20.42. Release 0.207
Fix a planning issue for queries where correlated references were used in
Remove support for legacy
JOIN ... USINGbehavior.
Change behavior for unnesting an array of
rowtype to produce multiple columns.
reorder_joinssession property and the
reorder-joinsconfiguration property. They are replaced by the
join_reordering_strategysession property and the
NONEmaintains the order of the joins as written and is equivalent to
ELIMINATE_CROSS_JOINSwill eliminate any unnecessary cross joins from the plan and is equivalent to
AUTOMATICwill use the new cost-based optimizer to select the best join order. To simplify migration, setting the
reorder_joinssession property overrides the new session and configuration properties.
distributed_joinssession property and the
distributed-joins-enabledconfiguration property. They are replaced by the
join_distribution_typesession property and the
PARTITIONEDturns on hash partitioned joins and is equivalent to
BROADCASTchanges the join strategy to broadcast and is equivalent to
AUTOMATICwill use the new cost-based optimizer to select the best join strategy. If no statistics are available,
AUTOMATICis the same as
REPARTITIONED. To simplify migration, setting the
distributed_joinssession property overrides the new session and configuration properties.
Add support for column properties.
optimizer.max-reordered-joinsconfiguration property to set the maximum number of joins that can be reordered at once using cost-based join reordering.
Add support for
Fail on startup when configuration for file based system access control is invalid.
Add support for securing communication between cluster nodes with Kerberos authentication.
Web UI Changes
Add peak total (user + system) memory to query details UI.
Hive Connector Changes
Fix handling of
VARCHAR(length)type in the optimized Parquet reader. Previously, predicate pushdown failed with
Mismatched Domain types: varchar(length) vs varchar.
Fail on startup when configuration for file based access control is invalid.
Add support for HDFS wire encryption.
Allow ORC files to have struct columns with missing fields. This allows the table schema to be changed without rewriting the ORC files.
Change collector for columns statistics to only consider a sample of partitions. The sample size can be changed by setting the
Memory Connector Changes
Add support for dropping schemas.
Remove deprecated table/view-level access control methods.
Change predicate in constraint for accessing table layout to be optional.
Change schema name in
ConnectorMetadatato be optional rather than nullable.