Row_number() issue - exponentially increases the record count

Question

Hello all -
I'm trying to use the row_number to split the amount by the number of months between the dates and create one new row for the each month.
The use of row_number() is causing the data to be exponentially increase based -i.e currently there are 3 distinct records and each record should only return 2 rows but the this returns 12 rows and so on if the data increases which is causing a performance issue . I could put the entire query below again as a derived table and add a where condition as where rn < = tot_months but I just want to know if there is any way to limit the rows to the number of months reason being - for 48 distinct rows it is creating apprx 12000 rows and the performance has become a killer. You can try by adding more rows and see what I mean especially when the span between months is more -
Any help is greatly appreciated -
WITH testdata AS
(SELECT 1 id, 1001 acct_no, 2001 cnt_id, to_date('01/17/2007','mm/dd/yyyy') start_date, to_date('03/16/2007', 'mm/dd/yyyy') end_date, 8000 amount
FROM dual
UNION ALL
SELECT 2 id, 1002 acct_no, 2002 cnt_id, to_date('01/18/2007','mm/dd/yyyy') start_date, to_date('03/19/2007', 'mm/dd/yyyy') end_date, 90000 amount
FROM dual
UNION ALL
SELECT 3 id, 1003 acct_no, 2003 cnt_id ,to_date('02/08/2007','mm/dd/yyyy') start_date, to_date('04/11/2007', 'mm/dd/yyyy') end_date, 125000 amount
FROM dual)
SELECT DISTINCT id,
acct_no,
cnt_id,
start_date,
end_date,
amount,
row_number() over(PARTITION BY id , start_date ORDER BY cnt_id) rn,
ROUND(months_between(end_date, start_date)) tot_months
FROM
(SELECT id,
acct_no,
cnt_id,
start_date,
end_date,
amount
FROM testdata t
where ROUND(months_between(end_date, start_date)) > 1 )
t CONNECT BY LEVEL <= ROUND(months_between(end_date, start_date))
order by acct_no,cnt_id, start_date
Thanks.

Answer 1

Then why is the query returning multiple rows instead
of 2 per account. Because you have in no way specified that the query must return 2 rows per account. You need to fully understand the implications of what you are doing.
CONNECT BY <= SOME_INTEGER is not just a generic method of "multiplying" rows by an integer value. True, it works in certain circumstances when used in a certain way, but there is still debate about whether it is even a valid SQL construct. See Multi-Rows from DUAL (and many others) for different approaches to generating rows.
But let's understand exactly what your query is doing. In essence, you are doing something similar to the following,where the ID signifies the number of months between start_date and end_date in your example:
  1  with testdata as (
  2     select 1 id from dual union all
  3     select 2 id from dual
  4  )
  5  select lpad(t.id, 2*level, ' ') as id
  6  , level
  7  from testdata t
  8* connect by level <= t.id
[email protected](162)> /
ID              LEVEL
1                  1
   2                2
2                  1
   2                2I've included the LEVEL pseudocolumn in the output and formatted by the depth of the tree so we can better see what's going on. The hierarchical query is processed in the following order:
1. Select the root rows of the hierarchy. In our case this includes all rows in the test table. So, for the both ID = 1 and ID = 2, we descend to LEVEL = 1, i.e. the root level, and we return both rows with LEVEL = 1.
2. Increment the LEVEL pseudocolumn and select the child rows of each root row, i.e. those rows which satisfy the CONNECT BY condition.
So we increment LEVEL to 2. For the both the rows, we descend to the CONNECT BY condition LEVEL = 2. Only ID = 2 satisfies the LEVEL <= ID condition, so this row is returned as a child of both ID = 1 and ID = 2.
So we see, the row with ID = 2 is both the parent AND the child for one of the returned rows. Note that this is, strictly speaking, a LOOP in the data and should raise the ORA-01436: CONNECT BY loop in user data error, but it does not. In my opinion this is a BUG, hence the "debate" mentioned above..
3. We return the rows with parents followed directly by their children.
So, in your general case of:
with t as (
   select N mnths -- average no of months
   from T -- T is a table with R rows
select *
from t
connect by level <= t.mnths, you'll return:
- R rows for level = 1
- R*R rows for level = 2
- R*R*R rows for level = 3
So your general number of rows is Rtot = R + R^2 + R^3 + ... + R^N
For 48 rows with 2 months per row, this gives 48 + 48^2 = 2352 rows.
For 48 rows with 3 months per row, this gives 48 + 48^2 + 48^3 = 112944 rows.
  1  with testdata as (
  2     select 3 id from dual union all
48     select 3 id from dual union all
49     select 3 id from dual
50  )
51  select lpad(t.id, 2*level, ' ') as id
52  , level
53  from testdata t
54* connect by level <= t.id
[email protected](162)> /
112944 rows selected.So you can see, your number of rows returned will grow extemely quickly if you have 3 or more months between start date and end date.
cheers,
Anthony