[ 
https://issues.apache.org/jira/browse/HIVE-18608?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16349660#comment-16349660
 ] 

Mithun Radhakrishnan edited comment on HIVE-18608 at 2/2/18 2:03 AM:
---------------------------------------------------------------------

I've attached an initial implementation, where dictionary encoding might be 
disabled via a table-property ({{'orc.skip.dictionary.for.columns'}}).

Note: I've only added support for top-level columns. Specifying this on a 
{{STRUCT}} will disable dictionary encoding for the entire sub-tree (i.e. all 
members of the STRUCT), recursively.

It might be good to support selection at an arbitrary depth, at a later date.
E.g. {{myInfoArray.__elem__.emailBody}}.


was (Author: mithun):
I've attached an initial implementation, where dictionary encoding might be 
disabled via a table-property ({{'orc.skip.dictionary.for.columns'}}.

Note: I've only added support for top-level columns. Specifying this on a 
{{STRUCT}} will disable dictionary encoding for the entire sub-tree (i.e. all 
members of the STRUCT), recursively.

It might be good to support selection at an arbitrary depth, at a later date.
E.g. {{myInfoArray.__elem__.emailBody}}.

> ORC should allow selectively disabling dictionary-encoding on specified 
> columns
> -------------------------------------------------------------------------------
>
>                 Key: HIVE-18608
>                 URL: https://issues.apache.org/jira/browse/HIVE-18608
>             Project: Hive
>          Issue Type: New Feature
>          Components: ORC
>    Affects Versions: 3.0.0, 2.4.0, 2.2.1
>            Reporter: Mithun Radhakrishnan
>            Assignee: Mithun Radhakrishnan
>            Priority: Major
>         Attachments: HIVE-18608.1-branch-2.2.patch
>
>
> Just as ORC allows the choice of columns to enable bloom-filters on, it would 
> be nice to have a way to specify which columns {{DICTIONARY_V2}} encoding 
> should be disabled on.
> Currently, the choice of dictionary-encoding depends on the results of 
> sampling the first row-stride within a stripe. If the user knows that a 
> column's cardinality is bound to prevent an effective dictionary, she might 
> choose to simply disable it on just that column, and avoid the cost of 
> sampling in the first row-stride.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to