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
The maximum retry time for the Hive S3 file system can be configured by setting
Fix Hive partition pruning for null keys (i.e.
Update Cassandra driver to 2.1.0.
TIMESTAMPtype to Presto
“Big Query” support#
We’ve added experimental support for “big” queries. This provides a separate queue controlled by the following properties:
There are separate configuration options for queries that are submitted with
experimental_big_query session property:
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.,
DISTINCT aggregates) to execute
against table metadata.
For example, if
key2 are partition keys, the following queries
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
to the coordinator config properties.
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.
Add support implicit joins. The following syntax is now allowed:
SELECT * FROM a, b WHERE a.id = b.id;
task.verbose-statsto enable verbose statistics collection for tasks. The default is
Format binary data in the CLI as a hex dump.
Add approximate numeric histogram function
task.max-partial-aggregation-memoryto configure the memory limit for the partial step of aggregations.
Fix exception when processing queries with an
UNNESToperation 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.