Presto 0.100 Documentation

10.21. Release 0.80

10.21. Release 0.80

New Hive ORC Reader

We have added a new ORC reader implementation. The new reader supports vectorized reads, lazy loading, and predicate push down, all of which make the reader more efficient and typically reduces wall clock time for a query. Although the new reader has been heavily tested, it is an extensive rewrite of the Apache Hive ORC reader, and may have some latent issues. If you are seeing issues, you can disable the new reader on a per-query basis by setting the <hive-catalog>.optimized_reader_enabled session property, or you can disable the reader by default by setting the Hive catalog property hive.optimized-reader.enabled=false.

Hive Changes

  • The maximum retry time for the Hive S3 file system can be configured by setting hive.s3.max-retry-time.
  • Fix Hive partition pruning for null keys (i.e. __HIVE_DEFAULT_PARTITION__).

Cassandra Changes

  • Update Cassandra driver to 2.1.0.
  • Map Cassandra TIMESTAMP type to Presto TIMESTAMP type.

“Big Query” Support

We’ve added experimental support for “big” queries. This provides a separate queue controlled by the following properties:

  • experimental.max-concurrent-big-queries
  • experimental.max-queued-big-queries

There are separate configuration options for queries that are submitted with the experimental_big_query session property:

  • experimental.big-query-initial-hash-partitions
  • experimental.big-query-max-task-memory

Queries submitted with this property will use hash distribution for all joins.

Metadata-Only Query Optimization

We now support an optimization that rewrites aggregation queries that are insensitive to the cardinality of the input (e.g., max(), min(), DISTINCT aggregates) to execute against table metadata.

For example, if key, key1 and key2 are partition keys, the following queries will benefit:

SELECT min(key), max(key) FROM t;

SELECT DISTINCT key FROM t;

SELECT count(DISTINCT key) FROM t;

SELECT count(DISTINCT key + 5) FROM t;

SELECT count(DISTINCT key) FROM (SELECT key FROM t ORDER BY 1 LIMIT 10);

SELECT key1, count(DISTINCT key2) FROM t GROUP BY 1;

This optimization is turned off by default. To turn it on, add optimizer.optimize-metadata-queries=true to the coordinator config properties.

Warning

This optimization will cause queries to produce incorrect results if the connector allows partitions to contain no data. For example, the Hive connector will produce incorrect results if your Hive warehouse contains partitions without data.

General Changes

  • Add support implicit joins. The following syntax is now allowed:

    SELECT * FROM a, b WHERE a.id = b.id;
    
  • Add property task.verbose-stats to enable verbose statistics collection for tasks. The default is false.

  • Format binary data in the CLI as a hex dump.

  • Add approximate numeric histogram function numeric_histogram().

  • Add array_sort() function.

  • Add map_keys() and map_values() functions.

  • Make row_number() completely streaming.

  • Add property task.max-partial-aggregation-memory to configure the memory limit for the partial step of aggregations.

  • Fix exception when processing queries with an UNNEST operation where the output was not used.

  • Only show query progress in UI after the query has been fully scheduled.

  • Add query execution visualization to the coordinator UI. It can be accessed via the query details page.