You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hive.apache.org by "Jesus Camacho Rodriguez (Jira)" <ji...@apache.org> on 2020/05/28 19:30:00 UTC
[jira] [Commented] (HIVE-23557) HiveVolcanoPlanner works with 2
different RelMetadataQuery types which may lead to problems
[ https://issues.apache.org/jira/browse/HIVE-23557?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17119024#comment-17119024 ]
Jesus Camacho Rodriguez commented on HIVE-23557:
------------------------------------------------
Thanks [~kgyrtkirk]. Interesting... Shouldn't the call just before that one to {{optCluster.invalidateMetadataQuery();}} invalidate the cached metadata?
> HiveVolcanoPlanner works with 2 different RelMetadataQuery types which may lead to problems
> -------------------------------------------------------------------------------------------
>
> Key: HIVE-23557
> URL: https://issues.apache.org/jira/browse/HIVE-23557
> Project: Hive
> Issue Type: Bug
> Reporter: Zoltan Haindrich
> Priority: Major
>
> switching to trace mode uncovers Volcano estimation issues; which lead to the usage of multiple (different) metadataquery sources ; the actual assertion is based on the fact that the new metadataquery gives a better estimation for the same rel.
> the planner switches over to the default metadataquery provider after some time; however this switch may cause trouble - as after this switch the already cached internal visited nodesets become incorrect.
> https://github.com/apache/hive/blob/f49d257c560c81c38259e95023b20c544acb4d10/ql/src/java/org/apache/hadoop/hive/ql/parse/CalcitePlanner.java#L2234
--
This message was sent by Atlassian Jira
(v8.3.4#803005)