12.18. Release 312

General Changes

Fix incorrect results for queries using IS [NOT] DISTINCT FROM. (#795)
Fix array_distinct, array_intersect semantics with respect to indeterminate values (i.e., NULL or structural types containing NULL). (#559)
Fix failure when the largest negative BIGINT value (-9223372036854775808) is used as a constant in a query. (#805)
Improve reliability for network errors when using Kerberos with Secure Internal Communication. (#838)
Improve performance of JOIN queries involving inline tables (VALUES). (#743)
Improve performance of queries containing duplicate expressions. (#730)
Improve performance of queries involving comparisons between values of different types. (#731)
Improve performance of queries containing redundant ORDER BY clauses in subqueries. This may affect the semantics of queries that incorrectly rely on implementation-specific behavior. The old behavior can be restored via the skip_redundant_sort session property or the optimizer.skip-redundant-sort configuration property. (#818)
Improve performance of IN predicates that contain subqueries. (#767)
Improve support for correlated subqueries containing redundant LIMIT clauses. (#441)
Add a new UUID type to represent UUIDs. (#755)
Add uuid() function to generate random UUIDs. (#786)
Add Phoenix Connector. (#672)
Make semantic error name available in client protocol. (#790)
Report operator statistics when experimental.work-processor-pipelines is enabled. (#788)

Raise required Java version to 8u161. This version allows unlimited strength crypto. (#779)
Show JVM configuration hint when JMX agent fails to start on Java 9+. (#838)
Skip starting JMX agent on Java 9+ if it is already configured via JVM properties. (#838)
Support configuring TrustStore for Secure Internal Communication using the internal-communication.https.truststore.path and internal-communication.https.truststore.key configuration properties. The path can point at a Java KeyStore or a PEM file. (#785)
Remove deprecated check for minimum number of workers before starting a coordinator. Use the query-manager.required-workers and query-manager.required-workers-max-wait configuration properties instead. (#95)

Fix SHOW GRANTS failure when metastore contains few tables. (#791)
Fix failure reading from information_schema.table_privileges table when metastore contains few tables. (#791)
Use Hive naming convention for file names when writing to bucketed tables. (#822)
Support new Hive bucketing conventions by allowing any number of files per bucket. This allows reading from partitions that were inserted into multiple times by Hive, or were written to by Hive on Tez (which does not create files for empty buckets).
Allow disabling the creation of files for empty buckets when writing data. This behavior is enabled by default for compatibility with previous versions of Presto, but can be disabled using the hive.create-empty-bucket-files configuration property or the create_empty_bucket_files session property. (#822)

Add support for pushing TABLESAMPLE into connectors via the ConnectorMetadata.applySample() method. (#753)