1. 集群机器监控
这通常用于那种对集群中机器状态,机器在线率有较高要求的场景,能够快速对集群中机器变化作出响应。这样的场景中,往往有一个监控系统,实时检测集群机器是否存活。
利用ZooKeeper有两个特性(读可监控,临时节点),就可以实现一种集群机器存活性监控系统:
1. 客户端在节点 x 上注册一个Watcher,那么如果x的子节点变化了,会通知该客户端
2. 创建EPHEMERAL类型的节点,一旦客户端和服务器的会话结束或过期,那么该节点就会消失
利用这两个特性,可以分别实现对客服端的状态变化、上下线进行监控。
例如,监控系统在 /Monitor 节点上注册一个Watcher,以后每动态加机器,那么就往 /Monitor 下创建一个 EPHEMERAL类型的节点:/Monitor/{hostname}. 这样,监控系统就能够实时知道机器的增减情况,至于后续处理就是监控系统的业务了。
2. Master选举
在分布式环境中,有些业务逻辑只需要集群中的某一台机器进行执行,其他的机器可以共享这个结果,这样可以大大减少重复计算,提高性能,于是就需要进行master选举。
利用ZooKeeper的强一致性,能够保证在分布式高并发情况下节点创建的全局唯一性,即:同时有多个客户端请求创建 /currentMaster 节点,最终一定只有一个客户端请求能够创建成功。利用这个特性,就能很轻易的在分布式环境中进行集群选举了。
此外,也可以利用Zookeeper的EPHEMERAL_SEQUENTIAL节点,实现动态选举:每个客户端都在/Master/下创建一个EPHEMERAL_SEQUENTIAL节点,由于ZooKeeper保证SEQUENTIAL的有序性,因此我们可以简单的把节点号最小的作为Master,就完成了选主。
3. 场景分析
假设我们要监控集群中的一群活动的业务进程,同时会在这群进程中选取一个进程作为监控的Master进程。每个进程使用IP地址加进程号标识,即{ip:pid}.当新的业务进程上线时,该进程会到/Monitor下创建一个临时有序(EPHEMERAL_SEQUENTIAL)的节点.并获取/Monitor下的子节点列表,如果发现自己创建的节点最小,则提升自己为Master进程,否则仍是业务进程。当进程退出时该节点会自动删除,其他进程则会尝试选主,保证当Master进程退出后,会提升一个新的Master进程。
举个例子,假设集群中一开始没有进程,
- 进程A1被创建,在/Monitor创建/Monitor/proc-1路径,由于/Monitor下只有一个路径,A1被提升为Master进程。
- 进程A2被创建,在/Monitor创建/Monitor/proc-2路径,选主不成功,作为Slave进程;同时A1监控/Monitor的子节点变化事件,会收到有新进程被创建 ,因此执行show_list。
- 进程A2被创建,在/Monitor创建/Monitor/proc-3路径,选主不成功,作为Slave进程;同时A1监控/Monitor的子节点变化事件,会收到有新进程被创建 ,因此执行show_list。
- 进程A1被Killed掉,其他进程监控到/Monitor的子节点变化事件,尝试选主,只有A2序号成功,因此A2选主成功,A3作为Slave进程。
- 进程A4被创建,在/Monitor创建/Monitor/proc-4路径,选主不成功,作为Slave进程;同时A2监控/Monitor的子节点变化事件,会收到有新进程被创建 ,因此执行show_list。
执行情况如下表所示:
A1 | A2 | A3 | A4 |
---|---|---|---|
create,show_list(M) | |||
show_list(M) | create | ||
show_list(M) | - | create | |
killed | show_list(M) | - | |
- | show_list(M) | - | create |
4. 动手实践
首先是获取本机的IP已经当前进程的进程号PID,并通过ip_pid返回。
void getlocalhost(char *ip_pid,int len) { char hostname[64] = {0}; struct hostent *hent ; gethostname(hostname,sizeof(hostname)); hent = gethostbyname(hostname); char * localhost = inet_ntoa(*((struct in_addr*)(hent->h_addr_list[0]))); snprintf(ip_pid,len,"%s:%lld",localhost,getpid()); }
选主函数,获取path下的所有子节点,选择序号最小的一个,取出它的ip_pid,如果和本进程相同,则本进程被选为Master。如果当前进程被选为Master,则进程中的全局变量g_mode会被赋值为MODE_MONITOR,否则不变。
void choose_mater(zhandle_t *zkhandle,const char *path) { struct String_vector procs; int i = 0; int ret = zoo_get_children(zkhandle,path,1,&procs); if(ret != ZOK || procs.count == 0){ fprintf(stderr,"failed to get the children of path %s!\n",path); }else{ char master_path[512] ={0}; char ip_pid[64] = {0}; int ip_pid_len = sizeof(ip_pid); char master[512]={0}; char localhost[512]={0}; getlocalhost(localhost,sizeof(localhost)); strcpy(master,procs.data[0]); for(i = 1; i < procs.count; ++i){ if(strcmp(master,procs.data[i])>0){ strcpy(master,procs.data[i]); } } sprintf(master_path,"%s/%s",path,master); ret = zoo_get(zkhandle,master_path,0,ip_pid,&ip_pid_len,NULL); if(ret != ZOK){ fprintf(stderr,"failed to get the data of path %s!\n",master_path); }else if(strcmp(ip_pid,localhost)==0){ g_mode = MODE_MONITOR; } } for(i = 0; i < procs.count; ++i){ free(procs.data[i]); procs.data[i] = NULL; } }
show_list为Master进程函数,所做的任务为打印path目录下所有子节点的ip_pid.
