@Vinay Reddy NaguruSaprk is far ahead in terms of perfomance but it still need to address some of the concern like memory management. for some use case mapreduce is preferred over spark e.g. ETL calculations where result sets are vast and may exceed total RAM of hadoop cluster,mapreduce can out perform Spark for this situation.Iterative machine learning where spark is not able to manage memory more proficiently is ideal use case for mapreduce. but spark is evolving so fast and trying address these concerns so i think it is not very distant when spark will replace MR completely but for now they can coexists in the cluster. ... View more

seems you dont have delete on alert_notice table can you check it with mysql in mysql prompt use information_schema select * from TABLE_PRIVILEGES; ... View more

try increasing nimbus.thrift.max_buffer_size from default 1MB to 2-5M and see if you are able to see expected.nimbus.thrift.max_buffer_size is the maximum buffer size thrift should use when reading messages. ... View more

it seems that you are hitting https://issues.apache.org/jira/browse/HIVE-12349 can you try to run your query after setting hive.optimize.index.filter=false ... View more