23.6. Release 0.157.1-t

Presto 0.157.1-t is equivalent to Presto release 0.157.1, with some additional features and patches.

SQL

  • Support correlated scalar aggregation subqueries
  • Support correlated scalar subqueries
  • Additional prepared statement syntax: DESCRIBE INPUT and DESCRIBE OUTPUT

Data Types

  • Native functions for Varchar(x) and Char(x)
  • Native Decimal functions

Security

  • Secured cluster communication over HTTPS

  • LDAP Authentication

    LDAP Authentication functionality in 0.157.1-t is same as 0.152.1-t, but the server properties have been renamed. The new properties and the only ones supported now are:

Property Example usage in Active Directory Example usage in OpenLDAP
http-server.authentication.type LDAP LDAP
authentication.ldap.url ldaps://ldapserver:636 ldaps://ldapserver:636
authentication.ldap.user-bind-pattern ${USER}@domain.com uid=${USER},ou=Asia,dc=domain,dc=com
authentication.ldap.user-base-dn ou=Asia,dc=domain,dc=com ou=Asia,dc=domain,dc=com
authentication.ldap.group-auth-pattern (&(objectClass=person)(sAMAccountName=${USER}) (memberof=cn=group,ou=America,dc=domain,dc=com)) (&(objectClass=inetOrgPerson)(uid=${USER}) (memberof=cn=group,ou=America,dc=domain,dc=com))
  • SHOW GRANTS support for Hive connector

Performance

  • Automatically choose between repartitioned and replicated based on Hive table statistics. Must have join-distribution-type set to automatic.

  • Avoid cross joins if possible by reordering joins. Must have reorder-joins set to true.

  • Optimize DictionaryBlock.copyPositions()

  • Faster Decimal implementation

  • Move certain filters in the WHERE clauses to be executed as part of the INNER JOIN rather than a filter after the join. For example: SELECT * FROM t t1 JOIN t t2 ON t1.a = t2.a WHERE (t1.b+t2.b)*5 > 100000000

  • Code generation for joins with filters

  • Merge non-identical windows (for the same partition by and order by but different frame)

  • Optimize execution order of Windowing functions to minimize the number of times data is repartitioned. For example:

    wfunc1() OVER (PARTITION BY A ORDER BY B)

    wfunc2() OVER (PARTITION BY C ORDER BY D)

    wfunc3() OVER (PARTITION BY A ORDER BY B)

    If the execution order is as written in the query, it results in partitioning the data 3 times. If we rearrange to execute wfunc1 -> wfunc3 -> wfunc2 then data is only partitioned twice. Must have optimizer.reorder-windows set to true.

  • Add support for running regular expression functions using the more efficent re2j-td library by setting the session variable regex_library to RE2J. The memory footprint can be adjusted by setting re2j_dfa_states_limit. Additionally, the number of times the re2j library falls back from its DFA algorithm to the NFA algorithm (due to hitting the states limit) before immediately starting with the NFA algorithm can be set with the re2j_dfa_retries session variable.

Scale

Beta - Spilling to disk for aggregations and avoid the memory limits within Presto. Must have beta.spill-enabled set to true.

Teradata JDBC Driver

The Teradata JDBC driver does not support batch queries.

Connectors

  • TPC-DS connector was added
  • Teradata has added official support for the Cassandra connector.

Presto Admin

Removed Sudo requirement for Presto Admin. See Presto Admin upgrade documentation.

Documentation

  • Query Optimizer
  • Query Performance Analysis
  • Presto Tuning
  • Installation of Presto via Presto Admin

Bugs Fixed

  • Fix issue “Hive table is corrupt. It is declared as being bucketed, but the files do not match the bucketing declaration. The number of files in the directory (1) does not match the declared.” by fixing support for Hive bucketed tables. See option hive.multi-file-bucketing.enabled in the Presto Hive connector documentation.
  • Allow empty partitions for clustered hive tables
  • Fix issue “low must be less than or equal to high” that can occur with ORC and Character data.
  • Fix incorrect stream property derivations from GroupIdNode
  • Remove broken %w specifier for MySQL date functions
  • RPM to include installation of the Memory connector.
  • Fix query failures for connectors that produce non-remotely accessible splits (0.158)
  • Fix non-linear performance issue when parsing certain SQL expressions. (0.158)

Unsupported Functionality

Some functionality from Presto 0.157.1 may work but is not officially supported by Teradata.

  • The installation method as documented on prestodb.io.
  • Web Connector for Tableau
  • The following connectors:
    • Accumulo Connector
    • Kafka Connector
    • Local File Connector
    • MongoDB Connector
    • Redis Connector

SQL/DML/DDL Limitations

  • The SQL keyword end is used as a column name in system.runtime.queries, so in order to query from that column, end must be wrapped in quotes
  • NATURAL JOIN is not supported
  • LIMIT ALL and OFFSET are not supported

Hive Connector Limitations

File Formats

Teradata supports data stored in the following formats:

  • Text files
  • ORC
  • RCFILE
  • PARQUET

TIMESTAMP limitations

Presto supports a granularity of milliseconds for the TIMESTAMP datatype, while Hive supports microseconds.

