Is there a maximum number of tables for a SELECT?

Question

I know that technically there is nothing preventing me from joining many tables. But I know there are other "limits" like memory, or processing time allowable. I've also read from Tom kyte's advice that as much as possible if it can be done in one SQL statement, do it in SQL. But is there a limit to this?
I have 10 tables that I need to work with. 2 of these tables contain millions of records. At the minimum, I need to at least join 4 of these tables then I could loop for each record and perform processing with the other tables. I could actually join more than 4 tables but I'm wary if it could still cause the system to crash. I'm using BULK COLLECT with LIMIT 1000 to control the fetching. The columns I'm joining are mostly PKs of each table. There are some transactional processing needed so I can't totally avoid a loop.
This is the logic:
<Cursor with joining 4 tables>
Loop
Fetch cursor BULK COLLECT INTO nested table LIMIT 1000
For each record Loop
Per record processing. I'll use the record elements to query from the other 6 tables.
End Loop
EXIT WHEN cursor NOTFOUND
END LOOP
Is it advisable to transfer as many tables as I can in the outside loop so there would be less processing in the inner loop? Or can a query be too big that it is better to manage the number of tables in a SELECT?
Thanks!
Edited by: user12090980 on Jun 3, 2011 6:23 AM

Answer 1

user12090980 wrote:
I have 10 tables that I need to work with. 2 of these tables contain millions of records. At the minimum, I need to at least join 4 of these tables then I could loop for each record and perform processing with the other tables. I could actually join more than 4 tables but I'm wary if it could still cause the system to crash. I'm using BULK COLLECT with LIMIT 1000 to control the fetching. The columns I'm joining are mostly PKs of each table. There are some transactional processing needed so I can't totally avoid a loop.Well, the very same loop structure you code in PL/SQL to "manually join" the required data sets of the 10+ tables, is also implemented in the SQL engine. It is called a nested loop.
So why think that your code can handle joining 10+ tables better than the SQL engine? Also consider the fact that the SQL engine has a number of other (more sophisticated) join algorithms it can use, like hash joins, merge joins and so on.
The biggest problem with the nested loop algorithm is scalability. If you nest the loop 10 deep - then a single loop iteration more in the main loop, can cause a 1000's or even potentially million more iterations in total ("exponential" impact on the loops nested in it).
So nested loop is at times the worse type of join algorithm to use. And IMO, always the wrong thing to code in PL/SQL - as PL/SQL code (irrespective of bulk processing) will always be inferior to the SQL engine when it comes to joining data.