void show_list(zhandle_t *zkhandle,const char *path) { struct String_vector procs; int i = 0; char localhost[512]={0}; getlocalhost(localhost,sizeof(localhost)); int ret = zoo_get_children(zkhandle,path,1,&procs); if(ret != ZOK){ fprintf(stderr,"failed to get the children of path %s!\n",path); }else{ char child_path[512] ={0}; char ip_pid[64] = {0}; int ip_pid_len = sizeof(ip_pid); printf("--------------\n"); printf("ip\tpid\n"); for(i = 0; i < procs.count; ++i){ sprintf(child_path,"%s/%s",path,procs.data[i]); //printf("%s\n",child_path); ret = zoo_get(zkhandle,child_path,0,ip_pid,&ip_pid_len,NULL); if(ret != ZOK){ fprintf(stderr,"failed to get the data of path %s!\n",child_path); }else if(strcmp(ip_pid,localhost)==0){ printf("%s(Master)\n",ip_pid); }else{ printf("%s\n",ip_pid); } } } for(i = 0; i < procs.count; ++i){ free(procs.data[i]); procs.data[i] = NULL; } }
监控函数如下,当发现path的子节点发生变化,就会尝试重新选主,如果当前进程被选为主,就立即执行show_list,打印path下的所有子节点对应的ip_pid.
void zktest_watcher_g(zhandle_t* zh, int type, int state, const char* path, void* watcherCtx) { /* printf("watcher event\n"); printf("type: %d\n", type); printf("state: %d\n", state); printf("path: %s\n", path); printf("watcherCtx: %s\n", (char *)watcherCtx); */ if(type == ZOO_CHILD_EVENT && state == ZOO_CONNECTED_STATE ){ choose_mater(zh,path); if(g_mode == MODE_MONITOR){ show_list(zh,path); } } }
完整代码如下:
1.monitor.c
#include<stdio.h> #include<string.h> #include<unistd.h> #include <netinet/in.h> #include <netdb.h> #include <arpa/inet.h> #include"zookeeper.h" #include"zookeeper_log.h" enum WORK_MODE{MODE_MONITOR,MODE_WORKER} g_mode; char g_host[512]= "172.17.0.36:2181"; //watch function when child list changed void zktest_watcher_g(zhandle_t* zh, int type, int state, const char* path, void* watcherCtx); //show all process ip:pid void show_list(zhandle_t *zkhandle,const char *path); //if success,the g_mode will become MODE_MONITOR void choose_mater(zhandle_t *zkhandle,const char *path); //get localhost ip:pid void getlocalhost(char *ip_pid,int len); void print_usage(); void get_option(int argc,const char* argv[]); /**********unitl*********************/ void print_usage() { printf("Usage : [monitor] [-h] [-m] [-s ip:port] \n"); printf(" -h Show help\n"); printf(" -m set monitor mode\n"); printf(" -s zookeeper server ip:port\n"); printf("For example:\n"); printf("monitor -m -s172.17.0.36:2181 \n"); } void get_option(int argc,const char* argv[]) { extern char *optarg; int optch; int dem = 1; const char optstring[] = "hms:"; //default g_mode = MODE_WORKER; 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': g_mode = MODE_MONITOR; break; case 's': strncpy(g_host,optarg,sizeof(g_host)); break; default: break; } } } void zktest_watcher_g(zhandle_t* zh, int type, int state, const char* path, void* watcherCtx) { /* printf("watcher event\n"); printf("type: %d\n", type); printf("state: %d\n", state); printf("path: %s\n", path); printf("watcherCtx: %s\n", (char *)watcherCtx); */ if(type == ZOO_CHILD_EVENT && state == ZOO_CONNECTED_STATE ){ choose_mater(zh,path); if(g_mode == MODE_MONITOR){ show_list(zh,path); } } } void getlocalhost(char *ip_pid,int len) { char hostname[64] = {0}; struct hostent *hent ; gethostname(hostname,sizeof(hostname)); hent = gethostbyname(hostname); char * localhost = inet_ntoa(*((struct in_addr*)(hent->h_addr_list[0]))); snprintf(ip_pid,len,"%s:%lld",localhost,getpid()); } void choose_mater(zhandle_t *zkhandle,const char *path) { struct String_vector procs; int i = 0; int ret = zoo_get_children(zkhandle,path,1,&procs); if(ret != ZOK || procs.count == 0){ fprintf(stderr,"failed to get the children of path %s!