Tuning sql with analytic function

Question

Dear friends I've developed one sql :
with REP as
(select /*+ MATERIALIZE */ branch_code,
   row_number() over(partition by branch_code, account order by bkg_date desc  ) R,
         account,
         bkg_date,
         lcy_closing_bal
    from history t
select REP1.branch_code,
       REP1.account,
       REP1.bkg_date,
       REP1.lcy_closing_bal,
         NULL  AS second,
       REP2.bkg_date        bkg_date2,
       REP2.lcy_closing_bal lcy_closing_bal2,
        NULL  AS third,
       REP3.bkg_date        bkg_date3,
       REP3.lcy_closing_bal lcy_closing_bal3
  from (SELECT * FROM REP WHERE R=1) REP1, (SELECT * FROM REP WHERE R=2) REP2, (SELECT * FROM REP WHERE R=3) REP3
where
       (REP1.BRANCH_CODE = REP2.BRANCH_CODE(+) AND REP1.ACCOUNT = REP2.ACCOUNT(+)) AND
       (REP1.BRANCH_CODE = REP3.BRANCH_CODE(+) AND REP1.ACCOUNT = REP3.ACCOUNT(+))The point is I want to restrict (tune) REP before it used ,because , as you can see I need maximum three value from REP (where R=1,R=2,R=3) . Which analytic function and with wich options I have to use to receive only 3 values in each branch_code,account groups at the materializing time ?

Answer 1

Radrigez wrote:
Dear friends I've developed one sql :
with REP as
from (SELECT * FROM REP WHERE R=1) REP1,
(SELECT * FROM REP WHERE R=2) REP2,
(SELECT * FROM REP WHERE R=3) REP3
where
(REP1.BRANCH_CODE = REP2.BRANCH_CODE(+) AND REP1.ACCOUNT = REP2.ACCOUNT(+)) AND
(REP1.BRANCH_CODE = REP3.BRANCH_CODE(+) AND REP1.ACCOUNT = REP3.ACCOUNT(+))
The first step is to put your subquery (which doesn't need to be materialized) into an inline view and restrict the result set on r in (1,2,3) as suggested by thtsang - you don't need to query the same result set three times.
Then you're looking at a simple pivot operation (assuming the number of rows you want per branch and account is fixed). If you're on 11g search the manuals for PIVOT, on earlier versions you can do this with a decode() or case() operator.
Step 1 (which could go into another factored subquery) would be something like:
select
        branch_code, account,
        case r = 1 then bkg_date end bkg_date,
        case r = 1 then lcy_closing_bal end lcy_closing_bal,
        case r = 2 then bkg_date end bkg_date2,
        case r = 2 then lcy_closing_bal end lcy_closing_bal2,
        case r = 3 then bkg_date end bkg_date3,
        case r = 3 then lcy_closing_bal end lcy_closing_bal3
from
        repThis gives you the eight necessary columns, but still (up to) three rows per branch/account.
Then you aggregate this (call it rep1) on branch and account.
select
        branch_code, account,
        max(bkg_date),
        max(lcy_closing_bal),
        max(bkg_date2),
        max(lcy_closing_bal2),
        max(bkg_date3),
        max(lcy_closing_bal3)
from
        rep1
group by
        branch_code, account
order by
        branch_code, accountRegards
Jonathan Lewis
http://jonathanlewis.wordpress.com
Author: <b><em>Oracle Core</em></b>