12.1. Release 329 (Jan 23 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)
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)
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)

Add ldap.bind-dn and ldap.bind-password LDAP properties to allow LDAP authentication access LDAP server using service account. (#1917)

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)
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)

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)