11.20. Release 0.210
General Changes
- Fix planning failure when aliasing columns of tables containing hiddencolumns (#11385).
- Fix correctness issue when
GROUP BY DISTINCT
terms contain references tothe same column using different syntactic forms (#11120). - Fix failures when querying
information_schema
tables using capitalized names. - Improve performance when converting between
ROW
types. - Remove user CPU time tracking as introduces non-trivial overhead.
- Select join distribution type automatically for queries involving outer joins.
Hive Connector Changes
- Fix a security bug introduced in 0.209 when using
hive.security=file
,which would allow any user to create, drop, or rename schemas. - Prevent ORC writer from writing stripes larger than the max configured sizewhen converting a highly dictionary compressed column to direct encoding.
- Support creating Avro tables with a custom schema using the
avro_schema_url
table property. - Support backward compatible Avro schema evolution.
- Support cross-realm Kerberos authentication for HDFS and Hive Metastore.
JDBC Driver Changes
- Deallocate prepared statement when
PreparedStatement
is closed. Previously,Connection
became unusable after many prepared statements were created. - Remove
getUserTimeMillis()
fromQueryStats
andStageStats
.
SPI Changes
SystemAccessControl.checkCanSetUser()
now takes anOptional<Principal>
rather than a nullablePrincipal
.- Rename
connectorId
tocatalogName
inConnectorFactory
,QueryInputMetadata
, andQueryOutputMetadata
. - Pass
ConnectorTransactionHandle
toConnectorAccessControl.checkCanSetCatalogSessionProperty()
. - Remove
getUserTime()
fromSplitStatistics
(referenced inSplitCompletedEvent
).
Note
These are backwards incompatible changes with the previous SPI.If you have written a plugin, you will need to update your codebefore deploying this release.