12.5. Release 329 (23 Jan 2020)
General Changes
- Fix incorrect result for
last_day_of_month() function for first day of month. (#2452)
- Fix incorrect results when handling
DOUBLE or REAL types with NaN values. (#2582)
- Fix query failure when coordinator hostname contains underscores. (#2571)
- Fix
SHOW CREATE TABLE failure when row types contain a field named after a
reserved SQL keyword. (#2130)
- Handle common disk failures during spill. When one disk fails but multiple
spill locations are configured, the healthy disks will be used for future queries.
(#2444)
- Improve performance and reduce load on external systems when
querying
information_schema. (#2488)
- Improve performance of queries containing redundant scalar subqueries. (#2456)
- Limit broadcasted table size to
100MB by default when using the AUTOMATIC
join type selection strategy. This avoids query failures or excessive memory usage when joining two or
more very large tables. (#2527)
- Enable cost based join reordering and join type selection
optimizations by default. The previous behavior can be restored by
setting
optimizer.join-reordering-strategy configuration property to ELIMINATE_CROSS_JOINS
and join-distribution-type to PARTITIONED. (#2528)
- Hide non-standard columns
comment and extra_info in the standard
information_schema.columns table. These columns can still be selected,
but will no longer appear when describing the table. (#2306)
Security Changes
- Add
ldap.bind-dn and ldap.bind-password LDAP properties to allow LDAP authentication
access LDAP server using service account. (#1917)
Hive Connector Changes
- Fix incorrect data returned when using S3 Select on uncompressed files. In our testing, S3 Select
was apparently returning incorrect results when reading uncompressed files, so S3 Select is disabled
for uncompressed files. (#2399)
- Fix incorrect data returned when using S3 Select on a table with
skip.header.line.count or
skip.footer.line.count property. S3 Select API does not support skipping footers or more than one
line of a header. In our testing, S3 Select was apparently sometimes returning incorrect results when
reading a compressed file with header skipping, so S3 Select is disabled when any of these table
properties is set to non-zero value. (#2399)
- Fix query failure for writes when one of the inserted
REAL or DOUBLE values
is infinite or NaN. (#2471)
- Fix performance degradation reading from S3 when the Kinesis connector is installed. (#2496)
- Allow reading data from Parquet files when the column type is declared as
INTEGER
in the table or partition, but is a DECIMAL type in the file. (#2451)
- Validate the scale of decimal types when reading Parquet files. This prevents
incorrect results when the decimal scale in the file does not match the declared
type for the table or partition. (#2451)
- Delete storage location when dropping an empty schema. (#2463)
- Improve performance when deleting multiple partitions by executing these actions concurrently. (#1812)
- Improve performance for queries containing
IN predicates over bucketing columns. (#2277)
- Add procedure
system.drop_stats() to remove the column statistics
for a table or selected partitions. (#2538)
Elasticsearch Connector Changes
- Add support for Array Types. (#2441)
- Reduce load on Elasticsearch cluster and improve query performance. (#2561)
PostgreSQL Connector Changes
- Fix mapping between PostgreSQL’s
TIME and Presto’s TIME data types.
Previously the mapping was incorrect, shifting it by the relative offset between the session
time zone and the Presto server’s JVM time zone. (#2549)