1. To get precise result, top n query has to scan a full cuboid (not cube).
There are very fast estimates result if 100% correctness is not required.

2. You are correct, that's why we recommend list filtering dimensions
before the rest in
http://www.slideshare.net/YangLi43/design-cube-in-apache-kylin

Cheers

On Thu, Apr 9, 2015 at 9:11 PM, 林澍荣 <[email protected]> wrote:

> hi, kylin team,
> 1. does a topk query scan all the data in the cube table?
> 2. does the order of the dimensions in rowkey effect the query speed? for
> example, a rowkey consists of A,B,C dimension in order, and will a query
> with "WHERE A=a" be finished faster than another with "WHERE C=c"? In my
> view, as the hbase table rows are sorting in the lexicographical order, and
> the former query will result in a partial scan with a rowkey scale from "a"
> to "aa", while the latter one will result in a full scan. But I don't know
> whether the hbase coprocessor will do anything to make the latter one
> efficient. I'd like to know how Kylin deals with it.
>
> if I get anything wrong, I will appreciate any correction.
>
> thanks, Shon
>

Reply via email to