[ https://issues.apache.org/jira/browse/HIVE-8795?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14208465#comment-14208465 ]
Xuefu Zhang commented on HIVE-8795: ----------------------------------- Hi [~szehon], You're right. As stated in the description, it needs SPARK_HOME to be set so that the scripts can be found. I think this is okay for the build machine to have the script and/or the env variable, or even a full Spark installation, if we can make this configurable in such a way that on the build machine spark.master=local-cluster. The default can be still spark.master=local, so that devs are not required to have such settings in their environments for unit tests, unless one wants to test with local-cluster specifically. I assume this can be achieved. > Switch precommit test from local to local-cluster [Spark Branch] > ---------------------------------------------------------------- > > Key: HIVE-8795 > URL: https://issues.apache.org/jira/browse/HIVE-8795 > Project: Hive > Issue Type: Sub-task > Components: Spark > Reporter: Xuefu Zhang > Assignee: Szehon Ho > > It seems unlikely that Spark community will provide MRMiniCluster equivalent > (SPARK-3691), and Spark local-cluster was the recommendation. Latest research > shows that Spark local-cluster works with Hive. Therefore, for now, we use > Spark local-cluster (instead of current local) for our precommit test. > It's previous belived (HIVE-7382) that a Spark installation is required and > SPARK_HOME env variable needs to set. Since Spark pulls in Spark's assembly > jar, it's believed now we only need a few script from Spark installation > instead. -- This message was sent by Atlassian JIRA (v6.3.4#6332)