You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@drill.apache.org by "Jason Altekruse (JIRA)" <ji...@apache.org> on 2014/09/08 22:50:28 UTC
[jira] [Commented] (DRILL-1388) Incorrect results when projecting
nulls
[ https://issues.apache.org/jira/browse/DRILL-1388?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14126075#comment-14126075 ]
Jason Altekruse commented on DRILL-1388:
----------------------------------------
I had generated the select list from looking at all of the schema elements listed by parquet. pig_schema is not actually a column in the file, so the parquet reader currently will be producing a column with the name that is null filled. It appears that the project operator might not be handling this correctly, so it should be reviewed. I downgraded the priority as there is not an issue reading the real data.
> Incorrect results when projecting nulls
> ---------------------------------------
>
> Key: DRILL-1388
> URL: https://issues.apache.org/jira/browse/DRILL-1388
> Project: Apache Drill
> Issue Type: Bug
> Reporter: Jason Altekruse
>
> While testing fixed for the parquet nullable support I ran into an issue with unexpected results. I was selecting several columns out of file parquet file, which supports project pushdown. Currently the planner still includes a project operation after the scan in this case (to properly modify schema in the case of array indexing, project pushdown into scans is currently not supposed to be changing structure). I pulled the physical plan from the query and ran it without the extra project (as I was not selecting any array values) and got the expected results.
> Here is the query I ran, the file is too large to attach so you can e-mail me to get a copy of it.
> select pig_schema,ss_sold_date_sk,ss_item_sk,ss_cdemo_sk,ss_addr_sk, ss_hdemo_sk from store_sales
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)