版本问题的坑

半夜手贱，调整了下集群配置，发现yarn挂了，nodemanager一直起不来，查了下log是个null pointer，没看出来撒，结果google到这玩意。
https://issues.apache.org/jira/browse/YARN-2816
然后又看到这个 https://sskaje.me/2014/11/yarn-nodemanager-failed-start/
原来。。。

And, in the start-up message part,

2014-10-30 21:23:07,141 INFO org.apache.hadoop.yarn.server.nodemanager.NodeManager: registered UNIX signal handlers for [TERM, HUP, INT]
2014-10-30 21:23:08,259 INFO org.apache.hadoop.yarn.server.nodemanager.recovery.NMLeveldbStateStoreService: Using state database at /tmp/hadoop-yarn/yarn-nm-recovery/yarn-nm-state for recover
2014-10-30 21:23:08,291 INFO org.apache.hadoop.yarn.server.nodemanager.recovery.NMLeveldbStateStoreService$LeveldbLogger: Recovering log #432
2014-10-30 21:23:08,309 INFO org.apache.hadoop.yarn.server.nodemanager.recovery.NMLeveldbStateStoreService$LeveldbLogger: Delete type=0 #432

2014-10-30 21:23:08,309 INFO org.apache.hadoop.yarn.server.nodemanager.recovery.NMLeveldbStateStoreService$LeveldbLogger: Delete type=3 #431

2014-10-30 21:23:08,321 INFO org.apache.hadoop.yarn.server.nodemanager.recovery.NMLeveldbStateStoreService: Loaded NM state version info 1.0

The solution is, stop the instance, delete ‘/tmp/hadoop-yarn/’ from local filesystem, start the instance.
将每个namenode下的这个目录都删除后，终于恢复了，可以睡觉了。。。