12.83. Release 0.163

General Changes

Fix data corruption when transporting dictionary-encoded data.
Fix potential deadlock when resource groups are configured with memory limits.
Improve performance for OUTER JOIN queries.
Improve exchange performance by reading from buffers in parallel.
Improve performance when only a subset of the columns resulting from a JOIN are referenced.
Make ALL, SOME and ANY non-reserved keywords.
Add from_big_endian_64() function.
Change xxhash64() return type from BIGINT to VARBINARY.
Change subscript operator for map types to fail if the key is not present in the map. The former behavior (returning NULL) can be restored by setting the deprecated.legacy-map-subscript config option.
Improve EXPLAIN ANALYZE to render stats more accurately and to include input statistics.
Improve tolerance to communication errors for long running queries. This can be adjusted with the query.remote-task.max-error-duration config option.

Fix issue that would prevent predicates from being pushed into Parquet reader.
Fix Hive metastore user permissions caching when tables are dropped or renamed.
Add experimental file based metastore which stores information in HDFS or S3 instead of a database.