[
https://issues.apache.org/jira/browse/DRILL-1309?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14101597#comment-14101597
]
Hanifi Gunes commented on DRILL-1309:
-------------------------------------
Trying this query with a simple filter
{code:sql}
explain plan for
select t1.columns[1]
from
dfs.`/Users/hgunes/workspaces/mapr/incubator-drill/data/t1.csv` t1
where t1.columns[0] = 10;
{code}
generates
{panel}
00-00 Screen
00-01 Project(EXPR$0=[ITEM($0, 1)])
00-02 SelectionVectorRemover
00-03 Filter(condition=[=(CAST(ITEM($0, 0)):INTEGER NOT NULL, 10)])
00-04 ProducerConsumer
00-05 Scan(groupscan=[EasyGroupScan [selectionRoot=/data/t1.csv,
columns = [SchemaPath [`columns`]]]])
{panel}
Looks like we need projection pushdown for csv just like DRILL-1283.
> Projected columns are not pushed into scan
> -------------------------------------------
>
> Key: DRILL-1309
> URL: https://issues.apache.org/jira/browse/DRILL-1309
> Project: Apache Drill
> Issue Type: Bug
> Reporter: Mehant Baid
> Assignee: Hanifi Gunes
> Fix For: 0.5.0
>
>
> While performing the following query on two text files, the projected column
> does not get pushed into the scan.
> explain plan for select t1.columns[1] from dfs.`/tmp/t1.csv` t1,
> dfs.`/tmp/t2.csv` t2 where t1.columns[0] = t2.columns[0];
> 00-00 Screen
> 00-01 Project(EXPR$0=[ITEM($0, 1)])
> 00-02 HashJoin(condition=[=($1, $2)], joinType=[inner])
> 00-04 Project(columns=[$0], $f2=[ITEM($0, 0)])
> 00-06 Scan(groupscan=[EasyGroupScan [selectionRoot=/tmp/t1.csv,
> columns = [SchemaPath [`columns`], SchemaPath [`columns`[0]]]]])
> 00-03 Project($f20=[$0])
> 00-05 Project($f2=[ITEM($0, 0)])
> 00-07 Scan(groupscan=[EasyGroupScan [selectionRoot=/tmp/t2.csv,
> columns = [SchemaPath [`columns`[0]]]]])
> In the above plan (00-06) we see that the scan contains the following column
> projections pushed into scan: 'columns' and 'columns[0]'.
> We should not push 'columns' into the scan, instead push 'columns[1]' which
> is the projected column into the scan.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
