网游内存数据库的设计(2)

续第一篇，前两天对核心存储做了些修改，以前只打算与关系数据库的行与表做对应,value类型只能使array或list,

现在把7种基本类型也加入到value支持的类型中，以使得数据库更通用.

当然，这都不是本文的核心，本篇主要介绍一个测试前端，以及测试的远程调用协议.

先贴出测试前端的服务器代码:

#include "netservice.h"
#include "msg_loop.h"
#include "datasocket.h"
#include "SysTime.h"
#include "db_protocal.h"
atomic_32_t wpacket_count = 0;
atomic_32_t rpacket_count = 0;
atomic_32_t buf_count = 0;
global_table_t gtb;
void server_process_packet(datasocket_t s,rpacket_t r)
{
//执行操作并返回结果

cache_protocal_t p;
uint32_t coro_id = rpacket_read_uint32(r);
uint8_t type = rpacket_read_uint8(r);
switch(type)
{
case CACHE_GET:
p = create_get();
break;
case CACHE_SET:
p = create_set();
break;
case CACHE_DEL:
p = create_del();
break;
}
wpacket_t ret = p->execute(gtb,r,coro_id);
if(NULL != ret)
data_send(s,ret);
destroy_protocal(&p);
}
void process_new_connection(datasocket_t s)
{
printf("w:%u,r:%u,b:%un",wpacket_count,rpacket_count,buf_count);
}
void process_connection_disconnect(datasocket_t s,int32_t reason)
{
release_datasocket(&s);
printf("w:%u,r:%u,b:%un",wpacket_count,rpacket_count,buf_count);
}
void process_send_block(datasocket_t s)
{
//发送阻塞,直接关闭

close_datasocket(s);
}
const char *ip;
uint32_t port;
int main(int argc,char **argv)
{
init_net_service();
ip = argv[1];
port = atoi(argv[2]);
netservice_t n = create_net_service(1);
gtb = global_table_create(65536);
int32_t i = 0;
char key[64];
for( ; i < 1000000; ++i)
{
basetype_t a = basetype_create_int32(i);
snprintf(key,64,"test%d",i);
a = global_table_insert(gtb,key,a,global_hash(key));
if(!a)
printf("error 1n");
basetype_release(&a);
}
net_add_listener(n,ip,port);
msg_loop_t m = create_msg_loop(server_process_packet,process_new_connection,process_connection_disconnect,process_send_block);
while(1)
{
msg_loop_once(m,n,100);
}
return 0;
}

前端的网络模块使用了在上一篇中介绍的网络框架，启动时先插入100W条32位整型的记录,然后进入消息循环，不断的处理从客户端发过来的操作请求.

目前只添加了三个协议,分别是获取:CACHE_GET;添加/修改:CACHE_SET;删除:CACHE_DEL.

服务器处理协议并将结果返回给客户端.

然后是测试客户端:

#include "db_protocal.h"
#include "dbtype.h"
#include <stdio.h>
#include "SocketWrapper.h"
#include "SysTime.h"
#include "KendyNet.h"
#include "Connector.h"
#include "Connection.h"
#include "common_define.h"
#include "netservice.h"
#include "msg_loop.h"
#include "co_sche.h"
sche_t g_sche = NULL;
uint32_t call_count = 0;
atomic_32_t wpacket_count = 0;
atomic_32_t rpacket_count = 0;
atomic_32_t buf_count = 0;
datasocket_t db_s;
int8_t test_select(const char *key,int32_t i)
{
coro_t co = get_current_coro();
wpacket_t wpk = get_wpacket(64);
wpacket_write_uint32(wpk,(int32_t)co);
wpacket_write_uint8(wpk,CACHE_GET);//ÉèÖÃ

wpacket_write_string(wpk,key);
data_send(db_s,wpk);
coro_block(co);
int8_t ret = rpacket_read_uint8(co->rpc_response);
rpacket_read_uint8(co->rpc_response);
int32_t val = rpacket_read_uint32(co->rpc_response);
if(val != i)
printf("errorn");
//printf("beginn");
rpacket_destroy(&co->rpc_response);
//printf("endn");
return ret;
}
void *test_coro_fun2(void *arg)
{
coro_t co = get_current_coro();
while(1)
{
char key[64];
int32_t i = rand()%1000000;
snprintf(key,64,"test%d",100);
if(0 == test_select(key,100))
++call_count;
}
}
void server_process_packet(datasocket_t s,rpacket_t r)
{
coro_t co = (coro_t)rpacket_read_uint32(r);
co->rpc_response = rpacket_create_by_rpacket(r);
coro_wakeup(co);
}
void process_new_connection(datasocket_t s)
{
printf("connect servern");
db_s = s;
g_sche = sche_create(20000,65536,NULL,NULL);
int i = 0;
for(; i < 20000; ++i)
{
sche_spawn(g_sche,test_coro_fun2,NULL);
}
}
void process_connection_disconnect(datasocket_t s,int32_t reason)
{
release_datasocket(&s);
}
void process_send_block(datasocket_t s)
{
//·¢ËÍ×èÈû,Ö±½Ó¹Ø±Õ

close_datasocket(s);
}
int main(int argc,char **argv)
{
init_net_service();
const char *ip = argv[1];
uint32_t port = atoi(argv[2]);
netservice_t n = create_net_service(1);
net_connect(n,ip,port);
msg_loop_t m = create_msg_loop(server_process_packet,process_new_connection,process_connection_disconnect,process_send_block);
uint32_t tick = GetSystemMs();
while(1)
{
msg_loop_once(m,n,1);
uint32_t now = GetSystemMs();
if(now - tick > 1000)
{
printf("call_count:%un",(call_count*1000)/(now-tick));
tick = now;
call_count = 0;
}
if(g_sche)
sche_schedule(g_sche);
}
return 0;
}

操作接口使用用户级线程实现，以支持同步调用接口，用户级线程发出请求后就阻塞自己，直到结果返回时才被唤醒:

关键部分在test_select,把自己的coro地址作为id打包到协议中，发往服务器，然后调用coro_block阻塞。服务器返回的数据包

中也带了对应的coro_id,以通知客户端的调度系统该唤醒哪个coro.coro被唤醒后从结果包中读取操作结果和数据,返回给上层调用者.

从测试结果来看,启动1W个coro的客户端，每秒平均能执行50W次的操作。对于一个万人在线的MMORPG游戏来说应该已经是够用的了。

如果还是不够，可以通过表空间的划分，启动多个内存数据库进程来服务请求。

 项目地址:https://github.com/sniperHW/kendylib/tree/master/dbcache