12.39. Release 0.207
General Changes
- Fix a planning issue for queries where correlated references were used in
VALUES.
- Remove support for legacy
JOIN ... USING behavior.
- Change behavior for unnesting an array of
row type to produce multiple columns.
- Deprecate the
reorder_joins session property and the reorder-joins
configuration property. They are replaced by the join_reordering_strategy
session property and the optimizer.join-reordering-strategy configuration
property. NONE maintains the order of the joins as written and is equivalent
to reorder_joins=false. ELIMINATE_CROSS_JOINS will eliminate any
unnecessary cross joins from the plan and is equivalent to reorder_joins=true.
AUTOMATIC will use the new cost-based optimizer to select the best join order.
To simplify migration, setting the reorder_joins session property overrides the
new session and configuration properties.
- Deprecate the
distributed_joins session property and the
distributed-joins-enabled configuration property. They are replaced by the
join_distribution_type session property and the join-distribution-type
configuration property. PARTITIONED turns on hash partitioned joins and
is equivalent to distributed_joins-enabled=true. BROADCAST changes the
join strategy to broadcast and is equivalent to distributed_joins-enabled=false.
AUTOMATIC will use the new cost-based optimizer to select the best join
strategy. If no statistics are available, AUTOMATIC is the same as
REPARTITIONED. To simplify migration, setting the distributed_joins
session property overrides the new session and configuration properties.
- Add support for column properties.
- Add
optimizer.max-reordered-joins configuration property to set the maximum number of joins that
can be reordered at once using cost-based join reordering.
- Add support for
char type to approx_distinct().
Security Changes
- 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
hive.partition-statistics-sample-size property.
Memory Connector Changes
- Add support for dropping schemas.
SPI Changes
- Remove deprecated table/view-level access control methods.
- Change predicate in constraint for accessing table layout to be optional.
- Change schema name in
ConnectorMetadata to be optional rather than nullable.
