[jira] [Commented] (KYLIN-3812) optimize the child CompareTupleFilter in a CompareTupleFilter

["Zhong Yanghong (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/KYLIN-3812?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16813482#comment-16813482 ] 

Zhong Yanghong commented on KYLIN-3812:
---------------------------------------

Hi [~Yifei_Wu94], at least for calcite 1.12.0, the 1=1 constant part is not optimized.

> optimize the child CompareTupleFilter in a CompareTupleFilter
> -------------------------------------------------------------
>
>                 Key: KYLIN-3812
>                 URL: https://issues.apache.org/jira/browse/KYLIN-3812
>             Project: Kylin
>          Issue Type: Improvement
>          Components: Query Engine
>            Reporter: Zhong Yanghong
>            Assignee: Zhong Yanghong
>            Priority: Major
>             Fix For: v3.0.0
>
>
> Currently it's not well supported for CompareTupleFilter to have a child of CompareTupleFilter. However, in some cases, it's better to support it.
> {code}
> where (colA = (1=1))
> {code}
> The *(1=1)* can be transformed to "true". And then this filter can be pushed down to hbase. Otherwise, the filter *(colA = (1=1))* does not work in hbase.
> And it may return incorrect results for the following SQL:
> {code}
> select colA
>        case
>            when colB = (1 = 1) then 'B'
>            when colC = (1 = 1) then 'C'
>            when colD = (1 = 1) then 'D'
>            else 'n/a'
>        end as phase,
>        count(*)
> from T
> where session_date between '2018-08-01' and '2018-08-31'
> group by colA
>        case
>            when colB = (1 = 1) then 'B'
>            when colC = (1 = 1) then 'C'
>            when colD = (1 = 1) then 'D'
>            else 'n/a'
>        end;
> {code}
> In the final result, all of the keys will become 'B'.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)