Update Column using Rownum - oracle11g

Question

I am trying to update a column based on the rownumber. Table Unpaid1 contain multiple transactions for alot of accounts. When i execute the following script, I often end up transactions With incorrect sequential number (Cycleage) like shown below. The cycleage should be in the order 1 to 4 from oldest to newest. What might be the cause of this issue? Is there another way to assign the cycleage?
Begin
For X in (Select * From temp_Unpaid1 order by trxnserno asc)
loop
Update temp_Unpaid1 t set t.cycleage = rownum where t.custid=x.custid;
end loop;
End;
+-----------+-----------+--------+--------+----------+
| TRXNSERNO | CUSTID | AMOUNT | DATE | CYCLEAGE |
+-----------+-----------+--------+--------+----------+
| 66 | 45 | -10 | 08.jan | 2 |
| 67 | 45 | -20 | 10.jan | 1 |
| 90 | 45 | -30 | 15.jan | 3 |
| 155 | 45 | -15 | 20.jan | 4 |
+-----------+-----------+--------+--------+----------+

Answer 1

The catch is that the ROWNUM in the update is independent of the SELECT statement and, in fact, it looks like you are doing unnecessary updates. Here's why:
The SELECT will return all rows in the order of TRXNSERNO and even if you have only the one custid, you'll get 4 rows with the same custid:
CUSTID TRXNSERNO ...(other columns)
45 66
45 67
45 90
45 155
Then for each of these rows, the update statement will run (so you'll be updating the same 4 rows 4 times.)
When the UPDATE runs, it finds the 4 rows for custid 45 (it doesn't know/care what the TRXNSERNO was in the SELECT) and updates them in whatever order it finds them. (There's nothing in the UPDATE statement that forces the order of the rows as they are updated.)
So, whatever order the (4th and last) update found them in is how the ROWNUM values will be assigned so if Oracle found them in the order of
CUSTID TRXNSERNO
45 67
45 66
45 90
45 155
Then the ROWNUM value will be in that order:
CUSTID TRXNSERNO ROWNUM
45 67 1
45 66 2
45 90 3
45 155 4
In another run (depending on circumstances) Oracle might execute the update and find the custid 45 rows in a completely different order like the following and assign them rownums accordingly:
CUSTID TRXNSERNO ROWNUM
45 155 1
45 66 2
45 90 3
45 67 4
You could keep your current code but just generate the ROWNUMs in the SELECT list and apply the value in the update to the custid/trxnserno pairs:
Begin
For X in
(Select tmp.*
, row_number() over (partition by tmp.custid order by tmp.trxnserno) as rn
From temp_Unpaid1 tmp)
loop
Update temp_Unpaid1 t
set t.cycleage = rn
where t.custid=x.custid
and t.trxnserno = x.trxnserno;
end loop;
End;
Alternatively, you could try this in a single SQL statement using something like MERGE and using only the WHEN MATCHED portion of it:
MERGE INTO temp_unpaid1 temp
USING
(Select t.*
, row_number() over (partition by custid order by trxnserno) as rn
From temp_Unpaid1
) rn_qry
ON (temp.custid = rn_qry.custid
AND temp.trxnserno = rn_qry.trxnserno
)
WHEN MATCHED THEN
UPDATE SET
temp.cyclage = rn_qry.rn
;
Note: unfortunately I can't test the accuracy of the syntax right now, but hopefully this is of some value. :)