Custom Spark SQL Warehouse Directory
If required, you must configure the Spark SQL warehouse location in the upgraded cluster.
If Applicable: Configure Custom Spark SQL Warehouse Directory
Since Spark 2.0.0, Spark references spark.sql.warehouse.dir as the default Spark SQL
Warehouse location. In HDP 2.6.5.x, the standard value for the
spark.sql.warehouse.dir property is
/apps/hive/warehouse
. On an Ambari cluster with HDP 2.6.5.x, this
property must be set manually in both Custom spark2-defaults
and
Custom spark2-thrift-sparkconf
. When upgrading from HDP 2.6.5.x to
HDP intermediate bits, Ambari migrates the spark.sql.warehouse.dir
property to Advanced spark2-defaults
and Advanced
spark2-thrift-sparkconf
, and changes the value of this property to
/apps/spark/warehouse
. This is done to accommodate the separate
Spark and Hive catalogs introduced in the HDP intermediate bits. If you used a custom
setting for the spark.sql.warehouse.dir property in HDP 2.6.5.x, the Ambari upgrade to
HDP intermediate bits ignores the custom setting and sets the value of the
spark.sql.warehouse.dir property to
/apps/spark/warehouse
in both Advanced
spark2-defaults
and Advanced spark2-thrift-sparkconf
.
If you want to use a custom Spark SQL warehouse after upgrading to the HDP intermediate
bits, select Spark2 > Configs, then use Add
Property in Advanced spark2-defaults
and
Advanced spark2-thrift-sparkconf
to update the value of the
spark.sql.warehouse.dir property with the custom setting.
If Hive is configured to run as a custom user, you must change the ownership of your new
Hive warehouse directory to be owned by that custom user. hdfs dfs -chown
cstm-hive /warehouse/tablespace/managed/hive/