18.67. Release 0.90
This release has a memory leak and should not be used.
- Initial support for partition and placement awareness in the query planner. This can
result in better plans for queries involving
GROUP BYover the same key columns.
- Improve planning of UNION queries.
- Add presto version to query creation and completion events.
- Add property
task.writer-countto configure the number of writers per task.
- Fix a bug when optimizing constant expressions involving binary types.
- Fix bug where a table writer commits partial results while cleaning up a failed query.
- Fix a bug when unnesting an array of doubles containing NaN or Infinity.
- Fix failure when accessing elements in an empty array.
- Fix “Remote page is too large” errors.
- Improve error message when attempting to cast a value to
- Update the
approx_distinct()documentation with correct standard error bounds.
- Disable falling back to the interpreter when expressions fail to be compiled
to bytecode. To enable this option, add
compiler.interpreter-enabled=trueto the coordinator and worker config properties. Enabling this option will allow certain queries to run slowly rather than failing.
- Improve JDBC Driver conformance. In particular, all unimplemented
methods now throw
Functions and Language Features
- Add standard SQL function
every()as an alias for
- Add support for casting
- Add support for unparenthesized expressions in
- Added SET SESSION, RESET SESSION and SHOW SESSION.
- Improve formatting of
EXPLAIN (TYPE DISTRIBUTED)output and include additional information such as output layout, task placement policy and partitioning functions.
- Disable optimized metastore partition fetching for non-string partition keys.
This fixes an issue were Presto might silently ignore data with non-canonical
partition values. To enable this option, add
hive.assume-canonical-partition-keys=trueto the coordinator and worker config properties.
- Don’t retry operations against S3 that fail due to lack of permissions.
Slicefor table writer fragments.
ConnectorPageSinkwhich is a more efficient interface for column-oriented sources.
This is a backwards incompatible change with the previous connector SPI. If you have written a connector, you will need to update your code before deploying this release.