昨天同事遇到一个hadoop故障，找了半天没看出问题，问到我这里，花了一会解决了一下，估计这是我给暴风的集群解决的最后的故障了，以后就不定给谁解决问题去了。


只截下来了Namenode的报错Log，Datanode的刷屏刷过去了，不过都差不多。

2013-09-03 18:11:44,021 WARN org.apache.hadoop.hdfs.StateChange: BLOCK* NameSystem.blockReceived: blk_8094241928859719036_2147969 is received from dead or unregistered node 192.168.1.99:50010
2013-09-03 18:11:44,022 ERROR org.apache.hadoop.security.UserGroupInformation: PriviledgedActionException as:hdfs cause:java.io.IOException: Got blockReceived message from unregistered or dead node blk_8094241928859719036_2147969
2013-09-03 18:11:44,022 INFO org.apache.hadoop.ipc.Server: IPC Server handler 2 on 9000, call blockReceived(DatanodeRegistration(192.168.1.99:50010, storageID=DS-1925877777-192.168.1.99-50010-1372745739682, infoPort=50075, ipcPort=50020), [Lorg.apache.hadoop.hdfs.protocol.Block;@4ec371c, [Ljava.lang.String;@301611ca) from 192.168.1.99:18853: error: java.io.IOException: Got blockReceived message from unregistered or dead node blk_8094241928859719036_2147969
java.io.IOException: Got blockReceived message from unregistered or dead node blk_8094241928859719036_2147969
        at org.apache.hadoop.hdfs.server.namenode.FSNamesystem.blockReceived(FSNamesystem.java:4188)
        at org.apache.hadoop.hdfs.server.namenode.NameNode.blockReceived(NameNode.java:1069)
        at sun.reflect.GeneratedMethodAccessor13.invoke(Unknown Source)
        at sun.reflect.DelegatingMethodAccessorImpl.invoke(Unknown Source)
        at java.lang.reflect.Method.invoke(Unknown Source)
        at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:578)
        at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1393)
        at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1389)
        at java.security.AccessController.doPrivileged(Native Method)
        at javax.security.auth.Subject.doAs(Unknown Source)
        at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1149)
        at org.apache.hadoop.ipc.Server$Handler.run(Server.java:1387)



看上去是个IPC的错误，从下往上看，都是报权限错误，然后无法注册Datanode，还有从未注册或死亡的Datanode上报了一个块已被接收的错误。同事就晕了，已经死亡的node怎么还上报啊。


然后重启datanode时间不长，就又挂掉了。


登录到datanode，先看了一下dfs的数据文件夹的权限，正确无误。然后看了一下df -h，发现/var文件夹满了，OPS很缺，只给分了20G的/var。结果Hadoop的log写不进去了，自然就挂了。删掉/var/log/hadoop/hdfs里面的历史日志，datanode启动正常。以后的解决办法只有两个，要么设置定时脚本每天删历史日志，要么就把/var/log/hadoop/hdfs文件夹软链到一个比较大的硬盘上。


马上要离开暴风影音了，心中有诸多槽点，以后慢慢吐了。