1.基本介绍
按照ZooKeeper典型应用场景一览里的说法，分布式队列有两种，一种是常规的先进先出队列，另一种是要等到队列成员聚齐之后的才统一按序执行。
第二种队列可以先建立一个/queue，赋值为n，表达队列的大小。然后每个队列成员加入时，就判断是否达到队列要求的大小，如果是可以进行下一步动作，否则继续等待队列成员的加入。比较典型的情况是，当一个大的任务可能需要很多的子任务完成才能开始进行。
比如汇总账单的时候，就必须先将用户的消费数据，积分数据等都统计完成后才能开始。汇总账单的程序建立一个队列/Queue,赋值为2，然后分别统计消费数据和积分数据的程序当完成任务时就往/Queue下创建一个临时节点。而汇总账单程序监测到/Queue的子节点个数为2时，就可以开始执行任务了。
实际上，我们也可以先建立一个数目为2的子节点。当一个子任务完成的时候，就删除一个子节点，当所有子节点都被删除的时候，主任务就可以开始执行了。这个过程可以形象的理解为拆除屏障。因此这种队列还有一个专门的词语描述，叫做屏障（barrier）。
2.场景分析
讲了那么多的关于屏障的认识，但是并不打算就去实现它，并且Zookeeper的官方文档也有相关的知识。这次的主要目标是常规的FIFO队列。我将实现队列的两个主要操作：push和pop。
1). int push(zhandle_t *zkhandle,const char *path,char *element)
-
zkhandle为zookeeper_init初始化后的句柄 -
path为队列的路径 -
element为要压入队列的内容
2). int pop(zhandle_t *zkhandle,const char *path,char *element_buffer,int *buffer_len)
-
zkhandle为zookeeper_init初始化后的句柄 -
path为队列的路径 -
element_buffer为要弹出的缓冲区 -
buffer_len为指向缓冲区的大小的指针
简单来说，假设队列的路径为/Queue,push就是就是创建一个临时有序的/Queue/queue-节点。pop就是取出/Queue/下序列号最小的节点。
我们知道在C++中stl里有一个queue的类，实现了push,pop等操作，然而它是非线程安全的，即多个线程同时push/pop的时候可能会出现错误。而由于ZooKeeper保证了创建节点和删除节点的一致性，因此可以说利用Zookeeper实现的队列是进程安全的。
3. 场景实践
来看push和pop的具体实现。push的实现很简单，就是在{path}下创建一个有序的{path}/queue-子节点.
int push(zhandle_t *zkhandle,const char *path,char *element)
{
char child_path[512] = {0};
char path_buffer[512] = {0};
int bufferlen = sizeof(path_buffer);
sprintf(child_path,"%s/queue-",path);
int ret = zoo_create(zkhandle,child_path,element,strlen(element),
&ZOO_OPEN_ACL_UNSAFE,ZOO_SEQUENCE,
path_buffer,bufferlen);
if(ret != ZOK){
fprintf(stderr,"failed to create the path %s!\n",path);
}else{
printf("create path %s successfully!\n",path);
}
return ret;
}
pop的功能则是取出{path}下序号最小的子节点，如果没有子节点，则返回-1.
int pop(zhandle_t *zkhandle,const char *path,char *element,int *len)
{
int i = 0;
struct String_vector children;
int ret = zoo_get_children(zkhandle,path,0,&children);
if(ret != ZOK){
fprintf(stderr,"failed to create the path %s!\n",path);
}else if (children.count == 0){
strcpy(element,"");
*len = 0;
ret = -1;
}else{
char *min = children.data[0];
for(i = 0; i < children.count; ++i){
printf("%s:%s\n",min,children.data[i]);
if(strcmp(min,children.data[i]) > 0){
min = children.data[i];
}
}
if(min != NULL){
char child_path[512]={0};
sprintf(child_path,"%s/%s",path,min);
ret = zoo_get(zkhandle,child_path,0,element,len,NULL);
if(ret != ZOK){
fprintf(stderr,"failed to get data of the path %s!\n",child_path);
}else{
ret = zoo_delete(zkhandle,child_path, -1);
if(ret != ZOK){
fprintf(stderr,"failed to delete the path %s!\n",child_path);
}
}
}
}
for(i = 0; i < children.count; ++i){
free(children.data[i]);
children.data[i] = NULL;
}
return ret;
}
最后，再来看看模拟队列操作的程序。和其他程序类似，它的选项有
-
-p:指定队列的路径 -
-m:指定操作是push还是pop -
-v:只在push时有用，用与指定要push的元素的值 -
-s:指定Zookeeper的服务器的ip：port.
