Release 0.203¶

General Changes¶

Fix spurious duplicate key errors from map().
Fix planning failure when a correlated subquery containing a LIMIT clause is used within EXISTS (#10696).
Fix out of memory error caused by missing pushback checks in data exchanges.
Fix execution failure for queries containing a cross join when using bucketed execution.
Fix execution failure for queries containing an aggregation function with DISTINCT and a highly selective aggregation filter. For example: sum(DISTINCT x) FILTER (WHERE y = 0)
Fix quoting in error message for SHOW PARTITIONS.
Eliminate redundant calls to check column access permissions.
Improve query creation reliability by delaying query start until the client acknowledges the query ID by fetching the first response link. This eliminates timeouts during the initial request for queries that take a long time to analyze.
Remove support for legacy ORDER BY semantics.
Distinguish between inner and left spatial joins in explain plans.

Fix sending authentication challenge when at least two of the KERBEROS, PASSWORD, or JWT authentication types are configured.
Allow using PEM encoded (PKCS #8) keystore and truststore with the HTTP server and the HTTP client used for internal communication. This was already supported for the CLI and JDBC driver.

Declare a dependency on uuidgen. The uuidgen program is required during installation of the Presto server RPM package and lack of it resulted in an invalid config file being generated during installation.

Fix complex type handling in the optimized Parquet reader. Previously, null values, optional fields, and Parquet backward compatibility rules were not handled correctly.
Fix an issue that could cause the optimized ORC writer to fail with a LazyBlock error.
Improve error message for max open writers.

Fix retry of requests when the remote Thrift server indicates that the error is retryable.