Release 307 (3 Apr 2019)#
General#
Fix cleanup of spill files for queries using window functions or
ORDER BY
. (#543)Optimize queries containing
ORDER BY
together withLIMIT
over anOUTER JOIN
by pushingORDER BY
andLIMIT
to the outer side of the join. (#419)Improve performance of table scans for data sources that produce tiny pages. (#467)
Improve performance of
IN
subquery expressions that contain aDISTINCT
clause. (#551)Expand support of types handled in
EXPLAIN (TYPE IO)
. (#509)Add support for outer joins involving lateral derived tables (i.e.,
LATERAL
). (#390)Add support for setting table comments via the COMMENT syntax. (#200)
Web UI#
Allow UI to work when opened as
/ui
(no trailing slash). (#500)
Security#
Make query result and cancellation URIs secure. Previously, an authenticated user could potentially steal the result data of any running query. (#561)
Server RPM#
Prevent JVM from allocating large amounts of native memory. The new configuration is applied automatically when Presto is installed from RPM. When Presto is installed another way, or when you provide your own
jvm.config
, we recommend adding-Djdk.nio.maxCachedBufferSize=2000000
to yourjvm.config
.
CLI#
JDBC driver#
Abort query synchronously when the
ResultSet
is closed or when theStatement
is cancelled. Previously, the abort was sent in the background, allowing the JVM to exit before the abort was received by the server. (#580)
Hive connector#
Add safety checks for Hive bucketing version. Hive 3.0 introduced a new bucketing version that uses an incompatible hash function. The Hive connector will treat such tables as not bucketed when reading and disallows writing. (#512)
Add support for setting table comments via the COMMENT syntax. (#200)
Other connectors#
These changes apply to the MySQL, PostgreSQL, Redshift, and SQL Server connectors.
Fix reading and writing of
timestamp
values. Previously, an incorrect value could be read, depending on the Presto JVM time zone. (#495)Add support for using a client-provided username and password. The credential names can be configured using the
user-credential-name
andpassword-credential-name
configuration properties. (#482)
SPI#
LongDecimalType
andIpAddressType
now useInt128ArrayBlock
instead ofFixedWithBlock
. Any code that creates blocks directly, rather than using theBlockBuilder
returned from theType
, will need to be updated. (#492)Remove
FixedWidthBlock
. Use one of the*ArrayBlock
classes instead. (#492)Add support for simple constraint pushdown into connectors via the
ConnectorMetadata.applyFilter()
method. (#541)