Link to home
Start Free TrialLog in
Avatar of jflanner
jflanner

asked on

Balance Snapshot and Lag Window Function

Having read the books - I understand that a daily snapshot should - by definition - be taken every day.  I am in the specialty credit card space.  Ours is not a daily use card.  Typically - our card is used only once or twice per month.

So I can't give it up.  I don't want to insert 31 rows where only a small number of changes have occurred.  I want to implement the snapshot as a view.  I would insert into a FactChange table only when a change in balance is observed.


For the below example uses cross join, left outer join and lag to accomplish ALMOST exactly what I want - a 20 row result set addressing 2 accounts over 10 days.  The lag gets the previous value and I can coalesce for the FIRST day of inactivity.  It fails on the second, third, yada yada because the previous day as null.

What I need is lag(c.Balance, LastNonNull)  That LastNonNull is the part I can't get my head around.  Can anyone help??


create table DimDate
(
    DateKey                  int not null,
    constraint PK_Date primary key clustered (DateKey)
);
go

insert into DimDate values
(20130601),
(20130602),
(20130603),
(20130604),
(20130605),
(20130606),
(20130607), 
(20130608),
(20130609),
(20130610);
go

create table DimAccount
(
    AccountKey                  int not null,
    constraint PK_Account primary key clustered (AccountKey)
);
go

insert into DimAccount values
(1),
(2);
go

create table FactBalanceChange
(
    DateKey                  int not null,
    AccountKey               int not null,
    Balance                  money not null,
    constraint PK_BalanceChange primary key clustered (DateKey, AccountKey),
    constraint FK_BalanceChange_DateKey foreign key (DateKey) references DimDate(DateKey),
    constraint FK_BalanceChange_AccountKey foreign key (AccountKey) references DimAccount(AccountKey)
);
go

insert into FactBalanceChange values
(20130601, 1, 10.00),
(20130601, 2, 10.00),
(20130604, 1, 5.00),
(20130607, 2, 5.00),
(20130609, 1, 4.00),
(20130609, 2, 4.00);
go

create view FactBalanceSnapshot as

    select 
        d.DateKey, 
        a.AccountKey, 
        coalesce(c.Balance, lag(c.Balance, 1) over (partition by a.AccountKey order by d.DateKey)) as Balance
    from 
            DimDate d
        cross join 
            DimAccount a
        left outer join
            FactBalanceChange c
                on d.DateKey = c.DateKey
               and a.AccountKey = c.AccountKey;  
go 


select * from FactBalanceSnapshot.
 

Open in new window

Avatar of PortletPaul
PortletPaul
Flag of Australia image

does this help your thinking? Here I've placed the lag() function into a subquery on FactBalanceChange which avoids the issue of non-matching dates I believe.

I've left the query as displaying 2 columns so you may 'see' what's going on
--create view FactBalanceSnapshot as

    select 
          d.DateKey
        , a.AccountKey
        --, coalesce(c.Balance, lag(c.Balance, 1) over (partition by a.AccountKey order by d.DateKey)) as Balance
        , c.balance
        , c.lag_balance
    from 
            DimDate d
        cross join 
            DimAccount a
        left outer join (
                        select
                          fbc.dateKey
                        , fbc.AccountKey
                        , fbc.Balance
                        , lag(fbc.Balance, 1) over (partition by fbc.AccountKey order by fbc.DateKey ASC)
                            as lag_balance
                        from FactBalanceChange as fbc
                        ) as c
                on d.DateKey = c.DateKey
               and a.AccountKey = c.AccountKey;  
;

Open in new window

thanks for the tables and data - very helpful.
you can play with that data and this query at: http://sqlfiddle.com/#!6/8a56a/5
ASKER CERTIFIED SOLUTION
Avatar of PortletPaul
PortletPaul
Flag of Australia image

Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial