前面讨论的定时方案都是以固定频率调用心搏函数tick,并在其中一次检测到期的定时器,然后执行到期定时器上的回调函数。设计定时器的另一中思路是,将所有定器超时时间最小的一个定时器的超时值作为心搏间隔。这样,一旦心搏函数tick执行,超时时间最小的定时器必然到期。我们就可以从剩余定时器中选出超时时间最小的一个,并将这个时间设为下一次心搏间隔。如此反复,就实现了较为精确的定时。
最小堆适合这种解决方案,下面直接给出最小堆方案的代码,并附有测试用例。不过测试用例我仍然只用了alarm来做测试。
#ifndef MIN_HEAP_H
#define MIN_HEAP_H
#include <iostream>
#include <netinet/in.h>
#include <time.h>
#include <assert.h>
#include <string.h>
const int BUFFER_SIZE = 1024;
class heap_timer;
//绑定socket和定时器
struct client_data {
sockaddr_in addr_;
int sockfd_;
char buf_[BUFFER_SIZE];
heap_timer* timer_;
};
//定时器类
class heap_timer {
public:
heap_timer(int delay) {
printf("birth time %d, delay: %d\n", time(NULL), delay);
expire_ = time(NULL) + delay; //注意,和之前的升序链表以及时间轮不同,这次我们在timer中初始化生效时间
}
public:
void (*timeout_callback_)(client_data*); //定时器的回调函数
public:
time_t expire_; //定时器生效的绝对时间
client_data* user_data_; //用户数据
};
class time_heap {
public:
time_heap(int cap) throw (std::exception) //构造函数之一,初始化一个大小为cap的空堆
: array_(new heap_timer*[cap]), capacity_(cap), cur_size_(0) {
memset(array_, 0, sizeof(array_)); //初始化指针
}
time_heap(heap_timer** init_array, int size, int capacity) throw (std::exception) //构造函数之二,使用已有数组来初始堆
: array_(new heap_timer*[capacity]), cur_size_(size), capacity_(capacity) {
assert(capacity >= size);
memset(array_, 0, sizeof(array_));
if(cur_size_ != 0){
for(int i=0; i<cur_size_; ++i)
array_[i] = init_array[i]; //初始化堆数组
for(int i=((cur_size_-1)>>1); i>=0; --i)
sift_down(i); //多数组中的(cur_size_-1)/2 ~ 0 个元素执行下滤操作
}
}
~time_heap() { //销毁时间堆
for(int i=0; i<cur_size_; ++i)
delete array_[i];
delete []array_;
}
public:
//添加目标定时器timer
void add_timer(heap_timer* timer) throw (std::exception) {
assert(timer != NULL);
if(cur_size_ >= capacity_) //如果当前堆数组容量不够,扩容
resize();
//新插入一个元素,当前堆大小加1,hole是新插入元素的位置
int hole = cur_size_++; //hole = cur_size_ - 1
int parent = 0;
//对从新插入位置到根节点路径上所有节点进行上滤操作
for(; hole > 0; hole = parent){
parent = (hole - 1) >> 1;
if(array_[parent]->expire_ <= timer->expire_)
break;
array_[hole] = array_[parent];
}
array_[hole] = timer;
}
//调整,老规矩,先删除,再添加一次,否则旧定时器还作祟
void adjust_timer(heap_timer* old_timer, heap_timer* new_timer) {
del_timer(old_timer);
add_timer(new_timer);
}
//删除目标定时器
void del_timer(heap_timer* timer) {
assert(timer != NULL);
//仅仅将目标定时器的回调函数设置为空,即所谓的延迟销毁。这将节省真正删除该定时器造成的开销,但这样做容易使堆数组膨胀
//说下为什么延迟销毁效率高:如果不延迟我们需要查找O(lgn)+删除下滤调整O(lgn);而采用延迟删除,由于其他定时器可能触发,其他定时器会调用真正的删除函数pop_heap来删除自己,所以我们采用了延迟删除的定时器慢慢会被调整到堆顶,这时候由于时间到达,执行回调函数(我们已经赋空,tick函数中不会执行空回调函数)。由于为空,什么也不做。然后被pop_heap,实际上仅仅消耗删除调整的代价,无需查找操作,时间复杂度是O(lgn)。
timer->timeout_callback_ = NULL;
}
heap_timer* top() const { //获得堆顶定时器
return empty() ? NULL : array_[0];
}
//删除堆顶定时器
void pop_timer() {
if(empty()) //有可能目前时空的,什么也不做
return ;
if(array_[0] != NULL){
printf("delete user: %d\n", array_[0]->user_data_->sockfd_);
delete array_[0];
//将原来的堆顶元素替换为堆数组中最后一个元素,提高效率
array_[0] = array_[--cur_size_];
printf("cur_size_: %d\n", cur_size_);
sift_down(0); //对新的堆顶元素执行下滤操作
}
}
//心搏函数
void tick() {
heap_timer* tmp = array_[0];
time_t cur = time(NULL); //循环处理堆中到期的定时器
while(!empty()){
