[
https://issues.apache.org/jira/browse/CALCITE-900?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14933917#comment-14933917
]
Sean Hsuan-Yi Chu commented on CALCITE-900:
-------------------------------------------
The motivation is that we found out infinite planning on DRILL side
(unfortunately, I have not been able to reproduce on Calcite). For this query
{code}
WITH year_total
AS (SELECT c.r_regionkey customer_id,
1 year_total
FROM table c
UNION ALL
SELECT c.r_regionkey customer_id,
1 year_total
FROM table c)
SELECT t_w_firstyear.customer_id
FROM year_total t_w_firstyear,
year_total t_w_secyear
WHERE t_w_firstyear.year_total = t_w_secyear.year_total
AND t_w_firstyear.year_total > 0 and t_w_secyear.year_total > 0
{code}
According to the trace, ProjectSetOpTransposeRule and FilterSetOpTransposeRule
run into a loop, where one generates new plan triggering the other one.
Basically, I was trying to find a place to cut the loop.
Let me show the trace in the attachment and explain what is happening in the
planning below:
rel#217 in SubSet#2 keeps triggering FilterSetOpTransposeRule and add new Rels
into SubSet#2 (who is the input of rel#126);
On the other hand, rel#126 in SubSet#11 keeps triggering
ProjectSetOpTransposeRule and add new Set (that means if you let planner keep
going, you would see more sets get added)
> ProjectSetOpTransposeRule can be improved when the input is a SetOperator
> with Trivial Project on top
> -----------------------------------------------------------------------------------------------------
>
> Key: CALCITE-900
> URL: https://issues.apache.org/jira/browse/CALCITE-900
> Project: Calcite
> Issue Type: Improvement
> Components: core
> Reporter: Sean Hsuan-Yi Chu
> Assignee: Julian Hyde
>
> When SetOperator with Trivial Project on top, pushing this trivial project
> does not necessary. An improvement can be done if we just discard this
> trivial project.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
