Release 0.201#


  • Change grouped aggregations to use IS NOT DISTINCT FROM semantics rather than equality semantics. This fixes incorrect results and degraded performance when grouping on NaN floating point values, and adds support for grouping on structural types that contain nulls.

  • Fix planning error when column names are reused in ORDER BY query.

  • System memory pool is now unused by default and it will eventually be removed completely. All memory allocations will now be served from the general/user memory pool. The old behavior can be restored with the deprecated.legacy-system-pool-enabled config option.

  • Improve performance and memory usage for queries using row_number() followed by a filter on the row numbers generated.

  • Improve performance and memory usage for queries using ORDER BY followed by a LIMIT.

  • Improve performance of queries that process structural types and contain joins, aggregations, or table writes.

  • Add session property prefer-partial-aggregation to allow users to disable partial aggregations for queries that do not benefit.

  • Add support for current_user (see Session information).


  • Change rules in the System access control for enforcing matches between authentication credentials and a chosen username to allow more fine-grained control and ability to define superuser-like credentials.


  • Replace ORC writer stripe minimum row configuration hive.orc.writer.stripe-min-rows with stripe minimum data size hive.orc.writer.stripe-min-size.

  • Change ORC writer validation configuration hive.orc.writer.validate to switch to a sampling percentage hive.orc.writer.validation-percentage.

  • Fix optimized ORC writer writing incorrect data of type map or array.

  • Fix SHOW PARTITIONS and the $partitions table for tables that have null partition values.

  • Fix impersonation for the simple HDFS authentication to use login user rather than current user.


  • Support resource group selection based on resource estimates.