if(tmp == NULL)
break;
printf("current time is: %d\n", cur);
if(tmp->expire_ > cur) //如果堆顶定时器没到期,则退出循环
break;
if(array_[0]->timeout_callback_ != NULL) //否则执行没有被延迟删除的定时器的任务
array_[0]->timeout_callback_(array_[0]->user_data_);
//将堆顶元素删除,同时生成新的堆顶定时器(array[0])
pop_timer();
tmp = array_[0];
}
}
bool empty() const { return cur_size_ == 0; }
private:
void sift_down(int hole) { //下滤操作
auto left_child = [] (int i) { return 2*i + 1; };
heap_timer* tmp = array_[hole];
int child = -1;
//注意比较的是left_child(hole)而不是hole
for(; left_child(hole)<cur_size_; hole=child){
child = left_child(hole);
if(child < cur_size_-1
&& array_[child]->expire_ > array_[child+1]->expire_)
++child;
if(array_[child]->expire_ < tmp->expire_)
array_[hole] = array_[child];
else
break;
}
array_[hole] = tmp;
}
//扩容函数
void resize() throw (std::exception) {
heap_timer** tmp = new heap_timer* [capacity_<<1];
memset(tmp, 0, sizeof(tmp));
capacity_ = capacity_ << 1;
for(int i=0; i<cur_size_; ++i)
tmp[i] = array_[i];
delete []array_;
array_ = tmp;
}
private:
heap_timer** array_; //堆数组
int capacity_; //堆数组的容量
int cur_size_; //堆数组当前包含的元素的个数
};
#endif
测试代码:
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <assert.h>
#include <stdio.h>
#include <signal.h>
#include <unistd.h>
#include <errno.h>
#include <string.h>
#include <fcntl.h>
#include <stdlib.h>
#include <sys/epoll.h>
#include <pthread.h>
#include <memory>
#include <vector>
#include "min_heap.h"
const int FD_LIMIT = 65535;
const int MAX_EVENT_NUMBER = 1024;
const int TIME_SLOT = 5;
const int INIT_HEAP_SIZE = 2;
static int pipefd[2];
static int epollfd = 0;
static std::shared_ptr<time_heap> timer_lst;
int setnonblocking( int fd )
{
int old_option = fcntl( fd, F_GETFL );
int new_option = old_option | O_NONBLOCK;
fcntl( fd, F_SETFL, new_option );
return old_option;
}
void addfd(int fd )
{
epoll_event event;
event.data.fd = fd;
event.events = EPOLLIN | EPOLLET;
epoll_ctl( epollfd, EPOLL_CTL_ADD, fd, &event );
setnonblocking( fd );
}
void sig_handler( int sig )
{
int save_errno = errno;
int msg = sig;
send( pipefd[1], ( char* )&msg, 1, 0 );
errno = save_errno;
}
void addsig( int sig )
{
struct sigaction sa;
memset( &sa, '\0', sizeof( sa ) );
sa.sa_handler = sig_handler;
sa.sa_flags |= SA_RESTART;
sigfillset( &sa.sa_mask );
assert( sigaction( sig, &sa, NULL ) != -1 );
}
void timer_handler()
{
timer_lst->tick();
alarm( TIME_SLOT );
}
void cb_func( client_data* user_data )
{
epoll_ctl( epollfd, EPOLL_CTL_DEL, user_data->sockfd_, 0 );
assert( user_data );
close( user_data->sockfd_ );
printf( "close fd %d\n", user_data->sockfd_ );
}
int main( int argc, char* argv[] )
{
if( argc <= 2 )
{
printf( "usage: %s ip_address port_number\n", basename( argv[0] ) );
return 1;
}
const char* ip = argv[1];
int port = atoi( argv[2] );
int ret = 0;
struct sockaddr_in address;
bzero( &address, sizeof( address ) );
address.sin_family = AF_INET;
inet_pton( AF_INET, ip, &address.sin_addr );
address.sin_port = htons( port );
int listenfd = socket( PF_INET, SOCK_STREAM, 0 );
assert( listenfd >= 0 );
int on = 1;
