13.62. Release 0.190

General Changes

  • Fix correctness issue for array_min() and array_max() when arrays contain NaN.
  • Fix planning failure for queries involving GROUPING that require implicit coercions in expressions containing aggregate functions.
  • Fix potential workload imbalance when using topology-aware scheduling.
  • Fix performance regression for queries containing DISTINCT aggregates over the same column.
  • Fix a memory leak that occurs on workers.
  • Improve error handling when a HAVING clause contains window functions.
  • Avoid unnecessary data redistribution when writing when the target table has the same partition property as the data being written.
  • Ignore case when sorting the output of SHOW FUNCTIONS.
  • Improve rendering of the BingTile type.
  • The approx_distinct() function now supports a standard error in the range of [0.0040625, 0.26000].
  • Add support for ORDER BY in aggregation functions.
  • Add dictionary processing for joins which can improve join performance up to 50%. This optimization can be disabled using the dictionary-processing-joins-enabled config property or the dictionary_processing_join session property.
  • Add support for casting to INTERVAL types.
  • Add ST_Buffer() geospatial function.
  • Allow treating decimal literals as values of the DECIMAL type rather than DOUBLE. This behavior can be enabled by setting the parse-decimal-literals-as-double config property or the parse_decimal_literals_as_double session property to false.
  • Add JMX counter to track the number of submitted queries.

Resource Groups Changes

  • Add priority column to the DB resource group selectors.
  • Add exact match source selector to the DB resource group selectors.

CLI Changes

  • Add support for setting client tags.

JDBC Driver Changes

  • Add getPeakMemoryBytes() to QueryStats.

Accumulo Changes

  • Improve table scan parallelism.

Hive Changes

  • Fix query failures for the file-based metastore implementation when partition column values contain a colon.
  • Improve performance for writing to bucketed tables when the data being written is already partitioned appropriately (e.g., the output is from a bucketed join).
  • Add config property hive.max-outstanding-splits-size for the maximum amount of memory used to buffer splits for a single table scan. Additionally, the default value is substantially higher than the previous hard-coded limit, which can prevent certain queries from failing.

Thrift Connector Changes

  • Make Thrift retry configurable.
  • Add JMX counters for Thrift requests.

SPI Changes

  • Remove the RecordSink interface, which was difficult to use correctly and had no advantages over the PageSink interface.

Note

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