11.3. Release 0.227

General Changes

  • Fix a bug where index joins would fail with the error driver should never block.Queries will now fail if the index is unable to load within the time specified bythe index-loader-timeout configuration property and the index_loader_timeoutsession property.
  • Fix a bug that could cause NullPointerException in spatial join with geometrycollections on the build side.
  • Improve performance for queries with FULL OUTER JOIN where join keys have theCOALESCE function applied.
  • Improve performance for UNNEST queries.
  • Improve performance of repartitioning data between stages. The optimization can beenabled by the optimized_repartitioning session property or theexperimental.optimized-repartitioning configuration property.
  • Add spatial join (broadcast and partitioned) support for ST_Equals(),ST_Overlaps(), ST_Crosses(), and ST_Touches().
  • Add task_partitioned_writer_count session property to allow setting the numberof concurrent writers for partitioned (bucketed) writes.
  • Add IPPREFIX type and ip_prefix() function.
  • Add differential_entropy() functions to compute differential entropy.
  • Remove syntax support for SET PATH and CURRENT_PATH. The path information wasnever used by Presto.

Hive Changes

  • Fix a bug that might lead to corruption when writing sorted table in the recoverablegrouped execution mode.
  • Fix ORC stripe skipping when using bloom filter.
  • Improve the CPU load on coordinator by reducing the cost of serializing HiveSplit.
  • Improve GC pressure from Parquet reader by constraining the maximum column read size.This can be configured by the parquet_max_read_block_size session property or thehive.parquet.max-read-block-size configuration property.
  • Add support for sub-field pruning when reading Parquet files, so that only necessarysub-fields are extracted from struct columns.
  • Add configuration property hive.s3-file-system-type=HADOOP_DEFAULT to allowusers to switch different Hadoop file system implementations for s3:// addresses.The corresponding Hadoop File system implementation should be specified in core-site.xml

Raptor Changes

  • Fix memory leak in file descriptor during shard compaction. The regression was introduced in 0.219.

Verifier Changes

  • Add support for auto-resolving query failures with HIVE_TOO_MANY_OPEN_PARTITIONS error.
  • Add support to perform additional determinism analysis for queries with LIMIT clause.
  • Add detailed determinism analysis result to VerifierOutputEvent.

SPI Changes

  • Move AggregationNode to SPI. Connectors can now push down aggregation to table scan.
  • Move ProjectNode to SPI. Connectors can now push down projection to table scan.
  • Rename Block#getObject to Block#getBlock and remove unnecessary clazz parameter.