hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Vineet Garg (JIRA)" <>
Subject [jira] [Commented] (HIVE-17043) Remove non unique columns from group by keys if not referenced later
Date Fri, 21 Sep 2018 20:33:00 GMT


Vineet Garg commented on HIVE-17043:

[~jcamachorodriguez] I agree it is ugly. The problem with {{RelMdColumnUniqueness}} is that
it only tells you if given set of columns are unique or not, for this optimization we need
to know the set of unique keys (if there are any for a given input). Therefore {{RelMdColumnUniqueness}}
wouldn't really work here.

Another possible solution I could think of was calling {{getColumnOrigin}} on each group key
to track lineage and build the set, then calling {{getTableOrigin}} to get to the base table
using which we can figure out the keys, get rid of the corresponding columns from group sets.
But this will be pretty expensive (calling getColumnOrigin on all the keys and then calling

I think we should keep RelMdUniqueKeys for determining unique keys based on the constraints,
it seems like it is designed for this. We can write (preferably in later patch) different
logic/methods for getRowCount to use (which will be based on stats) since  it only override
project to determine uniqueness based on statistics.

Let me know what you think.

> Remove non unique columns from group by keys if not referenced later
> --------------------------------------------------------------------
>                 Key: HIVE-17043
>                 URL:
>             Project: Hive
>          Issue Type: Sub-task
>          Components: Logical Optimizer
>    Affects Versions: 3.0.0
>            Reporter: Ashutosh Chauhan
>            Assignee: Vineet Garg
>            Priority: Major
>         Attachments: HIVE-17043.1.patch, HIVE-17043.2.patch, HIVE-17043.3.patch, HIVE-17043.4.patch
> Group by keys may be a mix of unique (or primary) keys and regular columns. In such cases
presence of regular column won't alter cardinality of groups. So, if regular columns are not
referenced later, they can be dropped from group by keys. Depending on operator tree may result
in those columns not being read at all from disk in best case. In worst case, we will avoid
shuffling and sorting regular columns from mapper to reducer, which still could be substantial
CPU and network savings.

This message was sent by Atlassian JIRA

View raw message