分布式EHCACHE系统在缓存同步上存在着不小的缺陷
2012-08-29 15:10阅读：
分布式EHCACHE系统,如何实现缓存数据同步?
这也是最常用的方式,配置简单
,关键一点,各EHCACHE的节点配置都是一样的
例子:
spring 配置中调用的ehcache文件
<bean id='cacheManager'
class='org.springframework.cache.ehcache.EhCacheManagerFactoryBean'>
<property name='configLocation'>
<value>classpath:ehcache_mc.xml</value>
</property>
</bean>
<bean id='userCache'
class='org.springframework.cache.ehcache.EhCacheFactoryBean'>
<property name='cacheManager'>
<ref
local='cacheManager'/>
</property>
<property name='cacheName'>
<value>userCache</value>
&l |
t;/property>
</bean>
ehcache_mc.xml:
<cache name='userCache'
maxElementsInMemory='10000'
maxElementsOnDisk='0'
eternal='true'
overflowToDisk='true'
diskSpoolBufferSizeMB='20'
memoryStoreEvictionPolicy='LFU'
transactionalMode='off'>
<cacheEventListenerFactory
class='net.sf.ehcache.distribution.RMICacheReplicatorFactory'
properties='replicateAsynchronously=true, replicatePuts=true,
replicateUpdates=true,replicateUpdatesViaCopy= true,
replicateRemovals= true ' />
<cacheEventListenerFactory
class='cn.com.gary.test.ehcache.EventFactory' /> <!--
打印ehcache 动作日志 -->
</cache>
<cacheManagerPeerProviderFactory
class='net.sf.ehcache.distribution.RMICacheManagerPeerProviderFactory'
properties='peerDiscovery=automatic,
multicastGroupAddress=230.0.0.1,
multicastGroupPort=4446,timeToLive=255'/>
<cacheManagerPeerListenerFactory
class='net.sf.ehcache.distribution.RMICacheManagerPeerListenerFactory'/>
|
原理:
这样当缓存改变时,ehcache会向230.0.0.1端口4446发RMI UDP组播包
这种组播方式的缺陷:
EHCACHE的组播做得比较初级,功能只是基本实现(比如简单的一个HUB,接两台单网卡的服务器,互相之间组播同步就没问题),
对一些复杂的环境(比如
多台服务器,每台服务器上多地址,尤其是集群,存在
一个集群地址带多个物理机,每台物理机又带多个虚拟站的子地址),就容易出现问题.
究其原因, 组播/广播转发是一个很复杂的过程. 简单的说,
一个组播缺省只能在一个网段内传输,不能跨网段.
举个简单的例子,
PC机网卡的自动获取地址,还有WINDOWS里的网上邻居,都属于典型的广播服务,所以这些服务都是不能跨网段(跨路由)的,当然也不是完全不行,借助一些工具,比如CISCO路由器上的udp-broadcast
helper,或者微软的netBIOS on Tcp/ip,就可以实现.
我们自己安装一些软件时,也经常遇到比如'将网卡的广播转发打开'之类的操作.
而在多网卡的主机,或同一网卡多IP的主机上,尽管地址可能是一个网段内的,但其实地址间已经存在跳数了(hop),其实就是从一个地址向另一个地址跳.
这时广播/组播就容易被阻断.
比如:
我们自己的WINDOWS上装一个VMWARE虚拟机,尽管IP地址是一个网段的,但因为虚拟机采用的桥模式不是标准的网桥模式(也可能是需要配置一下,但说实话懒得研究VMWARE了),所以广播/组播也经常出现不通的情况.
更何况在一些云计算的环境,集群的分布往往是跨网段的,甚至是跨地域的.这时更难以依赖这种初级的组播同步.
总之,分布式集群架构,建议EHCACHE改为PEER-2-PEER的同步方式.
其实就是每个节点和其他
n-1个节点都建立TCP的P2P PEER.
