23.25. Release 0.177
Warning
Query may incorrectly produce NULL
when no row qualifies for the aggregation
if the optimize_mixed_distinct_aggregations
session property or
the optimizer.optimize-mixed-distinct-aggregations
config option is enabled.
This optimization was introduced in Presto version 0.156.
General Changes
- Fix correctness issue when performing range comparisons over columns of type
CHAR
. - Fix correctness issue due to mishandling of nulls and non-deterministic expressions in
inequality joins unless
fast_inequality_join
is disabled. - Fix excessive GC overhead caused by lambda expressions. There are still known GC issues with captured lambda expressions. This will be fixed in a future release.
- Check for duplicate columns in
CREATE TABLE
before asking the connector to create the table. This improves the error message for most connectors and will prevent errors for connectors that do not perform validation internally. - Add support for null values on the left-hand side of a semijoin (i.e.,
IN
predicate with subqueries). - Add
SHOW STATS
to display table and query statistics. - Improve implicit coercion support for functions involving lambda. Specifically, this makes
it easier to use the
reduce()
function. - Improve plans for queries involving
ORDER BY
andLIMIT
by avoiding unnecessary data exchanges. - Improve performance of queries containing window functions with identical
PARTITION BY
andORDER BY
clauses. - Improve performance of certain queries involving
OUTER JOIN
and aggregations, or containing certain forms of correlated subqueries. This optimization is experimental and can be turned on via thepush_aggregation_through_join
session property or theoptimizer.push-aggregation-through-join
config option. - Improve performance of certain queries involving joins and aggregations. This optimization
is experimental and can be turned on via the
push_partial_aggregation_through_join
session property. - Improve error message when a lambda expression has a different number of arguments than expected.
- Improve error message when certain invalid
GROUP BY
expressions containing lambda expressions.
Hive Changes
- Fix handling of trailing spaces for the
CHAR
type when reading RCFile. - Allow inserts into tables that have more partitions than the partitions-per-scan limit.
- Add support for exposing Hive table statistics to the engine. This option is experimental and
can be turned on via the
statistics_enabled
session property. - Ensure file name is always present for error messages about corrupt ORC files.
Cassandra Changes
- Remove caching of metadata in the Cassandra connector. Metadata caching makes Presto violate
the consistency defined by the Cassandra cluster. It’s also unnecessary because the Cassandra
driver internally caches metadata. The
cassandra.max-schema-refresh-threads
,cassandra.schema-cache-ttl
andcassandra.schema-refresh-interval
config options have been removed. - Fix intermittent issue in the connection retry mechanism.
Web UI Changes
- Change cluster HUD realtime statistics to be aggregated across all running queries.
- Change parallelism statistic on cluster HUD to be averaged per-worker.
- Fix bug that always showed indeterminate progress bar in query list view.
- Change running drivers statistic to exclude blocked drivers.
- Change unit of CPU and scheduled time rate sparklines to seconds on query details page.
- Change query details page refresh interval to three seconds.
- Add uptime and connected status indicators to every page.
CLI Changes
- Add support for preprocessing commands. When the
PRESTO_PREPROCESSOR
environment variable is set, all commands are piped through the specified program before being sent to the Presto server.