\n",path); }else{ char master_path[512] ={0}; char ip_pid[64] = {0}; int ip_pid_len = sizeof(ip_pid); char master[512]={0}; char localhost[512]={0}; getlocalhost(localhost,sizeof(localhost)); strcpy(master,procs.data[0]); for(i = 1; i < procs.count; ++i){ if(strcmp(master,procs.data[i])>0){ strcpy(master,procs.data[i]); } } sprintf(master_path,"%s/%s",path,master); ret = zoo_get(zkhandle,master_path,0,ip_pid,&ip_pid_len,NULL); if(ret != ZOK){ fprintf(stderr,"failed to get the data of path %s!\n",master_path); }else if(strcmp(ip_pid,localhost)==0){ g_mode = MODE_MONITOR; } } for(i = 0; i < procs.count; ++i){ free(procs.data[i]); procs.data[i] = NULL; } } void show_list(zhandle_t *zkhandle,const char *path) { struct String_vector procs; int i = 0; char localhost[512]={0}; getlocalhost(localhost,sizeof(localhost)); int ret = zoo_get_children(zkhandle,path,1,&procs); if(ret != ZOK){ fprintf(stderr,"failed to get the children of path %s!\n",path); }else{ char child_path[512] ={0}; char ip_pid[64] = {0}; int ip_pid_len = sizeof(ip_pid); printf("--------------\n"); printf("ip\tpid\n"); for(i = 0; i < procs.count; ++i){ sprintf(child_path,"%s/%s",path,procs.data[i]); //printf("%s\n",child_path); ret = zoo_get(zkhandle,child_path,0,ip_pid,&ip_pid_len,NULL); if(ret != ZOK){ fprintf(stderr,"failed to get the data of path %s!\n",child_path); }else if(strcmp(ip_pid,localhost)==0){ printf("%s(Master)\n",ip_pid); }else{ printf("%s\n",ip_pid); } } } for(i = 0; i < procs.count; ++i){ free(procs.data[i]); procs.data[i] = NULL; } } 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,zktest_watcher_g, timeout, 0, (char *)"Monitor Test", 0); if (zkhandle ==NULL) { fprintf(stderr, "Error when connecting to zookeeper servers...\n"); exit(EXIT_FAILURE); } char path[512]="/Monitor"; int ret = zoo_exists(zkhandle,path,0,NULL); if(ret != ZOK){ ret = zoo_create(zkhandle,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",path); }else{ printf("create path %s successfully!\n",path); } } if(ret == ZOK && g_mode == MODE_WORKER){ char localhost[512]={0}; getlocalhost(localhost,sizeof(localhost)); char child_path[512]; sprintf(child_path,"%s/proc-",path); ret = zoo_create(zkhandle,child_path,localhost,strlen(localhost), &ZOO_OPEN_ACL_UNSAFE,ZOO_SEQUENCE|ZOO_EPHEMERAL, path_buffer,bufferlen); if(ret != ZOK){ fprintf(stderr,"failed to create the child_path %s,buffer:%s!\n",child_path,path_buffer); }else{ printf("create child path %s successfully!\n",path_buffer); } choose_mater(zkhandle,path); } if(g_mode == MODE_MONITOR){ show_list(zkhandle,path); } getchar(); zookeeper_close(zkhandle); return 0; }
2.Makefile
CC=gcc CFLAGS=-g ZOOKEEPER_INSTALL=/usr/local ZOOKEEPER_INC=-I${ZOOKEEPER_INSTALL}/include/zookeeper ZOOKEEPER_LIB= -L${ZOOKEEPER_INSTALL}/lib -lzookeeper_mt APP=monitor all: ${CC} monitor.c -DTHREAD ${CFLAGS} ${ZOOKEEPER_INC} ${ZOOKEEPER_LIB} -o ${APP} clean: rm -f ${APP}
可以单机上重复启动程序,它们的进程号都是不同的,也可以在集群中启动程序。
参数-s
表示Zookeeper的服务器的ip和端口,(注意不要理解成master的ip和端口哦)
参数-m
表示该进程是一个独立的监控进程,注意,指定这个参数的进程是不参加选主的,因为它不会在/Monitor目录下创建路径。
运行示例:monitor -s172.17.0.36:2181