下面是一个3节点的ehcache分布式部署:
节点1:
spring 配置中调用的ehcache文件
<bean id='cacheManager'
class='org.springframework.cache.ehcache.EhCacheManagerFactoryBean'>
<property name='configLocation'>
<value>classpath:ehcache_p2p_40001.xml</value>
</property>
</bean>
<bean id='userCache'
class='org.springframework.cache.ehcache.EhCacheFactoryBean'>
<property name='cacheManager'>
<ref
local='cacheManager'/>
</property>
<property name='cacheName'>
<value>userCache</value>
</property>
</bean>
ehcache_p2p_40001.xml
<cache name='userCache'
maxElementsInMemory='10000'
maxElementsOnDisk='0'
eternal='true'
overflowToDisk='true'
diskSpoolBufferSizeMB='20'
memoryStoreEvictionPolicy='LFU'
transactionalMode='off'>
<cacheEventListenerFactory
class='net.sf.ehcache.distribution.RMICacheReplicatorFactory'
properties='replicateAsynchronously=true, replicatePuts=true,
replicateUpdates=true,replicateUpdatesViaCopy= true,
replicateRemovals= true ' />
<cacheEventListenerFactory
class='cn.com.dwsoft.test.ehcache.EventFactory' />
</cache>
<!--调用ehcache2的RMI-->
<cacheManagerPeerProviderFactory
class='net.sf.ehcache.distribution.RMICacheManagerPeerProviderFactory'
properties='peerDiscovery=manual,rmiUrls=//192.168.0.251:40002/userCache|//192.168.0.251:40003/userCache'/>
<!--RMI监听40001端口-->
<cacheManagerPeerListenerFactory
class='net.sf.ehcache.distribution.RMICacheManagerPeerListenerFactory'
properties='hostName=192.168.0.122,port=40001,socketTimeoutMillis=2000'/>
|
节点2:
spring 配置中调用的ehcache文件
<bean id='cacheManager'
class='org.springframework.cache.ehcache.EhCacheManagerFactoryBean'>
<property name='configLocation'>
<value>classpath:ehcache_p2p_40002.xml</value>
</property>
</bean>
<bean id='userCache'
class='org.springframework.cache.ehcache.EhCacheFactoryBean'>
<property name='cacheManager'>
<ref
local='cacheManager'/>
</property>
<property name='cacheName'>
<value>userCache</value>
</property>
</bean>
ehcache_p2p_40002.xml
<cache name='userCache'
maxElementsInMemory='10000'
maxElementsOnDisk='0'
eternal='true'
overflowToDisk='true'
diskSpoolBufferSizeMB='20'
memoryStoreEvictionPolicy='LFU'
transactionalMode='off'>
<cacheEventListenerFactory
class='net.sf.ehcache.distribution.RMICacheReplicatorFactory'
properties='replicateAsynchronously=true, replicatePuts=true,
replicateUpdates=true,replicateUpdatesViaCopy= true,
replicateRemovals= true ' />
<cacheEventListenerFactory
class='cn.com.dwsoft.test.ehcache.EventFactory' />
</cache>
<!--调用ehcache2的RMI-->
<cacheManagerPeerProviderFactory
class='net.sf.ehcache.distribution.RMICacheManagerPeerProviderFactory'
properties='peerDiscovery=manual,rmiUrls=//192.168.0.122:40001/userCache|//192.168.0.251:40003/userCache'/>
<!--RMI监听40001端口-->
<cacheManagerPeerListenerFactory
class='net.sf.ehcache.distribution.RMICacheManagerPeerListenerFactory'
properties='hostName=192.168.0.251,port=40002,socketTimeoutMillis=2000'/>
|
节点3:
spring 配置中调用的ehcache文件
<bean id='cacheManager'
class='org.springframework.cache.ehcache.EhCacheManagerFactoryBean'>
<property name='configLocation'>
<value>classpath:ehcache_p2p_40003.xml</value>
