Fix regression in 0.214 that could cause queries to produce incorrect results for queries using map types.
Fix reporting of the processed input data for source stages in
Fail queries that use non-leaf resource groups. Previously, they would remain queued forever.
Improve CPU usage for specific queries (#11757).
Extend stats and cost model to support
row_number()window function estimates.
Improve the join type selection and the reordering of join sides for cases where the join output size cannot be estimated.
Add dynamic scheduling support to grouped execution. When a stage is executed with grouped execution and the stage has no remote sources, table partitions can be scheduled to tasks in a dynamic way, which can help mitigating skew for queries using grouped execution. This feature can be enabled with the
dynamic_schedule_for_grouped_executionsession property or the
Split the reporting of raw input data and processed input data for source operators.
Remove collection and reporting of raw input data statistics for the
Local Exchange, and
Local Merge Sortoperators.
EXPLAIN (TYPE IO)output when there are too many discrete components. This avoids large output at the cost of reduced granularity.
Add support for
UNION ALLto optimizer’s cost model.
Add support for estimating the cost of filters by using a default filter factor. The default value for the filter factor can be configured with the
default_filter_factor_enabledsession property or the
Add input validation checks to
ST_LineString()to conform with the specification.
Improve spatial join performance.
Enable spatial joins for join conditions expressed with the
Web UI Changes#
Fix Capture Snapshot button for showing current thread stacks.
Fix dropdown for expanding stage skew component on the query details page.
Improve the performance of the thread snapshot component on the worker status page.
Make the reporting of Cumulative Memory usage consistent on the query list and query details pages.
Remove legacy thread UI.
Add predicate pushdown support for the
DATEtype to the Parquet reader. This change also fixes a bug that may cause queries with predicates on
DATEcolumns to fail with type mismatch errors.
Prevent printing the value of the
redis.passwordconfiguration property to log files.