Fix planning failure for queries involving
GROUPINGthat require implicit coercions in expressions containing aggregate functions.
Fix potential workload imbalance when using topology-aware scheduling.
Fix performance regression for queries containing
DISTINCTaggregates over the same column.
Fix a memory leak that occurs on workers.
Improve error handling when a
HAVINGclause 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
Improve rendering of the
approx_distinct()function now supports a standard error in the range of
Add support for
ORDER BYin 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-enabledconfig property or the
Add support for casting to
Allow treating decimal literals as values of the
DECIMALtype rather than
DOUBLE. This behavior can be enabled by setting the
parse-decimal-literals-as-doubleconfig property or the
parse_decimal_literals_as_doublesession property to
Add JMX counter to track the number of submitted queries.
Add priority column to the DB resource group selectors.
Add exact match source selector to the DB resource group selectors.
Add support for setting client tags.
Improve table scan parallelism.
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-sizefor 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.
Make Thrift retry configurable.
Add JMX counters for Thrift requests.
RecordSinkinterface, which was difficult to use correctly and had no advantages over the
This is a backwards incompatible change with the previous connector SPI.
If you have written a connector that uses the
you will need to update your code before deploying this release.