[ https://issues.apache.org/jira/browse/HIVE-12812?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16513407#comment-16513407 ]
Zoltan Haindrich commented on HIVE-12812: ----------------------------------------- That sounds awesome [~wanghaihua]! about this ticket: it seems to me that this setting could be set at the site.xml - am I wrong? I'm not sure what's going sideways with tez/spark in your production workloads; but I think they have improved a lot lately - w.r.t. being more stable in producing results reliably. The problem with mr is that it doesn't got all the new features...because it's already complicated; I think mainly because mr doesn't understand dags. I would like to suggest to give a try to Hive-3 (or wait for Hive-3.1 - it will get a lot of stability fixes) note: with query reexecution you can enable fallback to mr in hive3 > Enable mapred.input.dir.recursive by default to support union with aggregate > function > ------------------------------------------------------------------------------------- > > Key: HIVE-12812 > URL: https://issues.apache.org/jira/browse/HIVE-12812 > Project: Hive > Issue Type: Bug > Components: Hive > Affects Versions: 1.2.1, 2.1.0 > Reporter: Chaoyu Tang > Priority: Major > Attachments: HIVE-12812.patch, HIVE-12812.patch, HIVE-12812.patch > > > When union remove optimization is enabled, union query with aggregate > function writes its subquery intermediate results to subdirs which needs > mapred.input.dir.recursive to be enabled in order to be fetched. This > property is not defined by default in Hive and often ignored by user, which > causes the query failure and is hard to be debugged. > So we need set mapred.input.dir.recursive to true whenever union remove > optimization is enabled. -- This message was sent by Atlassian JIRA (v7.6.3#76005)