22.131. Release 0.66

Type System

In this release we have replaced the existing simple fixed type system with a fully extensible type system and have added several new types. We have also expanded the function system to support custom arithmetic, comparison and cast operators. For example, the new date/time types include an operator for adding an INTERVAL to a TIMESTAMP.

Existing functions have been updated to operate on and return the newly added types. For example, the ANSI color functions now operate on a COLOR type, and the date/time functions operate on standard SQL date/time types (described below).

Finally, plugins can now provide custom types and operators in addition to connectors and functions. This feature is highly experimental, so expect the interfaces to change over the next few releases. Also, since in SQL there is only one namespace for types, you should be careful to make names for custom types unique as we will add other common SQL types to Presto in the near future.

Date/Time Types

Presto now supports all standard SQL date/time types: DATE, TIME, TIMESTAMP and INTERVAL. All of the date/time functions and language constructs now operate on these types instead of BIGINT and perform temporal calculations correctly. This was previously broken due to, for example, not being able to detect whether an argument was a DATE or a TIMESTAMP. This change comes at the cost of breaking existing queries that perform arithmetic operations directly on the BIGINT value returned from the date/time functions.

As part of this work, we have also added the date_trunc() function which is convenient for grouping data by a time span. For example, you can perform an aggregation by hour:

SELECT date_trunc('hour', timestamp_column), count(*)
FROM ...
GROUP BY 1

Time Zones

This release has full support for time zone rules, which are needed to perform date/time calculations correctly. Typically, the session time zone is used for temporal calculations. This is the time zone of the client computer that submits the query, if available. Otherwise, it is the time zone of the server running the Presto coordinator.

Queries that operate with time zones that follow daylight saving can produce unexpected results. For example, if we run the following query to add 24 hours using in the America/Los Angeles time zone:

SELECT date_add('hour', 24, TIMESTAMP '2014-03-08 09:00:00');
-- 2014-03-09 10:00:00.000

The timestamp appears to only advance 23 hours. This is because on March 9th clocks in America/Los Angeles are turned forward 1 hour, so March 9th only has 23 hours. To advance the day part of the timestamp, use the day unit instead:

SELECT date_add('day', 1, TIMESTAMP '2014-03-08 09:00:00');
-- 2014-03-09 09:00:00.000

This works because the date_add() function treats the timestamp as list of fields, adds the value to the specified field and then rolls any overflow into the next higher field.

Time zones are also necessary for parsing and printing timestamps. Queries that use this functionality can also produce unexpected results. For example, on the same machine:

SELECT TIMESTAMP '2014-03-09 02:30:00';

The above query causes an error because there was no 2:30 AM on March 9th in America/Los_Angeles due to a daylight saving time transition.

In addition to normal TIMESTAMP values, Presto also supports the TIMESTAMP WITH TIME ZONE type, where every value has an explicit time zone. For example, the following query creates a TIMESTAMP WITH TIME ZONE:

SELECT TIMESTAMP '2014-03-14 09:30:00 Europe/Berlin';
-- 2014-03-14 09:30:00.000 Europe/Berlin

You can also change the time zone of an existing timestamp using the AT TIME ZONE clause:

SELECT TIMESTAMP '2014-03-14 09:30:00 Europe/Berlin'
     AT TIME ZONE 'America/Los_Angeles';
-- 2014-03-14 01:30:00.000 America/Los_Angeles

Both timestamps represent the same instant in time; they differ only in the time zone used to print them.

The time zone of the session can be set on a per-query basis using the X-Presto-Time-Zone HTTP header, or via the PrestoConnection.setTimeZoneId(String) method in the JDBC driver.

Localization

In addition to time zones, the language of the user is important when parsing and printing date/time types. This release adds localization support to the Presto engine and functions that require it: date_format() and date_parse(). For example, if we set the language to Spanish:

SELECT date_format(TIMESTAMP '2001-01-09 09:04', '%M'); -- enero

If we set the language to Japanese:

SELECT date_format(TIMESTAMP '2001-01-09 09:04', '%M'); -- 1月

The language of the session can be set on a per-query basis using the X-Presto-Language HTTP header, or via the PrestoConnection.setLocale(Locale) method in the JDBC driver.

Optimizations

  • We have upgraded the Hive connector to Hive 0.12 which includes performance improvements for RCFile.
  • GROUP BY and JOIN operators are now compiled to byte code and are significantly faster.
  • Reduced memory usage of GROUP BY and SELECT DISTINCT, which previously required several megabytes of memory per operator, even when the number of groups was small.
  • The planner now optimizes function call arguments. This should improve the performance of queries that contain complex expressions.
  • Fixed a performance regression in the HTTP client. The recent HTTP client upgrade was using inadvertently GZIP compression and has a bug in the buffer management resulting in high CPU usage.

SPI changes

In this release we have made a number of backward incompatible changes to the SPI:

  • Added Type and related interfaces
  • ConnectorType in metadata has been replaced with Type
  • Renamed TableHandle to ConnectorTableHandle
  • Renamed ColumnHandle to ConnectorColumnHandle
  • Renamed Partition to ConnectorPartition
  • Renamed PartitionResult to ConnectorPartitionResult
  • Renamed Split to ConnectorSplit
  • Renamed SplitSource to ConnectorSplitSource
  • Added a ConnectorSession parameter to most ConnectorMetadata methods
  • Removed most canHandle methods

General Bug Fixes

  • Fixed CLI hang after using USE CATALOG or USE SCHEMA
  • Implicit coercions in aggregations now work as expected
  • Nulls in expressions work as expected
  • Fixed memory leak in compiler
  • Fixed accounting bug in task memory usage
  • Fixed resource leak caused by abandoned queries
  • Fail queries immediately on unrecoverable data transport errors

Hive Bug Fixes

  • Fixed parsing of timestamps in the Hive RCFile Text SerDe (ColumnarSerDe) by adding configuration to set the time zone originally used when writing data

Cassandra Bug Fixes

  • Auto-reconnect if Cassandra session dies
  • Format collection types as JSON