Fix regression that could cause high CPU usage for join queries when dictionary processing for joins is enabled.
bit_count()for bits between 33 and 63.
query.low-memory-killer.enabledconfig property has been replaced with
total-reservationto continue using the previous policy of killing the largest query. There is also a new policy,
total-reservation-on-blocked-nodes, which kills the query that is using the most memory on nodes that are out of memory (blocked).
Add support for grouped join execution. When both sides of a join have the same table partitioning and the partitioning is addressable, partial data can be loaded into memory at a time, making it possible to execute the join with less peak memory usage. The colocated join feature must be enabled with the
colocated-joins-enabledconfig property or the
colocated_joinsession property, and the
concurrent_lifespans_per_tasksession property must be specified.
Allow connectors to report the amount of physical written data.
Add ability to dynamically scale out the number of writer tasks rather than allocating a fixed number of tasks. Additional tasks are added when the the average amount of physical data per writer is above a minimum threshold. Writer scaling can be enabled with the
scale-writersconfig property or the
scale_writerssession property. The minimum size can be set with the
writer-min-sizeconfig property or the
writer_min_sizesession property. The tradeoff for writer scaling is that write queries can take longer to run due to the decreased writer parallelism while the writer count ramps up.
Add query type to the exact match source selector in the DB resource group selectors.
Improve display of values of the Geometry type.
Add support for grouped join execution for Hive tables when both sides of a join have the same bucketing property.
Report physical written data for the legacy RCFile writer, optimized RCFile writer, and optimized ORC writer. These writers thus support writer scaling, which can both reduce the number of written files and create larger files. This is especially important for tables that have many small partitions, as small files can take a disproportionately longer time to read.
Add page size distribution metrics.
MySQL, PostgreSQL, Redshift, and SQL Server#
information_schema.columnsif there are tables with no columns or no supported columns.