[ https://issues.apache.org/jira/browse/HIVE-14167?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15378453#comment-15378453 ]
Siddharth Seth commented on HIVE-14167: --------------------------------------- {code} + if (LlapProxy.isDaemon()) { + localDirList = HiveConf.getVar(conf, HiveConf.ConfVars.LLAP_DAEMON_WORK_DIRS); + if (localDirList != null && !localDirList.isEmpty()) { + return localDirList; + } // otherwise, fall back to use tez work dirs + } {code} This can be improved further for LLAP. WIll create a follow up jira. {code} if (conf.get("hive.execution.engine").equals("tez")) { {code} HiveConf.get( ... instead of conf.get ? {code} return null; {code} I suspect this will not work too well with HiveOnSpark. > Use work directories provided by Tez instead of directly using YARN local dirs > ------------------------------------------------------------------------------ > > Key: HIVE-14167 > URL: https://issues.apache.org/jira/browse/HIVE-14167 > Project: Hive > Issue Type: Improvement > Affects Versions: 2.1.0 > Reporter: Siddharth Seth > Assignee: Wei Zheng > Attachments: HIVE-14167.1.patch > > > HIVE-13303 fixed things to use multiple directories instead of a single tmp > directory. However it's using yarn-local-dirs directly. > I'm not sure how well using the yarn-local-dir will work on a secure cluster. > Would be better to use Tez*Context.getWorkDirs. This provides an app specific > directory - writable by the user. > cc [~sershe] -- This message was sent by Atlassian JIRA (v6.3.4#6332)