kylin-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From 赵天烁 <zhaotians...@meizu.com>
Subject Re: Re: support filter on a dimension which is not being grouped
Date Wed, 03 Aug 2016 01:55:02 GMT
sorry, the sql I provide is made up by myself to simulate the actual one which is to complicated
, it's my fault to add crash_point column to the select list which isn't contain in the actual
sql,but it still give me an sql compile error when I execute as below:
select  fid from crash_analysis_record where package_ver='5.1' group by fid
already update to 1.5.3 by folow the upgrade guide,is there anything else I may miss?
________________________________
赵天烁
Kevin Zhao
zhaotianshuo@meizu.com<mailto:zhaotianshuo@meizu.com>

珠海市魅族科技有限公司
MEIZU Technology Co., Ltd.
广东省珠海市科技创新海岸魅族科技楼
MEIZU Tech Bldg., Technology & Innovation Coast
Zhuhai, 519085, Guangdong, China
meizu.com

From: ShaoFeng Shi<mailto:shaofengshi@apache.org>
Date: 2016-08-03 09:03
To: user<mailto:user@kylin.apache.org>
Subject: Re: support filter on a dimension which is not being grouped
Hi Tianshuo,

Apache Kylin already support the case you mentioned;

Please note the SQL you provided (in below) is invalid, because the "crash_point" should be
in "group by" if it is a dimension, or it should be a measure like SUM(crash_point).

select  fid,crash_point from crash_analysis_record where package_ver='5.1' group by fid



2016-08-02 16:20 GMT+08:00 赵天烁 <zhaotianshuo@meizu.com<mailto:zhaotianshuo@meizu.com>>:
right now I could achieve the same result by two step query:
first ,execute a two dimensions group by query
select distinct fid,count(1) from crash_analysis_record where package_ver='5.1' group by FID,package_ver
then execute the second query to filter the result base on fid return above:
select  fid,crash_point from crash_analysis_record where fid in (1,2,3) group by fid order
by count(1) desc

but it's gona be very convenient if I could just do a filter without group by it.is<http://it.is>
it possible to support this?
btw why order by field must appear in the select field list?in the first query actually I
just want the distinct fid
________________________________
赵天烁
Kevin Zhao
zhaotianshuo@meizu.com<mailto:zhaotianshuo@meizu.com>

珠海市魅族科技有限公司
MEIZU Technology Co., Ltd.
广东省珠海市科技创新海岸魅族科技楼
MEIZU Tech Bldg., Technology & Innovation Coast
Zhuhai, 519085, Guangdong, China
meizu.com<http://meizu.com>

发件人: 赵天烁<mailto:zhaotianshuo@meizu.com>
发送时间: 2016-08-02 15:49
收件人: user<mailto:user@kylin.apache.org>
主题: support filter on a dimension which is not being grouped
currently kylin don't support filter on a dimension which is not being grouped,for example
I have a sql like this:
select  fid,crash_point from crash_analysis_record where package_ver='5.1' group by fid order
by count(1) desc
fid and package_ver are all predefine dimensions,it can be done if we find those cuboid data
with fid and package_ver, eliminate the data which are not package_ver=5.0,then runtime aggragate
to fid
I just propose some thought,this approach may have profermance issue,maybe you guys already
have a better way todo this?

________________________________
赵天烁
Kevin Zhao
zhaotianshuo@meizu.com<mailto:zhaotianshuo@meizu.com>

珠海市魅族科技有限公司
MEIZU Technology Co., Ltd.
广东省珠海市科技创新海岸魅族科技楼
MEIZU Tech Bldg., Technology & Innovation Coast
Zhuhai, 519085, Guangdong, China
meizu.com<http://meizu.com>



--
Best regards,

Shaofeng Shi

Mime
View raw message