Note that this method will always return TRUE because a connection is not actually made at call time. See this bug report for more information: https://bugs.php.net/bug.php?id=58193
(PECL memcache >= 2.0.0)
Memcache::addServer — 向连接池中添加一个memcache服务器
$host
[, int $port = 11211
[, bool $persistent
[, int $weight
[, int $timeout
[, int $retry_interval
[, bool $status
[, callback $failure_callback
[, int $timeoutms
]]]]]]]] )Memcache::addServer()增加一个服务器到连接池中。通过Memcache::addServer() 打开的连接将会在脚本执行结束后自动关闭，也可以使用Memcache::close()进行手动关闭。 您也可以使用memcache_add_server()来添加服务器。
当使用这个方法的时候(与Memcache::connect()和Memcache::pconnect()相反) 网络连接并不会立刻建立，而是直到真正使用的时候才建立。 因此在加入大量服务器到连接池中时也是没有开销的，因为它们可能并不会被使用。
故障转移可能在方法的任何一个层次发生，通常只要其他服务器可用用户就不会感受到。任何的socket或memcache服务器级别的错误 （比如内存溢出）都可能导致故障转移。而一般的客户端错误比如使用Memcache::add尝试增加一个已经存在的key则不会导致故障转移。
Note:
这个方法在2.0.0版本加入Memcache。
host
要连接的memcached服务端监听的主机位置。这个参数通常指定其他类型的传输比如Unix域套接字使用
unix:///path/to/memcached.sock，这种情况下参数port
必须设置为0。
port要连接的memcached服务端监听的端口。当使用UNIX域套接字连接时设置为0。
persistent
控制是否使用持久化连接。默认TRUE。
weight为此服务器创建的桶的数量，用来控制此服务器被选中的权重，单个服务器被选中的概率是相对于所有服务器weight总和而言的。
timeout连接持续（超时）时间（单位秒），默认值1秒，修改此值之前请三思，过长的连接持续时间可能会导致失去所有的缓存优势。
retry_interval
服务器连接失败时重试的间隔时间，默认值15秒。如果此参数设置为-1表示不重试。此参数和persistent参数在扩展以
dl()函数动态加载的时候无效。
每个失败的连接结构有自己的超时时间，并且在它失效之前选择后端服务请求时该结构会被跳过。一旦一个连接失效，
它将会被成功重新连接或被标记为失败连接以在下一个retry_interval秒重连。
典型的影响是每个web服务子进程在服务于一个页面时将会每retry_interval秒
尝试重新连接一次。
status
控制此服务器是否可以被标记为在线状态。设置此参数值为FALSE并且retry_interval参数
设置为-1时允许将失败的服务器保留在一个池中以免影响key的分配算法。对于这个服务器的请求会进行故障转移或者立即失败，
这受限于memcache.allow_failover参数的设置。该参数默认TRUE，表明允许进行故障转移。
failure_callback允许用户指定一个运行时发生错误后的回调函数。回调函数会在故障转移之前运行。回调函数会接受到两个参数，分别是失败主机的 主机名和端口号。
timeoutms
成功时返回 TRUE， 或者在失败时返回 FALSE。
Example #1 Memcache::addServer() 示例
<?php
/* OO API */
$memcache = new Memcache;
$memcache->addServer('memcache_host', 11211);
$memcache->addServer('memcache_host2', 11211);
/* procedural API */
$memcache_obj = memcache_connect('memcache_host', 11211);
memcache_add_server($memcache_obj, 'memcache_host2', 11211);
?>
Note that this method will always return TRUE because a connection is not actually made at call time. See this bug report for more information: https://bugs.php.net/bug.php?id=58193
The $timeoutms argument can be used to specify the timeout in milliseconds, but isn't available in all versions. For example, it exists in php_memcache 2.2.6, but not in 3.0.4. In 2.2.6, if you specify it, then it overrides $timeout.
Caveat emptor: If $timeoutms is not specified, it defaults to the value of memcache.default_timeout_ms in php.ini, which defaults to 1000 if not set. This also overrides $timeout, which has the curious effect that $timeout is always ignored in php_memcache 2.2.6 (either in favor of $timeoutms, memcache.default_timeout_ms or the value 1000, in that order of priority).
The Memcache client library is responsible for picking the right server to set/get data. That's why addServer is what you want to use rather than connect, when you have more than one Memcache server. A subsequent set/get will then connect on demand to the appropriate instance as needs be. Disconnection to whatever servers were connected to happening when you close or your script terminates.
Memcache instances added to your Memcache object via addServer should be added in the same order in your application to ensure that the same server is picked for use with the same key.
A client library may be implemented to run a CRC on the key and do a modulus over the number of instances in the list to select an instance from the list for the set/get. This ensures that data is spread nicely across the nodes.
That all works nicely behind the scenes for you in your PHP code, as long as you add your list of Memcache instances in a consistent manner with addServer.
Actually if you have two memcached servers from which one of them is on localhost, and the other is on a remote machine you can communicate with both even if you specify the loopback address for the local one.
<?php
$memcache_obj = memcache_connect("127.0.0.1", 11211);
memcache_add_server($memcache_obj, "memcache_remote_host");
$memcache_obj->set('var_key', time());
?>
This WILL communicate with both hosts but however there are two aspects that must be taken into account:
1. the communication will be done through different network interfaces with the two hosts. It will use the loopback interface for the "127.0.0.1" host (lo in my case on Linux) and the external interface for the "memcache_remote_host" (eth0 in my case). Only if you want to use the same network interface to communicate with both hosts you must use the external IPs of both machines (and all communication will go out through the eth0 interface).
2. the connection with the two hosts will be established differently because of how memcache_connect() and memcache_add_server() work. Therefore the memcache_connect() will initiate the connection to localhost through the loopback interface when it is called, while memcache_add_server() will just add the second server to the pool, but it will not send any package through the network until it's absolutely needed (for example when a memcache_set() command is issued).
The weight of the server must be greater than 0.
If there is no memcached server to use, and you try to set/add variables, the apache will be crashed, with the error message "[notice] child pid 18725 exit signal Segmentation fault (11)" in error_log file.
The default value for the "weight" argument is 1