[ https://issues.apache.org/jira/browse/HIVE-13921?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15320037#comment-15320037 ]
Rui Li commented on HIVE-13921: ------------------------------- Hey [~ashutoshc], the failure happens only if the target file somehow exists before the moving. That means we need multiple tests writing to the same directory to trigger it. For spark, we have {{index_bitmap3.q}} and {{index_bitmap_auto.q}} that both write to {code}${system:test.tmp.dir}/index_result{code} If you run any of them alone, it'll pass successfully. But the latter will fail if you run both of them. You can also reproduce the failure for {{TestMinimrCliDriver}} by running the two tests together. We can remove the directory after each test to clean up the environment. I verified it can fix the failure. But ideally we don't need to do it because the query is an {{INSERT OVERWRITE DIRECTORY}}. That's why I think the code change makes sense. What's your opinion? > Fix spark on yarn tests for HoS > ------------------------------- > > Key: HIVE-13921 > URL: https://issues.apache.org/jira/browse/HIVE-13921 > Project: Hive > Issue Type: Test > Reporter: Rui Li > Assignee: Rui Li > Attachments: HIVE-13921.1.patch > > > {{index_bitmap3}} and {{constprog_partitioner}} have been failing. Let's fix > them here. -- This message was sent by Atlassian JIRA (v6.3.4#6332)