21.69. Release 0.192
- Fix performance regression in split scheduling introduced in 0.191. If a query scans a non-trivial number of splits (~1M splits in an hour), the coordinator CPU utilization can be very high, leading to elevated communication failures.
- Fix correctness issue in the
geometry_to_bing_tiles()function that causes it to return irrelevant tiles when bottom or right side of the bounding box of the geometry is aligned with the tile border.
- Fix handling of invalid WKT (well-known text) input in geospatial functions.
- Fix an issue that can cause long-running queries to hang when writer scaling is enabled.
- Fix cast from
DECIMALto conform to the SQL standard. For example, previously
cast (double '100000000000000000000000000000000' as decimal(38))would return
100000000000000005366162204393472. Now it returns
- Fix bug in validation of resource groups that prevented use of the
- Fail queries properly when the coordinator fails to fetch data from workers. Previously, it would return an HTTP 500 error to the client.
- Improve memory tracking for queries involving
row_number()that could cause over-committing memory resources for short time periods.
- Improve performance for queries involving
- Improve buffer utilization calculation for writer scaling.
- Remove tracking of per-driver peak memory reservation.
resource-groups.max-refresh-intervalconfig option to limit the maximum acceptable staleness of resource group configuration.
dictionary-processing-joins-enabledconfiguration option and
Web UI Changes
- Fix incorrect reporting of input size and positions in live plan view.
- Fix update of prompt after
- Fix correctness issue when rendering arrays of Bing tiles that causes the first entry to be repeated multiple times.
- Fix reading partitioned table statistics from newer Hive metastores.
- Do not treat file system errors as corruptions for ORC.
- Prevent reads from tables or partitions with
- Add support for validating ORC files after they have been written. This behavior can
be turned on via the
- Expose ORC writer statistics via JMX.
- Add configuration options to control ORC writer min/max rows per stripe and row group, maximum stripe size, and memory limit for dictionaries.
- Allow reading empty ORC files.
- Handle ViewFs when checking file system cache expiration.
- Improve error reporting when the target table of an insert query is dropped.
- Remove retry when creating Hive record reader. This can help queries fail faster.
- Remove support for
TIME WITH TIME ZONEand
TIMESTAMP WITH TIME ZONEtypes due to MySQL types not being able to store timezone information.
- Add support for
REALtype, which maps to MySQL’s
- Add support for
VARBINARYtype, which maps to PostgreSQL’s
- Fix support for pushing down inequality operators for string types.
- Add support for reading documents as
- Add support for MongoDB’s
- Treat document and array of documents as
- Allow nulls in history table values.