12.33. Release 0.213
General Changes
- Fix split scheduling backpressure when plan contains colocated join. Previously, splits
for the second and subsequent scan nodes (in scheduling order) were scheduled continuously
until completion, rather than pausing due to sufficient pending splits.
- Fix query execution failure or indefinite hang during grouped execution when all splits
for any lifespan are completely filtered out.
- Fix grouped execution to respect the configured concurrent lifespans per task.
Previously, it always used a single lifespan per task.
- Fix execution failure when using grouped execution with right or full outer joins
where the right side is not partitioned on the join key.
- Fix a scenario where too many rows are returned to clients in a single response.
- Do not allow setting invalid property values with SET SESSION.
- Disable stats calculator by default as it can cause a planning failure for
certain complex queries. It can be enabled with the
experimental.enable-stats-calculator
configuration property or the enable_stats_calculator session property.
- Avoid making guesses when estimating filters for joins. Previously, if nothing
was known about the filter, a
0.9 coefficient was applied as a filter factor.
Now, if nothing is known about a filter, the estimate will be unknown. A 0.9
coefficient will be applied for all additional conjuncts if at least a single
conjunct can be reasonably estimated.
- Improve inference of predicates for inner joins.
- Improve
EXPLAIN ANALYZE output by adding CPU time and enhancing accuracy of CPU fraction.
- Include stats and cost estimates in textual plans created on query completion.
- Enhance
SHOW STATS to support IN and BETWEEN predicates in the
WHERE condition of the SELECT clause.
- Remove transaction from explain plan for indexes joins.
- Add
max_drivers_per_task session property, allowing users to limit concurrency by
specifying a number lower than the system configured maximum. This can cause the
query to run slower and consume less resources.
- Add
join-max-broadcast-table-size configuration property and
join_max_broadcast_table_size session property to control the maximum estimated size
of a table that can be broadcast when using AUTOMATIC join distribution type (#11667).
- Add experimental config option
experimental.reserved-pool-enabled to disable the reserved memory pool.
- Add
targetResultSize query parameter to /v1/statement endpoint to control response data size.
Geospatial Changes
- Fix
ST_Distance() function to return NULL if any of the inputs is an
empty geometry as required by the SQL/MM specification.
- Add
ST_MultiPoint() function to construct multi-point geometry from an array of points.
- Add
geometry_union() function to efficiently union arrays of geometries.
- Add support for distributed spatial joins (#11072).
Server RPM Changes
- Allow running on a JVM from any vendor.
Web UI Changes
- Remove legacy plan UI.
- Add support for filtering queries by all error categories.
- Add dialog to show errors refreshing data from coordinator.
- Change worker thread list to not show thread stacks by default to improve page peformance.
Hive Connector Changes
- Fix LZO and LZOP decompression to work with certain data compressed by Hadoop.
- Fix ORC writer validation percentage so that zero does not result in 100% validation.
- Fix potential out-of-bounds read for ZSTD on corrupted input.
- Stop assuming no distinct values when column null fraction statistic is less than
1.0.
- Treat
-1 as an absent null count for compatibility with statistics written by
Impala.
- Preserve original exception for metastore network errors.
- Preserve exceptions from Avro deserializer
- Categorize text line length exceeded error.
- Remove the old Parquet reader. The
hive.parquet-optimized-reader.enabled
configuration property and parquet_optimized_reader_enabled session property
no longer exist.
- Remove the
hive.parquet-predicate-pushdown.enabled configuration property
and parquet_predicate_pushdown_enabled session property.
Pushdown is always enabled now in the Parquet reader.
- Enable optimized ORC writer by default. It can be disabled using the
hive.orc.optimized-writer.enabled configuration property or the
orc_optimized_writer_enabled session property.
- Use ORC file format as the default for new tables or partitions.
- Add support for Avro tables where the Avro schema URL is an HDFS location.
- Add
hive.parquet.writer.block-size and hive.parquet.writer.page-size
configuration properties and parquet_writer_block_size and
parquet_writer_page_size session properties for tuning Parquet writer options.
Memory Connector Changes
- Improve table data size accounting.
Thrift Connector Changes
- Include constraint in explain plan for index joins.
- Improve readability of columns, tables, layouts, and indexes in explain plans.
Verifier Changes
- Rewrite queries in parallel when shadowing writes.
