You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hive.apache.org by "Ádám Szita (Jira)" <ji...@apache.org> on 2021/11/02 13:26:00 UTC
[jira] [Work started] (HIVE-25651) Enable LLAP cache affinity for
Iceberg ORC splits
[ https://issues.apache.org/jira/browse/HIVE-25651?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Work on HIVE-25651 started by Ádám Szita.
-----------------------------------------
> Enable LLAP cache affinity for Iceberg ORC splits
> -------------------------------------------------
>
> Key: HIVE-25651
> URL: https://issues.apache.org/jira/browse/HIVE-25651
> Project: Hive
> Issue Type: Improvement
> Reporter: Ádám Szita
> Assignee: Ádám Szita
> Priority: Major
>
> Since HiveIcebergInputformat doesn't implement any LLAP marker interfaces, cache affinity is never tried, and so any split containing ORC file parts may go to a random LLAP daemon, causing subpar hit ratio later.
> So we should:
> * let HS2 know that cache affinity is required for this inputformat
> * prevent Iceberg from grouping separate files together in one combined split in case of LLAP execution
> * provide proper getPath() result for HiveIcebergSplit, so that HostAffinitySplitLocationProvider calculates different hashes for different files (right now getPath() returns table location only)
--
This message was sent by Atlassian Jira
(v8.3.4#803005)