Hi Meng, kylin have not use multiple columns store multiple measures just only F1:M so kylin using "rowkey:dim1+dim2, value:sum1+","+count2+",??+sum3?? as you list
Best regards Steven Zhang > On Aug 28, 2015, at 10:59 AM, [email protected] wrote: > > sorry about my poor english, what i means is: > > example: dim1,dim2,sum1,count2,sum3 > > will kylin store the aggr result like: > rowkey:dim1+dim2,value:sum1 > rowkey:dim1+dim2,value:count2 > rowkey:dim1+dim2,value:sum3 > > or: > rowkey:dim1+dim2, value:sum1+","+count2+","+sum3 > > the second way seem to save a lot of storage because it only store rowkey > once; > > > > ???????????????????? ???????? ???? > [email protected] > > ???????? steven zhang > ?????????? 2015-08-28 10:33 > ???????? dev > ?????? Re: one question, how kylin store data in hbase? > Hi Meng, > > To my understanding: > > kylin using RowKey:=cuboid+dimensions RowValue:=measures(name+value) > > For example > select class_id,sum(score) as total_score from student_score group by class_id > > 1.kylin will generate SQLDigest(include parameters) when execute sql > > 2.If the dimension index of class_id is 4 so the cuboid is 00010000 = 16 > ,cuboid length = 8 bytes ??00100000" is the last byte and the rest of rowkey > store the dimension value of class_id > > 3.How many rows is affected by numbers of dimensions . if dimensions > numbers = 5 so the combination is 2^5 = 32 > > 4.Hbase scan range (startRowKey and stopRowKey). if group by > dimension1,dimension2 (dimension1 index=7 , dimension2 index=5) so > startRowKey = 0000000 96 00 stopRowKey = 0000000 96 FF0 > and the measure value is after group by dimension1,dimension2 (high??>low) > . calculate the item such as SUM(Price),Min(Price) etc. > > Best regards > > Steven Zhang > > >> On Aug 28, 2015, at 9:20 AM, [email protected] wrote: >> >> for example??if i have 3 calculation item in cube, will kylin store each >> dimention of calculations in 3 rows or only 1 row(and split it when query)? >> >> >> >> >> ???????????????????? ???????? ???? >> [email protected] >
