[
https://issues.apache.org/jira/browse/DRILL-1487?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14170266#comment-14170266
]
Timothy Chen commented on DRILL-1487:
-------------------------------------
Hi [~jni], so if I understand correctly you're saying that when order by is
added I need to include all rows in the same order by value computing the
aggregation function that matches the partition by column?
In other words, with dataset
employee id | position_id | salary
1 1 4
2 1 4
3 2 2
The sum with partition by and order by on position id returns:
1 1 8
2 1 8
3 2 2
instead of
1 1 4
2 1 8
3 2 2
Where the 2nd table is a sliding window sum?
And also what happens when we add configurable offsets?
> Drill window functions return wrong results
> -------------------------------------------
>
> Key: DRILL-1487
> URL: https://issues.apache.org/jira/browse/DRILL-1487
> Project: Apache Drill
> Issue Type: Bug
> Affects Versions: 0.6.0
> Reporter: Neeraja
> Assignee: Timothy Chen
>
> Executing the following window function with the requirement to see how a
> given employee salary would compare to the avg(salary) in his/her position.
> Query executes fine however returns wrong results(expect the avg(salary) to
> stay same for a given window (i.e position id)
> 0: jdbc:drill:zk=local> SELECT employee_id,position_id, salary, avg(salary)
> OVER (PARTITION BY position_id order by position_id) FROM cp.`employee.json`
> order by employee_id;
> +-------------+-------------+------------+------------+
> | employee_id | position_id | salary | EXPR$3 |
> +-------------+-------------+------------+------------+
> | 1 | 1 | 80000.0 | 80000.0 |
> | 2 | 2 | 40000.0 | 37500.0 |
> | 4 | 2 | 40000.0 | 38333.333333333336 |
> | 5 | 2 | 35000.0 | 35000.0 |
> | 6 | 3 | 25000.0 | 25000.0 |
> | 7 | 4 | 15000.0 | 15000.0 |
> | 8 | 11 | 10000.0 | 14333.333333333334 |
> | 9 | 11 | 17000.0 | 17000.0 |
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
