我有两台机器建了RAC，客户端配了负载均衡的TAF。关掉某一台机器的listener或者database，客户端连接都可以正常切换实例。但是当主机重启时，客户端就会连部上数据库。
我用网络监听软件观察一下，是由于客户端发出tcp连接请求后，一直在等待主机回应，但是主机关闭后，没有任何TCP回应，所以客户端一直等到tcp超时，就报告一个错误。
我的问题是这样我如果对RAC主机进行维护，不就影响业务运行了吗？RAC不会就是这样吧，有没有解决的办法呢？

从现象来看是IP没有切换,你用cluvfy stage -post crsinst all 检查

If you use vip instead of ip address, in your tnsname.ora file, you can avoid this problem.

you have two cases in here.

1) shutdown instance, but listener still up in server1
when client connections connect to listener in server1, listener in server 1 will relocate those connections to instance in server2.

Check remote_listener in init parameter file

2) shutdown intances and listener in server1, but vip still up.
vip will inform the incoming connection to connect to the second ip address in tnsname.ora file.

You can try to use outbound_connect_timeout in sqlnet.ora file in client side.