I think I understand. So if I wanted to group by “country” Kylin would group by “country, state, city” and then roll up? I suppose that’s OK if state and city are of medium cardinality, or if people never want to access by country.
It would be more compelling if you could find an example where no one ever wanted to group by X and Y separately, only X and Y together. > On Jul 4, 2016, at 2:46 AM, ShaoFeng Shi <[email protected]> wrote: > > Hi Julian, > > Say I have a cube which has many dimensions, among them there are 3 > dimensions: "country", "state", "city"; As we know, these three dimensions > are a hierarchy, so in the past we would suggest user to declare them as a > "hierarchy" relationship in Kylin, then Kylin will calculate the following > combinations (for these 3 dimensions): > > * > country > country, state > country, state, city > > Using "hierarchy" helps to reduce the combinations from 2^3 to 4, this is > good; > > While, my report or dashboard will group by (or filter on) all these 3 > dimensions when doing location-based analysis; this means some combinations > with "hierarchy" are usless; In this case, I would declare them as a "joint", > then Kylin will calculation these combinations (for these 3 dimensions): > > * > country, state, city > > You can see now the combination number is 2, that's a further optimization on > "hierarchy", quite simple and effective. > > Hope this can help. > > > > 2016-07-02 0:20 GMT+08:00 Julian Hyde <[email protected] > <mailto:[email protected]>>: > A real(ish) world example would help me understand. Can you give an example > of a joint dimension in terms of sales, customers, products, etc.? > >> On Jul 1, 2016, at 2:01 AM, ShaoFeng Shi <[email protected] >> <mailto:[email protected]>> wrote: >> >> "joint" just tell Kylin to prune certain combinations; for example, "joint >> AB" will prune all combinations that only has A or only has B; the order in >> "joint" doesn't matter; >> >> for case 3), it should not be allowed on Kylin's GUI, can you verify? >> >> >> >> 2016-07-01 14:47 GMT+08:00 Richard Calaba (Fishbowl) <[email protected] >> <mailto:[email protected]>>: >> Ok, so AB are joined, if: >> >> >> >> 1) Both A and B have to be specified in either in WHERE-clause or GROUP >> BY-clause ; or one in WHERE and the other in GROUP-BY >> >> >> >> 2) The order of the Joint Dimension is or is NOT important ??? If AB is >> joint, then BA is also joint, right? >> >> >> >> Meaning "select ... group by B, A" or "select ... where B = xx group by A" >> is also valid for AB as joint-dimension ??? >> >> >> >> 3) If AB is joint and AC is joint: >> >> a. It does NOT mean that ABC is necessarily joint group, right ??? >> >> b. Also BC doesn’t have to be joint, correct ?? >> >> >> >> Thanx, Richard. >> >> >> >> From: ShaoFeng Shi [mailto:[email protected] >> <mailto:[email protected]>] >> Sent: Thursday, June 30, 2016 7:45 PM >> To: [email protected] <mailto:[email protected]> >> Subject: Re: Joint VS Derived >> >> >> >> This slide has introduction on the Derived Dim: >> http://www.slideshare.net/YangLi43/design-cube-in-apache-kylin >> <http://www.slideshare.net/YangLi43/design-cube-in-apache-kylin> >> >> >> "Joint" is introduced since 1.5.1; Using "joint" to combine multiple >> dimensions into "one" ; Say if you have Dim A, B, C; among them AB are >> always appeared together, e.g, "select ... group by A, B" or "select ... >> where A = xx group by B"; in this case you can declare AB as "joint"; for >> Kylin it looks like a 2 dim cube: AB, C, the combinations are decreased from >> 2^3 to 2^2; >> >> >> >> 2016-06-30 6:56 GMT+08:00 Alberto Ramón <[email protected] >> <mailto:[email protected]>>: >> >> Hi >> >> I don't understand the difference between: >> >> - Joint Dim (from Dimension Step ) >> - Derived Dim (from Advance Setings Step) >> >> Some example ? :) >> >> >> >> >> >> >> -- >> >> Best regards, >> >> >> >> Shaofeng Shi >> >> >> >> No virus found in this message. >> Checked by AVG - www.avg.com <http://www.avg.com/> >> Version: 2016.0.7640 / Virus Database: 4613/12530 - Release Date: 07/01/16 >> >> >> >> >> -- >> Best regards, >> >> Shaofeng Shi >> > > > > > -- > Best regards, > > Shaofeng Shi >
