This release has the potential for data loss in the Hive connector when writing bucketed sorted tables.
Fix an issue with memory accounting that would lead to garbage collection pauses and out of memory exceptions.
Fix an issue that produces incorrect results when
push_aggregation_through_joinis enabled (#10724).
Preserve field names when unnesting columns of type
Make the cluster out of memory killer more resilient to memory accounting leaks. Previously, memory accounting leaks on the workers could effectively disable the out of memory killer.
Improve planning time for queries over tables with high column count.
Add a limit on the number of stages in a query. The default is
100and can be changed with the
query.max-stage-countconfiguration property and the
Add a cluster memory leak detector that logs queries that have possibly accounted for memory usage incorrectly on workers. This is a tool to for debugging internal errors.
Add support for correlated subqueries requiring coercions.
Add experimental support for running on Linux ppc64le.
Fix creation of the history file when it does not exist.
PRESTO_HISTORY_FILEenvironment variable to override location of history file.
Remove size limit for writing bucketed sorted tables.
Support writer scaling for Parquet.
Improve stripe size estimation for the optimized ORC writer. This reduces the number of cases where tiny ORC stripes will be written.
Provide the actual size of CHAR, VARCHAR, and VARBINARY columns to the cost based optimizer.
Collect column level statistics when writing tables. This is disabled by default, and can be enabled by setting the
Include error message from remote server in query failure message.