[
https://issues.apache.org/jira/browse/DRILL-3318?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Deneche A. Hakim updated DRILL-3318:
------------------------------------
Description:
I have a parquet file that contains an INT field with large enough values so
that the sum of the values overflows the INT limits, and a VARCHAR column with
one single value (to force all rows to be part of the same partition).
Computing the sums without casting will give similar results:
{noformat}
SELECT SUM(col_int) OVER(PARTITION BY col_char) FROM `3278.parquet` LIMIT 1;
+--------------+
| EXPR$0 |
+--------------+
| -3216087191 |
+--------------+
{noformat}
{noformat}
SELECT SUM(col_int) FROM `3278.parquet`;
+--------------+
| EXPR$0 |
+--------------+
| -3216087191 |
+--------------+
{noformat}
But if we cast the column before doing the sum, the results are now different:
{noformat}
SELECT SUM(CAST(col_int AS INT)) OVER(PARTITION BY col_char) FROM
`3278.parquet` LIMIT 1;
+-------------+
| EXPR$0 |
+-------------+
| 1078880105 |
+-------------+
{noformat}
{noformat}
SELECT SUM(CAST(col_int AS INT)) FROM `3278.parquet`;
+--------------+
| EXPR$0 |
+--------------+
| -3216087191 |
+--------------+
{noformat}
was:
I have a parquet file that contains an INT field with large enough values so
that the sum of the values overflows the INT limits, and a VARCHAR column with
one single value (to force all rows to be part of the same partition).
Computing the sums without casting will give similar results:
{noformat}
SELECT SUM(col_int) OVER(PARTITION BY col_char) FROM dfs.tmp.`3277` LIMIT 1;
+--------------+
| EXPR$0 |
+--------------+
| -3216087191 |
+--------------+
{noformat}
{noformat}
SELECT SUM(col_int) FROM dfs.tmp.`3277`;
+--------------+
| EXPR$0 |
+--------------+
| -3216087191 |
+--------------+
{noformat}
But if we cast the column before doing the sum, the results are now different:
{noformat}
SELECT SUM(CAST(col_int AS INT)) OVER(PARTITION BY col_char) FROM
dfs.tmp.`3277` LIMIT 1;
+-------------+
| EXPR$0 |
+-------------+
| 1078880105 |
+-------------+
{noformat}
{noformat}
SELECT SUM(CAST(col_int AS INT)) FROM dfs.tmp.`3277`;
+--------------+
| EXPR$0 |
+--------------+
| -3216087191 |
+--------------+
{noformat}
> SUM(CAST(col as INT)) shows different results when used in window functions
> ---------------------------------------------------------------------------
>
> Key: DRILL-3318
> URL: https://issues.apache.org/jira/browse/DRILL-3318
> Project: Apache Drill
> Issue Type: Bug
> Components: Query Planning & Optimization
> Affects Versions: 1.1.0
> Reporter: Deneche A. Hakim
> Assignee: Jinfeng Ni
> Priority: Minor
> Fix For: Future
>
>
> I have a parquet file that contains an INT field with large enough values so
> that the sum of the values overflows the INT limits, and a VARCHAR column
> with one single value (to force all rows to be part of the same partition).
> Computing the sums without casting will give similar results:
> {noformat}
> SELECT SUM(col_int) OVER(PARTITION BY col_char) FROM `3278.parquet` LIMIT 1;
> +--------------+
> | EXPR$0 |
> +--------------+
> | -3216087191 |
> +--------------+
> {noformat}
> {noformat}
> SELECT SUM(col_int) FROM `3278.parquet`;
> +--------------+
> | EXPR$0 |
> +--------------+
> | -3216087191 |
> +--------------+
> {noformat}
> But if we cast the column before doing the sum, the results are now different:
> {noformat}
> SELECT SUM(CAST(col_int AS INT)) OVER(PARTITION BY col_char) FROM
> `3278.parquet` LIMIT 1;
> +-------------+
> | EXPR$0 |
> +-------------+
> | 1078880105 |
> +-------------+
> {noformat}
> {noformat}
> SELECT SUM(CAST(col_int AS INT)) FROM `3278.parquet`;
> +--------------+
> | EXPR$0 |
> +--------------+
> | -3216087191 |
> +--------------+
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