</property>
</bean>
<bean id='userCache'
class='org.springframework.cache.ehcache.EhCacheFactoryBean'>
<property name='cacheManager'>
<ref
local='cacheManager'/>
</property>
<property name='cacheName'>
<value>userCache</value>
</property>
</bean>
ehcache_p2p_40003.xml
<cache name='userCache'
maxElementsInMemory='10000'
maxElementsOnDisk='0'
eternal='true'
overflowToDisk='true'
diskSpoolBufferSizeMB='20'
memoryStoreEvictionPolicy='LFU'
transactionalMode='off'>
<cacheEventListenerFactory
class='net.sf.ehcache.distribution.RMICacheReplicatorFactory'
properties='replicateAsynchronously=true, replicatePuts=true,
replicateUpdates=true,replicateUpdatesViaCopy= true,
replicateRemovals= true ' />
<cacheEventListenerFactory
class='cn.com.dwsoft.test.ehcache.EventFactory' />
</cache>
<!--调用ehcache2的RMI-->
<cacheManagerPeerProviderFactory
class='net.sf.ehcache.distribution.RMICacheManagerPeerProviderFactory'
properties='peerDiscovery=manual,rmiUrls=//192.168.0.122:40001/userCache|//192.168.0.251:40002/userCache'/>
<!--RMI监听40001端口-->
<cacheManagerPeerListenerFactory
class='net.sf.ehcache.distribution.RMICacheManagerPeerListenerFactory'
properties='hostName=192.168.0.251,port=40003,socketTimeoutMillis=2000'/>
|
测试
节点1添加一条缓存条目
节点1日志:
2012-08-28 17:00:39,859 INFO
CustomActionSupport.getParameter(188)-/admin/inputEntry.do?
submit=提交&cachekey=mac1228&cachevalue=1228
2012-08-28 17:00:39,859 INFO
TestAction.inputEntry(97)-mac1228:1228
2012-08-28 17:00:39,875 INFO CacheEvent.log(65)-in
notifyElementPut[ key = mac1228, value=1228, version=1, hitCount=0,
CreationTime = 1346144439875, LastAccessTime = 1346144439875
]
2012-08-28 17:00:39,875DEBUG
ServletDispatcherResult.debug(57)-Forwarding to location /index.jsp
|
节点2日志:
2011-09-22 14:25:51,262 INFO CacheEvent.log(65)-in
notifyElementPut[ key = mac1228, value=1228, version=1, hitCount=0,
CreationTime = 1346144440000, LastAccessTime = 1316672751253 ]
|
节点3日志:
2011-09-22 14:25:51,198 INFO CacheEvent.log(65)-in
notifyElementPut[ key = mac1228, value=1228, version=1, hitCount=0,
CreationTime = 1346144440000, LastAccessTime = 1316672751198 ]
|
|
节点2添加一条缓存条目
节点1日志:
2012-08-28 16:54:55,890 INFO CacheEvent.log(65)-in
notifyElementPut[ key = mac25181904, value=25181904, version=1,
hitCount=0, CreationTime = 1316672407000, LastAccessTime =
1346144095890 ] |
节点2日志:
2011-09-22 14:20:06,041 INFO
CustomActionSupport.getParameter(188)-/admin/inputEntry.do?
submit=提交&cachekey=mac25181904&cachevalue=25181904
2011-09-22 14:20:06,042 INFO
TestAction.inputEntry(97)-mac25181904:25181904
2011-09-22 14:20:06,050DEBUG
ServletDispatcherResult.debug(57)-Forwarding to location /index.jsp
|
节点3日志:
2011-09-22 14:20:06,568 INFO CacheEvent.log(65)-in
notifyElementPut[ key = mac25181904, value=25181904, version=1,
hitCount=0, CreationTime = 1316672407000, LastAccessTime =
1316672406568 ] |
|
总结:
上面说了,
组播方式同步不可靠.
P2P方式其实也存在不可靠的地方.这就是P2P要求每个节点的EHCACHE要指向其他的N-1个节点,
当在
云环境,或集群域下, 多个子节点部署项目都是被自动发布的,这时
很难做到不同节点有不同的配置,因为自动发布,配置往往都是相同的,这样P2P就很难实现.
总之,
这种同步型应用是很难适应大规模分布式部署的,还是建议采用一些集中软件比如MEMCACHED.
