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 task.split-concurrency-adjustment-interval can be used to change how frequently adjustments happen. The session properties initial_splits_per_node and split_concurrency_adjustment_interval can also be used.

General Changes

• Fix planning bug that causes some joins to not be redistributed when distributed-joins-enabled is true.
• Fix rare leak of stage objects and tasks for queries using LIMIT.
• Add experimental task.join-concurrency config which can be used to increase concurrency for the probe side of joins.

Hive Changes

• Remove cursor-based readers for ORC and DWRF file formats, as they have been replaced by page-based readers.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.