Pulling out two records at a time

Question

I have a TABLE with multiple records in it, of which i want to pull out two records at a time (as if they were one record), of those records that meet the WHERE criteria. Order is not important. Basically, i have a TABLE of things that wither need processing or have been processed already. I can process two at a time. (In a related search, that i am not up to yet, i can do 20 at a time.) How would i go about doing that?
My first thought was a CTE with ROW_NUMBER() RN, using CONNECT BY RN = PRIOR RN + 2, and have a sub-query (or joined TABLE) pull out the second record:
WITH Moo (SELECT ... ROW_NUMBER RN FROM Table WHERE ...)
SELECT ..., (SELECT ... FROM FROM Moo Moo2 WHERE Moo2.Id = Moo.Id + 1)
FROM Moo CONNECT BY Id = PRIOR Id + 2;
Or a join. Is there a better way to approach it?

Answer 1

Pretty much everything you just said is wrong.
>
I generally keep away from CURSORs, as i want the DB to handle the set processing for me, as opposed to a CURSOR which tells the DB how to do it's job.
>
If you do any DB queries then you are using cursors. You have the wrong idea about them. All result sets are obtained using a cursor. A cursor does NOT tell a DB how to do a job; that is what the query and execution plan do.
>
(And, Tom Kyt's mantra of "You should do it in a single SQL statement if at all possible") But that's just a rule of thumb.
>
No - that isn't a rule of thumb. That is exactly what you should do if possible which is what I said. But I also said
>
Well the general use is to use SQL instead of PL/SQL but since you already need to use a function you might as well make that function as efficient as possible.
>
If you are already using a function then you should get the most bang for the buck from it and make it as efficient as possible.
>
If the CURSOR is the most efficient, so be it. But as you point out, if i'm already going to use a CURSOR, i might as well do all the work in the CURSOR LOOP. Why bother switching contexts.
>
All queries are done using cursors. It isn't a question of whether they are more efficient; it is the only way possible. I think you are referring to a user creating an explicit cursor and performing explicit manipulation of that cursor. That isn't normally what you want to do.
A cursor loop is the slowest way to process results especially if all you are doing is row-by-row processing to perform essentially SQL-related work. PL/SQL should be used for the transactional part of the processing. In your case that transactional use is consolidating two records together for specialized processing.
As for context switches if you perform any PL/SQL you are going to have a context switch: functions, procedures, triggers. The function you already planned to use would have caused a context switch. What I am saying is that if you are going to perform a context switch then you want to maximize the amount of work (transactional work) done in the PL/SQL engine so that you minimize the context switching.
You still haven't explained exactly why 2 records at a time is any more ideal than 1 or if you could also process 50, 100, or 10000 at a time. Is the limit what needs to be passed somewhere else? What is the reason for a max of two? What exactly are you doing with the data to prepare it before you have to pass it on?
Bulk collect processing is more efficient, with fewer context switches, if you process 20, 50, or 100 records in the loop at a time.
See Chap 13 Using Pipelined and parallel Table Functions in the Data Cartridge Dev Guide
http://docs.oracle.com/cd/B28359_01/appdev.111/b28425/pipe_paral_tbl.htm