Presto 0.100 Documentation

10.31. Release 0.70

10.31. Release 0.70


This release contained a packaging error that resulted in an unusable server tarball. Do not use this release.


We have added support for creating views within Presto. Views are defined using Presto syntax but are stored (as blobs) by connectors. Currently, views are supported by the Raptor and Hive connectors. For the Hive connector, views are stored within the Hive metastore as Hive views, but they cannot be queried by Hive, nor can Hive views be queried by Presto.

See 建视图 and 删视图 for details and examples.

DUAL Table

The synthetic DUAL table is no longer supported. As an alternative, please write your queries without a FROM clause or use the VALUES syntax.

Presto Verifier

There is a new project, Presto校验器, which can be used to verify a set of queries against two different clusters.

Connector Improvements

  • Connectors can now add hidden columns to a table. Hidden columns are not displayed in DESCRIBE or information_schema, and are not considered for SELECT *. As an example, we have added a hidden row_number column to the tpch connector.
  • Presto contains an extensive test suite to verify the correctness. This test suite has been extracted into the presto-test module for use during connector development. For an example, see TestRaptorDistributedQueries.

Machine Learning Functions

We have added two new machine learning functions, which can be used by advanced users familiar with LIBSVM. The functions are learn_libsvm_classifier and learn_libsvm_regressor. Both take a parameters string which has the form key=value,key=value

General Changes

  • New comparison functions: greatest() and least()
  • New window functions: first_value(), last_value(), and nth_value()
  • We have added a config option to disable falling back to the interpreter when expressions fail to be compiled to bytecode. To set this option, add  compiler.interpreter-enabled=false to etc/ This will force certain queries to fail rather than running slowly.
  • DATE values are now implicitly coerced to TIMESTAMP and TIMESTAMP WITH TIME ZONE by setting the hour/minute/seconds to 0 with respect to the session timezone.
  • Minor performance optimization when planning queries over tables with tens of thousands of partitions or more.
  • Fixed a bug when planning ORDER BY ... LIMIT queries which could result in duplicate and un-ordered results under rare conditions.
  • Reduce the size of stats collected from tasks, which dramatically reduces garbage generation and improves coordinator stability.
  • Fix compiler cache for expressions.
  • Fix processing of empty or commented out statements in the CLI.

Hive Changes

  • There are two new configuration options for the Hive connector, hive.max-initial-split-size, which configures the size of the initial splits, and hive.max-initial-splits, which configures the number of initial splits. This can be useful for speeding up small queries, which would otherwise have low parallelism.
  • The Hive connector will now consider all tables with a non-empty value for the table property presto_offline to be offline. The value of the property will be used in the error message.
  • We have added support for for DROP TABLE in the hive connector. By default, this feature is not enabled. To enable it, set hive.allow-drop-table=true in your Hive catalog properties file.
  • Ignore subdirectories when generating splits (this now matches the non-recursive behavior of Hive).
  • Fix handling of maps with null keys.