Dynamic split concurrency#
The number of running leaf splits per query is now dynamically adjusted to improve
overall cluster throughput.
task.initial-splits-per-node can be used to set
the initial number of splits, and
can be used to change how frequently adjustments happen. The session properties
also be used.
Fix planning bug that causes some joins to not be redistributed when
Fix rare leak of stage objects and tasks for queries using
task.join-concurrencyconfig which can be used to increase concurrency for the probe side of joins.
Remove cursor-based readers for ORC and DWRF file formats, as they have been replaced by page-based readers.
Fix creating tables on S3 with CREATE TABLE AS.