ret = setsockopt(listenfd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
assert(ret != -1);
ret = bind( listenfd, ( struct sockaddr* )&address, sizeof( address ) );
assert( ret != -1 );
ret = listen( listenfd, 5 );
assert( ret != -1 );
epoll_event events[ MAX_EVENT_NUMBER ];
epollfd = epoll_create( 5 );
assert( epollfd != -1 );
addfd(listenfd );
ret = socketpair( PF_UNIX, SOCK_STREAM, 0, pipefd );
assert( ret != -1 );
setnonblocking( pipefd[1] );
addfd(pipefd[0] );
// add all the interesting signals here
addsig( SIGALRM );
addsig( SIGTERM );
bool stop_server = false;
std::vector<client_data> users(FD_LIMIT);
bool timeout = false;
alarm( TIME_SLOT );
//////////////////////////////////////////////////////
timer_lst.reset(new time_heap(INIT_HEAP_SIZE));
//////////////////////////////////////////////////////
while( !stop_server )
{
int number = epoll_wait( epollfd, events, MAX_EVENT_NUMBER, -1 );
if ( ( number < 0 ) && ( errno != EINTR ) )
{
printf( "epoll failure\n" );
break;
}
for ( int i = 0; i < number; i++ )
{
int sockfd = events[i].data.fd;
if( sockfd == listenfd )
{
struct sockaddr_in client_address;
socklen_t client_addrlength = sizeof( client_address );
int connfd = accept( listenfd, ( struct sockaddr* )&client_address, &client_addrlength );
addfd(connfd);
users[connfd].addr_ = client_address;
users[connfd].sockfd_ = connfd;
printf("user: %d\n", connfd);
heap_timer *timer = new heap_timer(3 * TIME_SLOT);
timer_lst->add_timer(timer);
timer->user_data_ = &users[connfd];
timer->timeout_callback_ = cb_func;
users[connfd].timer_ = timer;
}
else if( ( sockfd == pipefd[0] ) && ( events[i].events & EPOLLIN ) )
{
int sig;
char signals[1024];
ret = recv( pipefd[0], signals, sizeof( signals ), 0 );
if( ret == -1 )
{
// handle the error
continue;
}
else if( ret == 0 )
{
continue;
}
else
{
for( int i = 0; i < ret; ++i )
{
switch( signals[i] )
{
case SIGALRM:
{
timeout = true;
break;
}
case SIGTERM:
{
stop_server = true;
}
}
}
}
}
else if( events[i].events & EPOLLIN )
{
memset( users[sockfd].buf_, '\0', BUFFER_SIZE );
ret = recv( sockfd, users[sockfd].buf_, BUFFER_SIZE-1, 0 );
printf( "get %d bytes of client data %s from %d\n", ret, users[sockfd].buf_, sockfd );
heap_timer* timer = users[sockfd].timer_;
if( ret < 0 )
{
if( errno != EAGAIN )
{
cb_func( &users[sockfd] );
if( timer )
{
timer_lst->del_timer( timer );
}
}
}
else if( ret == 0 )
{
cb_func( &users[sockfd] );
if( timer )
{
timer_lst->del_timer( timer );
}
}
else
{
//send( sockfd, users[sockfd].buf, BUFFER_SIZE-1, 0 );
if( timer )
{
printf("user: %d\n", sockfd);
heap_timer* new_timer = new heap_timer(3*TIME_SLOT);
timer_lst->adjust_timer(timer, new_timer);
new_timer->user_data_ = &users[sockfd];
new_timer->timeout_callback_ = cb_func;
users[sockfd].timer_ = new_timer;
}
}
}
else
{
// others
}
}
if( timeout )
{
timer_handler();
timeout = false;
}
}
close( listenfd );
close( epollfd );
close( pipefd[1] );
close( pipefd[0] );
return 0;
}
对时间堆而言,添加一个定时器的时间复杂度是O(lgn),删除一个定时器的时间复杂度是O(1)(采用了延迟删除),执行一个定时器的时间复杂度是O(1)。因此,时间堆的效率是很高的。