TIMESTAMP values in tables are parsed according to the server’s timezone. If this is not what you want, you must start Presto in the UTC timezone. To do this, set the JVM timezone to UTC: -Duser.timezone=UTC and also add the following property in the Hive connector properties file: hive.time-zone=UTC.

Presto’s method for declaring timestamps with/with out timezone is not sql standard. In Presto, both are declared using the word TIMESTAMP, e.g. TIMESTAMP '2003-12-10 10:32:02.1212' or TIMESTAMP '2003-12-10 10:32:02.1212 UTC'. The timestamp is determined to be with or without timezone depending on whether you include a time zone at the end of the timestamp. In other systems, timestamps are explicitly declared as TIMESTAMP WITH TIME ZONE or TIMESTAMP WITHOUT TIME ZONE (with TIMESTAMP being an alias for one of them). In these systems, if you declare a TIMESTAMP WITHOUT TIMEZONE, and your string has a timezone at the end, it is silently ignored. If you declare a TIMESTAMP WITH TIME ZONE and no time zone is included, the string is interpreted in the user time zone.

INSERT INTO ... VALUES limitations

The data types must be exact, i.e. must use 2.0 for double, cast('2015-1-1' as date) for date, and you must supply a value for every column.

INSERT INTO ... SELECT limitations

INSERT INTO creates unreadable data (unreadable both by Hive and Presto) if a Hive table has a schema for which Presto only interprets some of the columns (e.g. due to unsupported data types). This is because the generated file on HDFS will not match the Hive table schema.

If called through JDBC, executeUpdate does not return the count of rows inserted.

Hive Parquet Issues

PARQUET support in Hive imposes more limitations than the other file types.

DATE and BINARY datatypes are not supported

PostgreSQL and MySQL Connectors Limitations

Known Bugs PostgreSQL connector describe table reports Table has no supported column types inappropriately. https://github.com/facebook/presto/issues/4082

Security

Presto connects to MySQL and PostgreSQL using the credentials specified in the properties file. The credentials are used to authenticate the users while establishing the connection. Presto runs queries as the “presto” service user and does not pass down user information to MySQL or PostgreSQL connectors.

Datatypes

PostgreSQL and MySQL each support a wide variety of datatypes (PostgreSQL datatypes, MySQL datatypes). Many of these types are not supported in Presto. Table columns that are defined using an unsupported type are not visible to Presto users. These columns are not shown when describe table or select * SQL statements are executed.

CREATE TABLE

CREATE TABLE (...) does not work, but CREATE TABLE AS SELECT does.

DROP TABLE

DROP TABLE is not supported.

Limited SQL push-down

Presto does not “push-down” aggregate calculations to PostgreSQL or MySQL. This means that when a user executes a simple query such as SELECT COUNT(*) FROM lineitem the entire table will be retrieved and the aggregate calculated by Presto. If the table is large or the network slow, this may take a very long time.

MySQL Catalogs

MySQL catalog names are mapped to Presto schema names.

23.7. Release 0.157.1-t.0.1

The following has been added to 0.157.1-t.0.1:

  • Set join_distribution_type default to repartitioned
  • Tuning and minor improvements for when join_distribution_type is automatic
  • Documentation improvements
  • Fix wrong results for nested except
  • Fix principal error in Hive Kerberos Authentication

23.8. Release 0.157.1-t.0.3

The following has been added to 0.157.1-t.0.3:

  • Fix incorrect result that may result as part of Decimal data type coercion
  • Additional documentation

23.9. Release 0.157.1-t.0.4

The following has been added to 0.157.1-t.0.4:

  • Renamed the “experimental.*” spill related properties to ”.beta”. “experimental” was the historic name and “beta” more accurately reflects this first phase of spill to disk.
  • Updated Sandbox VM documentation for Presto Admin 2.0

23.10. Release 0.157.1-t.0.5

The following has been added to 0.157.1-t.0.5:

23.11. Release 0.157.1-t.0.6

The following has been added to 0.157.1-t.0.6:

  • Fix ConnectorMetadata#beginQuery behavior. Now it is called before ConnectorSplitManager#getSplits.

23.12. Release 0.157.1-t.0.7

The following has been added to 0.157.1-t.0.7:

  • Fix incorrect pruning of join output columns.

23.13. Release 0.157.1-t.0.8

The following has been added to 0.157.1-t.0.8:

  • Fix creating external tables so that they are properly recognized by the Hive metastore. The Hive table property EXTERNAL is now set to TRUE in addition to the setting the table type. Any previously created tables need to be modified to have this property.
  • Fix HDFS configuration loading in the Hive connector. Without this fix, there sometimes were problems with connecting to HA HDFS.
  • Optimize window functions to avoid copying data unnecessarily for better CPU and memory utilization.
  • Optimize LIKE predicates with escape characters.

23.14. Release 0.157.1-t.1

The following has been added to 0.157.1-t.1:

  • Add support for SQL Server connector.
  • Fix bug where partially created files on S3 are not removed after failed CTAS.