[
https://issues.apache.org/jira/browse/DRILL-724?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14010280#comment-14010280
]
Aman Sinha commented on DRILL-724:
----------------------------------
Since the values of the the two columns get swapped for certain rows, I
suspected that some column projections may not be right. If I insert a
Project right after the SelectionVectorRemover and before the
SingleMergeExchange, the results are correct. The modified part of the plan is
below (the new Project is id: 100). This indicates that in the absence of a
Project, somehow the order of the columns fed into the SingleMergeExchange is
not right on all the nodes.
...
...
}, {
"pop" : "external-sort",
"@id" : 65538,
"child" : 65539,
"orderings" : [ {
"order" : "ASC",
"expr" : "`c_row`",
"nullDirection" : "UNSPECIFIED"
} ],
"reverse" : false,
"initialAllocation" : 1000000,
"maxAllocation" : 10000000000
}, {
"pop" : "selection-vector-remover",
"@id" : 65537,
"child" : 65538,
"initialAllocation" : 1000000,
"maxAllocation" : 10000000000
}, {
"pop" : "project",
"@id" : 100,
"exprs" : [ {
"ref" : "`c_row`",
"expr" : "`c_row`"
}, {
"ref" : "`c_int`",
"expr" : "`c_int`"
} ],
"child" : 65537,
"initialAllocation" : 1000000,
"maxAllocation" : 10000000000
}, {
"pop" : "single-merge-exchange",
"@id" : 2,
"child" : 100,
"orderings" : [ {
"order" : "ASC",
"expr" : "`c_row`",
"nullDirection" : "UNSPECIFIED"
} ],
"initialAllocation" : 1000000,
"maxAllocation" : 10000000000
}, {
...
...
> select two columns with order by produce wrong results
> ------------------------------------------------------
>
> Key: DRILL-724
> URL: https://issues.apache.org/jira/browse/DRILL-724
> Project: Apache Drill
> Issue Type: Bug
> Reporter: Chun Chang
> Priority: Blocker
>
> postgres:
> foodmart=# select c_row, c_int from data order by c_row;
> c_row | c_int
> -------+-------------
> 1 | 0
> 2 | 1
> 3 | -1
> 4 | 12
> 5 | 123
> 6 | 92032039
> 7 | -23395000
> 8 | -99392039
> 9 | -2147483648
> 10 | 2147483647
> 11 | 32767
> 12 | -32767
> 13 | 49032
> 14 | -4989385
> 15 | 69834830
> 16 | 243
> 17 | -426432
> 18 | -3904
> 19 | 489392758
> 20 | 589032574
> 21 | 340000504
> 22 | 0
> 23 | 1
> (23 rows)
> drill:
> 0: jdbc:drill:schema=dfs> select c_row, c_int from data order by c_row;
> +------------+------------+
> | c_int | c_row |
> +------------+------------+
> | 1 | 0 |
> | 1 | 2 |
> | -1 | 3 |
> | 12 | 4 |
> | 123 | 5 |
> | 92032039 | 6 |
> | -23395000 | 7 |
> | -99392039 | 8 |
> | -2147483648 | 9 |
> | 10 | 2147483647 |
> | 32767 | 11 |
> | -32767 | 12 |
> | 13 | 49032 |
> | -4989385 | 14 |
> | 69834830 | 15 |
> | 243 | 16 |
> | -426432 | 17 |
> | -3904 | 18 |
> | 489392758 | 19 |
> | 589032574 | 20 |
> | 340000504 | 21 |
> | 22 | 0 |
> | 1 | 23 |
> +------------+------------+
> 23 rows selected (10.932 seconds)
> notice for drill, on some of the rows, the columns are exchanged.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
