11.14. Release 0.216
General Changes
- Fix correctness issue for
array_intersect()
andarray_distinct()
when input containsboth zeros and nulls. - Fix
count(*)
aggregation on empty relation whenoptimize_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 fromsystem.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 writinghighly compressible data using dictionary encoding (#11930).
- Fail queries reading Parquet files if statistics in those Parquet files arecorrupt (e.g., min > max). To disable this behavior, set the configurationproperty
hive.parquet.fail-on-corrupted-statistics
or session propertyparquet_fail_with_corrupted_statistics
to false. - Add support for S3 select pushdown, which enables pushing downprojections and predicates into S3 for text files.
Kudu Connector Changes
- Add
number_of_replicas
table property toSHOW CREATE TABLE
output.
Cassandra Connector Changes
- Add
cassandra.splits-per-node
andcassandra.protocol-version
configuration propertiesto 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 toupdate your code before deploying this release.