Deleting "n" Rows From a Db2 Table

I regularly receive database- and Db2-related questions via e-mail. And that is great, but I don't always get a chance to respond to everything. If you've sent me a question and I haven't replied, I apologize. But every now and then, I will use one of the e-mail questions in my in-box and write about it in the blog. Today's question is this:

How do you delete N rows from a Db2 table?

Also, how do you retrieve bottom N rows from a Db2 table without sorting the table on key?

And here is my response:

First things first, you need to refresh your knowledge of "relational" database systems and Db2. There really is no such thing as the "top" or "bottom" N rows in a table. Tables are sets of data that have no inherent logical order.

With regard to the result set though, there is a top and a bottom. You can use the FETCH FIRST N ROWS ONLY clause to retrieve only the first N rows, but to retrieve only the bottom N rows is a bit more difficult. For that, you would have to use scrollable cursors.

A scrollable cursor allows you to move back and forth through the results set without first having to read/retrieve all of the rows before. I suggest that you read up on scrollable cursors in the Db2 SQL Reference manual and the Db2 Application Programming manual. All Db2 manuals can be downloaded in Adobe PDF format for free over the IBM web site.

Basically, you would want to FETCH LAST from the scrollable cursor and then loop through with a FETCH PRIOR statement executing the loop N-1 times. That would give you the "bottom" N of any results set -- sorted or not.

As for your other question, I am confused as to why you would want to delete N rows from a table. Doesn't it matter what the data in the rows is? My guess is that you are asking how you would limit a DELETE to a subset of the rows that would apply to the WHERE condition of the DELETE. The answer is, you cannot, at least not without writing some code.

You would have to open a cursor with the same WHERE conditions specifying FOR UPDATE OF. Then you would FETCH and DELETE WHERE CURRENT OF cursor for that row in a loop that occurs N times. Of course, that means you have to write a program to embed that SQL in.

Hope this answer helps...

The DB2 12 for z/OS Blog Series - Part 9: Piece-wise Deletion

Adding the FETCH FIRST clause to the DELETE statement at first appears to be a very simple enhancement, but upon closer examination it is really quite powerful.

Prior to DB2 12 for z/OS, the FETCH FIRST n ROWS ONLY clause could be specified on a SELECT statement. The clause had two impacts:

1. the number specified for n is used by the optimizer to formulate an access path
2. the result set of the SELECT is limited to no more than n rows

With the advent of FETCH FIRST being allowed in a DELETE statement, the number n limits the number of rows that will be deleted in a single DELETE statement. So let's assume that there are 1000 rows for employees in department 200. When we issue this DELETE statement

DELETE FROM EMP

WHERE  EMPDEPT = '200';

All 1000 rows will be deleted. However, if we issue this version of the same statement

DELETE FROM EMP

WHERE  EMPDEPT = '200'

FETCH FIRST 500 ROWS ONLY;

Only 500 rows will be deleted... you could then run it again to delete the remaining 500 rows.

So why would you want to do this? Well, our little example here is not really a good case for using FETCH FIRST on DELETE . Instead, it is primarily designed for situations where a large number of rows would be impacted. For example, assume that instead of 1000 rows there were 2 millions rows. Using FETCH FIRST  to DELETE the rows in batches, instead of 2 million all at once, can make an impossible task possible. The lock management when deleting 2 million rows can render a big, bulk deletion unwieldy as it impacts concurrent access of the data on the same pages where rows are being deleted.

So keep FETCH FIRST  in your arsenal of DB2 12 SQL tools that can help when you need to DELETE  a large number of rows.