AnsweredAssumed Answered

Hive table directory owned by mapr account

Question asked by sourcedelica on Jun 14, 2015
Latest reply on Jun 15, 2015 by Hao Zhu
Our team is new to Hadoop and our cluster is new too (so it might be a configuration issue).

When I create a table in Hive and the directory does not already exist, the directory gets created owned by the `mapr` account.  Then if I do a `INSERT INTO ... SELECT` it fails with

    MapReduce Total cumulative CPU time: 120 minutes 58 seconds 590 msec
    Ended Job = job_1434266640981_0007
    Loading data to table default.ttt (product=null, coupon=null)
    14:48:29,5737 ERROR JniCommon fs/client/fileclient/cc/jni_MapRClient.cc:1306
    Thread: 24678 mkdirs failed for /user/fikzdev/spp/product=FHLG20/coupon=4.0,
    error 13 Failed with exception copyFiles: error while checking/creating destination
    directory!!! FAILED: Execution Error, return code 1 from
    org.apache.hadoop.hive.ql.exec.MoveTask

Should `mapr` own the directory?  I have been able to work around it by creating the directory in HDFS first, but it's easy to forget and can be a lot of time wasted, as it doesn't give you the error until after it has done the map/reduce. 

Outcomes