There is a performance regression in this release for
GROUP BY and
queries when the length of the keys is between 16 and 31 bytes. This is fixed
in Release 0.130.
- Fix a planner issue that could cause queries involving
OUTER JOINto return incorrect results.
- Some queries, particularly those using
min_by(), now accurately reflect their true memory usage and thus appear to use more memory than before.
- Fix SHOW SESSION to not show hidden session properties.
- Fix hang in large queries with
- Fix an issue when casting empty arrays or arrays containing only
NULLto other types.
- Table property names are now properly treated as case-insensitive.
- Minor UI improvements for query detail page.
- Do not display useless stack traces for expected exceptions in verifier.
- Improve performance of queries involving
UNION ALLthat write data.
- Introduce the
P4HyperLogLogtype, which uses an implementation of the HyperLogLog data structure that trades off accuracy and memory requirements when handling small sets for an improvement in performance.
JDBC Driver Changes#
- Always clear screen when canceling query with
- Make client request timeout configurable.
Network Topology Aware Scheduling#
The scheduler can now be configured to take network topology into account when
scheduling splits. This is set using the
config. See Tuning Presto for more information.
- The S3 region is no longer automatically configured when running in EC2.
To enable this feature, use
hive.s3.pin-client-to-current-region=truein your Hive catalog properties file. Enabling this feature is required to access S3 data in the China isolated region, but prevents accessing data outside the current region.
- Server-side encryption is now supported for S3. To enable this feature,
hive.s3.sse.enabled=truein your Hive catalog properties file.
- Add support for the
- Add support for S3