[ https://issues.apache.org/jira/browse/HIVE-22770?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17024468#comment-17024468 ]
Ashutosh Chauhan commented on HIVE-22770: ----------------------------------------- Configuration conf = ((KryoWithHooks) kryo).getConf(); + if (conf != null && conf.get(HiveConf.ConfVars.HIVE_EXECUTION_ENGINE.varname, "").equalsIgnoreCase("spark")) { + mapWork.internFields(); + } + Lets remove above code block. With that removed we won't need {{new Configuration}} which is an expensive call. That means we will never intern fields. > Skip interning of MapWork fields for tez during deserialization > --------------------------------------------------------------- > > Key: HIVE-22770 > URL: https://issues.apache.org/jira/browse/HIVE-22770 > Project: Hive > Issue Type: Improvement > Reporter: László Bodor > Assignee: László Bodor > Priority: Major > Attachments: HIVE-22770.01.patch, HIVE-22770.01.patch, > HIVE-22770.02.patch, HIVE-22770.02.patch > > > HIVE-19937 introduced some interning logic into mapwork deserialization > process, but it's only related to spark, maybe we should skip this for tez, > reducing the cpu pressure in tez tasks. -- This message was sent by Atlassian Jira (v8.3.4#803005)