[
https://issues.apache.org/jira/browse/DRILL-4191?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15053840#comment-15053840
]
Khurram Faraaz commented on DRILL-4191:
---------------------------------------
Also when we order by the same column as the partitioning (partition by) column
we get the right result for last value. The column being c1 in the query below.
{code}
0: jdbc:drill:schema=dfs.tmp> select c1, last_value(c2) over(partition by c1
order by c1) lstval from md_627;
+-----+-------------+
| c1 | lstval |
+-----+-------------+
| 1 | 2015-12-12 |
| 1 | 2015-12-12 |
| 1 | 2015-12-12 |
+-----+-------------+
3 rows selected (0.458 seconds)
{code}
> Last value function returns incorrect results.
> ----------------------------------------------
>
> Key: DRILL-4191
> URL: https://issues.apache.org/jira/browse/DRILL-4191
> Project: Apache Drill
> Issue Type: Bug
> Components: Execution - Flow
> Affects Versions: 1.4.0
> Environment: 4 node cluster on CentOS
> Reporter: Khurram Faraaz
>
> Last value function returns incorrect results.
> {noformat}
> [root@centos-01 ~]# cat MD627.csv
> 1,2015-01-01
> 1,2015-01-02
> 1,2015-12-12
> git commit id : b9068117
> create table md_627 as select cast(columns[0] as int) c1, cast(columns[1] as
> date) c2 from `MD627.csv`;
> [root@centos-01 parquet-tools]# ./parquet-schema ../md627/0_0_0.parquet
> message root {
> optional int32 c1;
> optional int32 c2 (DATE);
> }
> 0: jdbc:drill:schema=dfs.tmp> select * from md_627;
> +-----+-------------+
> | c1 | c2 |
> +-----+-------------+
> | 1 | 2015-01-01 |
> | 1 | 2015-01-02 |
> | 1 | 2015-12-12 |
> +-----+-------------+
> 3 rows selected (0.265 seconds)
> 0: jdbc:drill:schema=dfs.tmp> select c1, last_value(c2) over(partition by c1
> order by c2) lstval from md_627;
> +-----+-------------+
> | c1 | lstval |
> +-----+-------------+
> | 1 | 2015-01-01 |
> | 1 | 2015-01-02 |
> | 1 | 2015-12-12 |
> +-----+-------------+
> 3 rows selected (0.405 seconds)
> 0: jdbc:drill:schema=dfs.tmp> explain plan for select c1, last_value(c2)
> over(partition by c1 order by c2) lstval from md_627;
> +------+------+
> | text | json |
> +------+------+
> | 00-00 Screen
> 00-01 Project(c1=[$0], lstval=[$1])
> 00-02 Project(c1=[$0], lstval=[$1])
> 00-03 Project(c1=[$0], $1=[$2])
> 00-04 Window(window#0=[window(partition {0} order by [1] range
> between UNBOUNDED PRECEDING and CURRENT ROW aggs [LAST_VALUE($1)])])
> 00-05 SelectionVectorRemover
> 00-06 Sort(sort0=[$0], sort1=[$1], dir0=[ASC], dir1=[ASC])
> 00-07 Project(c1=[$1], c2=[$0])
> 00-08 Scan(groupscan=[ParquetGroupScan
> [entries=[ReadEntryWithPath [path=maprfs:///tmp/md_627]],
> selectionRoot=maprfs:/tmp/md_627, numFiles=1, usedMetadataFile=false,
> columns=[`c1`, `c2`]]])
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
