[
https://issues.apache.org/jira/browse/DRILL-315?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13835868#comment-13835868
]
Jason Altekruse commented on DRILL-315:
---------------------------------------
I do apologize, I had thought I finished posting to review board, but
apparently it did not go through. It should not be available publicly for
comments. I did put in syntax for predicate push-down while I was adding the
plan syntax for limit and column selection. I did not plan on implementing it
right away fro parquet, because we likely will not get that great a performance
gain from doing the predicates on individual values before coping them into
VVs. The problem is right now we seem to able to get objects or byte arrays out
of the interface how we are using them now. If we copy out individual values to
evaluate a predicate, I believe we'll get worse performance than just copying
out all of the data and sending it along to the filter operation in a lot of
cases. This is definitely something worth implementing for json though, as the
files will likely be smaller, and we have to copy/translate individual values
anyway.
> Reading only select columns from a parquet file
> -----------------------------------------------
>
> Key: DRILL-315
> URL: https://issues.apache.org/jira/browse/DRILL-315
> Project: Apache Drill
> Issue Type: New Feature
> Reporter: Jason Altekruse
> Assignee: Jason Altekruse
> Attachments:
> 0001-Parquet-files-now-read-by-column-rather-than-buffer-.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.1#6144)