如：
向队列/Queue中压人一个元素，元素的值为"Hello":
>myqueue -s 172.17.0.36:2181 -p /Queue -m push -v Hello
将队列/Queue弹出一个元素
>myqueue -s 172.17.0.36:2181 -p /Queue -m pop
最后附上完整的源代码：
#include<stdio.h>
#include<string.h>
#include<unistd.h>
#include"zookeeper.h"
#include"zookeeper_log.h"
char g_host[512]= "172.17.0.36:2181";
char g_path[512]= "/Queue";
char g_value[512]="msg";
enum MODE{PUSH_MODE,POP_MODE} g_mode;
void print_usage();
void get_option(int argc,const char* argv[]);
/**********unitl*********************/
void print_usage()
{
printf("Usage : [myqueue] [-h] [-m mode] [-p path ] [-v value][-s ip:port] \n");
printf(" -h Show help\n");
printf(" -p Queue path\n");
printf(" -m mode:push or pop\n");
printf(" -v the value you want to push\n");
printf(" -s zookeeper server ip:port\n");
printf("For example:\n");
printf(" push the message \"Hello\" into the queue Queue:\n");
printf(" >myqueue -s172.17.0.36:2181 -p /Queue -m push -v Hello\n");
printf(" pop one message from the queue Queue:\n");
printf(" >myqueue -s172.17.0.36:2181 -p /Queue -m pop\n");
}
void get_option(int argc,const char* argv[])
{
extern char *optarg;
int optch;
int dem = 1;
const char optstring[] = "hv:m:p:s:";
g_mode = PUSH_MODE;
while((optch = getopt(argc , (char * const *)argv , optstring)) != -1 )
{
switch( optch )
{
case 'h':
print_usage();
exit(-1);
case '?':
print_usage();
printf("unknown parameter: %c\n", optopt);
exit(-1);
case ':':
print_usage();
printf("need parameter: %c\n", optopt);
exit(-1);
case 'm':
if(strcasecmp(optarg,"push")==0){
g_mode = PUSH_MODE;
}else{
g_mode = POP_MODE;
}
break;
case 's':
strncpy(g_host,optarg,sizeof(g_host));
break;
case 'p':
strncpy(g_path,optarg,sizeof(g_path));
break;
case 'v':
strncpy(g_value,optarg,sizeof(g_value));
break;
default:
break;
}
}
}
int push(zhandle_t *zkhandle,const char *path,char *element)
{
char child_path[512] = {0};
char path_buffer[512] = {0};
int bufferlen = sizeof(path_buffer);
sprintf(child_path,"%s/queue-",path);
int ret = zoo_create(zkhandle,child_path,element,strlen(element),
&ZOO_OPEN_ACL_UNSAFE,ZOO_SEQUENCE,
path_buffer,bufferlen);
if(ret != ZOK){
fprintf(stderr,"failed to create the path %s!\n",path);
}else{
printf("create path %s successfully!\n",path);
}
return ret;
}
int pop(zhandle_t *zkhandle,const char *path,char *element,int *len)
{
int i = 0;
struct String_vector children;
int ret = zoo_get_children(zkhandle,path,0,&children);
if(ret != ZOK){
fprintf(stderr,"failed to create the path %s!\n",path);
}else if (children.count == 0){
strcpy(element,"");
*len = 0;
ret = -1;
}else{
char *min = children.data[0];
for(i = 0; i < children.count; ++i){
printf("%s:%s\n",min,children.data[i]);
if(strcmp(min,children.data[i]) > 0){
min = children.data[i];
}
}
if(min != NULL){
char child_path[512]={0};
sprintf(child_path,"%s/%s",path,min);
ret = zoo_get(zkhandle,child_path,0,element,len,NULL);
if(ret != ZOK){
fprintf(stderr,"failed to get data of the path %s!\n",child_path);
}else{
ret = zoo_delete(zkhandle,child_path, -1);
if(ret != ZOK){
fprintf(stderr,"failed to delete the path %s!\n",child_path);
}
}
}
}
for(i = 0; i < children.count; ++i){
free(children.data[i]);
children.data[i] = NULL;
}
return ret;
}
int front(zhandle_t *zkhandle,char *path,char *element,int *len)
{
int i = 0;
struct String_vector children;
int ret = zoo_get_children(zkhandle,path,0,&children);
if(ret != ZOK){
fprintf(stderr,"failed to create the path %s!\n",path);
}else if(children.count == 0){
strcpy(element,"");
*len = 0;
ret = -1;
}else{
char *min = NULL;
for(i = 0; i < children.count; ++i){
if(strcmp(min,children.data[i]) > 0){
min = children.data[i];
}
}
if(min != NULL){
char child_path[512]={0};
sprintf(child_path,"%s/%s",path,min);
ret = zoo_get(zkhandle,child_path,0,element,len,NULL);
if(ret != ZOK){
fprintf(stderr,"failed to get data of the path %s!\n",child_path);
}
}
}
for(i = 0; i < children.count; ++i){
free(children.data[i]);
children.data[i] = NULL;
}
return ret;
}
int main(int argc, const char *argv[])
{
int timeout = 30000;
char path_buffer[512];
int bufferlen=sizeof(path_buffer);
zoo_set_debug_level(ZOO_LOG_LEVEL_WARN); //设置日志级别,避免出现一些其他信息
get_option(argc,argv);
zhandle_t* zkhandle = zookeeper_init(g_host,NULL, timeout, 0, (char *)"lock Test", 0);
if (zkhandle ==NULL)
{
fprintf(stderr, "Error when connecting to zookeeper servers...\n");
exit(EXIT_FAILURE);
}
int ret = zoo_exists(zkhandle,g_path,0,NULL);
if(ret != ZOK){
ret = zoo_create(zkhandle,g_path,"1.0",strlen("1.0"),
&ZOO_OPEN_ACL_UNSAFE,0,
path_buffer,bufferlen);
if(ret != ZOK){
fprintf(stderr,"failed to create the path %s!\n",g_path);
}else{
printf("create path %s successfully!\n",g_path);
}
}
if(g_mode == PUSH_MODE){
push(zkhandle,g_path,g_value);
printf("push:%s\n",g_value);
}else{
int len = sizeof(g_value);
ret = pop(zkhandle,g_path,g_value,&len) ;
if(ret == ZOK){
printf("pop:%s\n",g_value);
}else if( ret == -1){
printf("queue is empty\n");
}
}
zookeeper_close(zkhandle);
return 0;
}
