13.36. Release 0.216

General Changes

  • Fix correctness issue for array_intersect() and array_distinct() when input contains both zeros and nulls.
  • Fix count(*) aggregation on empty relation when optimize_mixed_distinct_aggregation is enabled.
  • Improve table scan performance for structural types.
  • Improve performance for array_intersect().
  • Add reduce_agg() aggregate function.
  • Add millisecond() function.
  • Add an optimizer rule to filter the window partitions before executing the window operators.
  • Remove ON keyword for SHOW STATS.
  • Restrict WHERE clause in SHOW STATS to filters that can be pushed down to the connectors.
  • Remove node_id column from system.runtime.queries table.
  • Return final results to clients immediately for failed queries.

Web UI

  • Fix rendering of live plan view for queries involving index joins.

Hive Connector Changes

  • Fix accounting of time spent reading Parquet data.
  • Fix a corner case where the ORC writer fails with integer overflow when writing highly compressible data using dictionary encoding (#11930).
  • Fail queries reading Parquet files if statistics in those Parquet files are corrupt (e.g., min > max). To disable this behavior, set the configuration property hive.parquet.fail-on-corrupted-statistics or session property parquet_fail_with_corrupted_statistics to false.
  • Add support for S3 select pushdown, which enables pushing down projections and predicates into S3 for text files.

Kudu Connector Changes

  • Add number_of_replicas table property to SHOW CREATE TABLE output.

Cassandra Connector Changes

  • Add cassandra.splits-per-node and cassandra.protocol-version configuration properties to allow connecting to Cassandra servers older than 2.1.5.

MySQL, PostgreSQL, Redshift, and SQL Server Changes

  • Add support for predicate pushdown for columns of char(x) type.

Verifier Changes

  • Add run-teardown-on-result-mismatch configuration property to facilitate debugging. When set to false, temporary tables will not be dropped after checksum failures.

SPI Changes

  • Make ConnectorBucketNodeMap a top level class.
  • Use list instead of map for bucket-to-node mapping.

Note

This is a backwards incompatible change with the previous connector SPI. If you have written a connector that uses bucketing, you will need to update your code before deploying this release.