I guess I can assume I don't rebuild for historic data or that I only do it
for major changes which should be rare.

But what about derived dimensions? I was thinking of something like that :

Fact table with storeID and measures.

Hierarchy with

   - storeID
   - city
   - country
   - region

Derived dimension with storeID <--> storeName

Let's say the store name can change and I can have 1000's of stores
so statistically I can have changes every day. I was thinking that I can
have the storeID in the cube (this doesn't change) and the storeName in a
simple key/value table, at query time the user sees the store name but the
internal query is on the ID, the key/value table name table can be updated
every day easily and the cube stays the same. I thought derived dimensions
could be used like this. If I change a derived dimension do I need to
recompute as well?


On Tue, Jun 9, 2015 at 4:37 AM, Li Yang <[email protected]> wrote:

> It depends on whether you want the rebranding on historic data. If you hope
> the sells in the past be re-grouped according to the new branding, then
> like Shaofeng said, a rebuild is required. The reason is obvious as the
> pre-calculated sums are now invalid.
>
> However, if new branding only take effect since a certain date, then there
> no need to rebuild old cube segments. Just make sure your dimension table
> is updated on that date, and any subsequent build will start to use the new
> branding.
>
> Cheers
> Yang
>
> On Mon, Jun 8, 2015 at 10:15 PM, Shi, Shaofeng <[email protected]> wrote:
>
> > In this case, your cube need a rebuild; Kylin assumes the lookup table is
> > stable (only append new records, not change history data); Once history
> > data be changed, the cube will be out of date, the query result might be
> > wrong;
> >
> > So far there is no way to manage such changes easily I think;
> >
> > On 6/8/15, 9:33 PM, "alex schufo" <[email protected]> wrote:
> >
> > >Hello,
> > >
> > >I was wondering how the hierarchy dimensions are managed in the cube and
> > >what happens if I update the Hive table that contains the dimension.
> > >
> > >Let's say I have a fact table with a storeID and then a dimension table
> > >that goes like this :
> > >
> > >   - storeID
> > >   - city
> > >   - country
> > >   - region
> > >
> > >and I want the sum of products sold either by store, city, region, etc.
> > >
> > >This seems fine with Kylin.
> > >
> > >For regions, internally is a row going to be created with key as
> something
> > >like region and the pre-calculated sum?
> > >
> > >Now let's say there is rebranding and the region is renamed, or that the
> > >strategy changes and that a country that was part of region is now part
> of
> > >another region.
> > >
> > >Every time some of those changes occur, should I rebuild completely my
> > >cube
> > >or is there any way to manage those changes?
> >
> >
>

Reply via email to