一、错误起因
Active NameNode日志出现异常IPC‘s epoch [X] is less than the last promised epoch [X+1]，出现短期的双Active
我配置的ha自动切换，但是发现STandByNameNode是active，我强制手动切换了三次，STandByNameNode就无法访问了，估计是这个问题。
二.内部原因
【HDFS机制】：该问题属于hdfs对于脑列的异常保护，属于正常行为，不影响业务。
1）ZKFC1对NameNode1(Active)进行健康检查，因为长时间监控不到NN1的回复，认为该NameNode1不健康，主动释 放zk中的ActiveStandbyElectorLock，此时NN1还是active（因为zkfc与NameNode1连接异常，不能将其 shutdown）。
zkfc log：
2014-06-16 02:11:02,720 WARN org.apache.hadoop.ha.HealthMonitor: Transport-level exception trying to monitor health of NameNode at namenode01/172.21.248.14:9005: Call From namenode01/1 72.21.248.14 to namenode02:9005 failed on socket timeout exception: java.net.SocketTimeoutException: 45000 millis timeout while waiting for channel to be ready for read. ch : java.nio.channels.SocketChannel[co nnected local=/172.21.248.14:47271 remote=namenode01/172.21.248.14:9005]; For more details see: http://wiki.apache.org/hadoop/SocketTimeout 2014-06-16 02:12:12,825 WARN org.apache.hadoop.ha.FailoverController: Unable to gracefully make NameNode at namenode02/172.21.248.13:9005 standby (unable to connect) java.net.SocketTimeoutException: Call From namenode01/172.21.248.14 to namenode02:9005 failed on socket timeout exception: java.net.SocketTimeoutException: 5000 millis timeout while waiting for channel to be ready for read. ch : java.nio.channels.SocketChannel[connected local=/172.21.248.14:59156 remote=namenode02/172.21.248.13:9005]; For more details see: http://wiki.apache.org/hadoop/SocketTimeout
2）ZKFC2在zk中竞争到ActiveStandbyElectorLock，将NameNode2（原来的Standby）变成Active，同时会更新JN中的epoch使其+1。
3）NameNode1（原先的Active）再次去操作JournalNode的editlog时发现自己的epoch比JN的epoch小1，促使自己重启，成为Standby NameNode。
NN1 log:
2014-08-26 12:20:59,017 FATAL org.apache.hadoop.hdfs.server.namenode.FSEditLog: Error: flush failed for required journal (JournalAndStream(mgr=QJM to [10.1.1.107:8485, 192.10.1.208:8485, 192.10.1.209:8485], stream=QuorumOutputStream starting at txid 22795230)) org.apache.hadoop.hdfs.qjournal.client.QuorumException: Got too many exceptions to achieve quorum size 2/3. 3 exceptions thrown: 192.10.1.208:8485: IPC‘s epoch 115 is less than the last promised epoch 116
三.解决方案
可以在core-site.xml文件中修改ha.health-monitor.rpc-timeout.ms参数值，来扩大zkfc监控检查超时时间。
<property> <name>ha.health-monitor.rpc-timeout.ms</name> <value>180000</value> </property>
![【问题记录与解决】KeyError: ‘acc‘ plt.plot(N[150:], H.history[“acc“][150:], label=“train_acc“) # KeyError: ‘](https://ucc.alicdn.com/pic/developer-ecology/imggxscun33bi_01583d145c1d40cdb642133a270a22d0.jpg?x-oss-process=image/resize,h_160,m_lfit)
