 #define EVENTSIZE 100 void process(); int fd, cfd,opt=1; int shmid; char *shmaddr; struct shmid_ds buf; int num = 0 ; int main(int argc, char *argv[]) { shmid = shmget(KEY,SIZE,IPC_CREAT|0600); /* 建立共享内存 */ if(shmid == -1){ printf("create share memory failed/n"); } shmaddr = (char *)shmat(shmid,NULL,0); if(shmaddr == (void *)-1){ printf("connect to the share memory failed: %s",strerror(errno)); return 0; } strcpy(shmaddr,"1/n"); struct sockaddr_in sin, cin; socklen_t sin_len = sizeof(struct sockaddr_in); if ((fd = socket(AF_INET, SOCK_STREAM, 0)) <= 0) { fprintf(stderr, "socket failed/n"); return -1; } memset(&sin, 0, sizeof(struct sockaddr_in)); sin.sin_family = AF_INET; sin.sin_port = htons((short)(PORT)); sin.sin_addr.s_addr = INADDR_ANY; if (bind(fd, (struct sockaddr *)&sin, sizeof(sin)) != 0) { fprintf(stderr, "bind failed/n"); return -1; } if (listen(fd, 32) != 0) { fprintf(stderr, "listen failed/n"); return -1; } int i ; for(i = 0; i < 2; i++) { int pid = fork(); if(pid == 0) { process(); } } while(1) ; return 0; } void process() { struct epoll_event ev; struct epoll_event events[1000]; int kdpfd = epoll_create(1000); int len = sizeof(struct sockaddr_in); ev.events = EPOLLIN | EPOLLET; ev.data.fd = fd; int new_fd; if((fcntl(fd, F_GETFL, 0)&O_NONBLOCK)) printf("ok non block/n"); else printf("wrong non block/n"); printf("sub socket is %d /n", fd); if (epoll_ctl(kdpfd, EPOLL_CTL_ADD, fd, &ev) < 0) { fprintf(stderr, "epoll set insertion error: fd=%d/n", fd); return ; } else { printf("监听 socket 加入 epoll 成功！/n"); } struct sockaddr_in my_addr, their_addr; while (1) { /* 等待有事件发生 */ int nfds = epoll_wait(kdpfd, events, 20, 500); if (nfds == -1) { perror("epoll_wait"); break; } /* 处理所有事件 */ //printf("num of event is :%d /n",nfds); int n; for (n = 0; n < nfds; ++n) { if (events[n].data.fd == fd) { new_fd = accept(fd, (struct sockaddr *) &their_addr,&len); if (new_fd < 0) { printf("accept error/n"); continue; } else { printf("%d create new socket: %d/n", getpid(), new_fd); } } } } }

 

当请求过来的时候，会出现accept error,而且我发现这个时候的socket id 都是 -1。我们可以在这里添加判断来解决问题，至于为什么epoll会引入这个问题，暂时还不清楚，需要进一步的学习。