For the original error, it appears I misread the stack trace the first time. Hive appears to be wanting to write to the local filesystem (on the NodeManager the task executes on), instead of HDFS, for some part of its work: > Caused by: java.io.IOException: Mkdirs failed to create file:/tmp/training/hive_2015-12-10_08-07-28_115_8039040536647708382/_task_tmp.-ext-10001 > at org.apache.hadoop.fs.ChecksumFileSystem.create(ChecksumFileSystem.java:434) Thereby, can you ensure the local /tmp directory exists on all your cluster host root filesystems with the drwxrwxrwt permissions? Also try clearing out local directory /tmp/training from every host and re-run the query. > instead the job is just going in pending state If you notice your RM screenshot, it tells there are 0 active nodes. This means your NodeManager is unavailable/dead/not-started, and the RM has no resources to allocate to (thereby the hang in PENDING state, as it is waiting for some NodeManager to come along and satisfy the requested resources of the application). You may want to restart NodeManager services, and/or check its logs if its gone down for some FATAL reason.
... View more
