I ran with debug logging, and this is interesting, there was a loss of connection to the metastore client RIGHT before the partition mention above... as data was looking to be moved around... I wonder if the timing on that is bad?
14/09/09 12:47:37 [main]: INFO exec.MoveTask: Partition is: {day=null, source=null} 14/09/09 12:47:38 [main]: INFO metadata.Hive: Renaming src:maprfs:/user/hive/scratch/hive-mapr/hive_2014-09-09_12-38-30_860_3555291990145206535-1/-ext-10000/day=2012-11-30/source=20121119_SWAirlines_Spam/000004_0;dest: maprfs:/user/hive/warehouse/intel_flow.db/pcaps/day=2012-11-30/source=20121119_SWAirlines_Spam/000004_0;Status:true 14/09/09 12:48:02 [main]: WARN metastore.RetryingMetaStoreClient: MetaStoreClient lost connection. Attempting to reconnect. org.apache.thrift.transport.TTransportException: java.net.SocketTimeoutException: Read timed out at org.apache.thrift.transport.TIOStreamTransport.read(TIOStreamTransport.java:129) On Tue, Sep 9, 2014 at 11:02 AM, John Omernik <j...@omernik.com> wrote: > I am doing a dynamic partition load in Hive 0.13 using ORC files. This has > always worked in the past both with MapReduce V1 and YARN. I am working > with Mesos now, and trying to trouble shoot this weird error: > > > > Failed with exception AlreadyExistsException(message:Partition already > exists > > > > What's odd is is my insert is an insert (without Overwrite) so it's like > two different reducers have data to go into the same partition, but then > there is a collision of some sort? Perhaps there is a situation where the > partition doesn't exist prior to the run, but when two reducers have data, > they both think they should be the one to create the partition? Shouldn't > if a partition already exists, the reducer just copies it's file into the > partition? I am struggling to see why this would be an issue with Mesos, > but not on Yarn, or MRv1. > > > Any thoughts would be welcome. > > > John >