For performance purposes

Question

Hello friends at www.oracle.com ,
let's suppose you're updating a big set of rows. In this situation, you're updating a row, and going to the next row to update it.
If the PL/SQL block meets a situation where its variables will do an useless update - for example, update the expression "TEST123" to "TEST123" (that is, will update that row to the same thing), what's better: let Oracle update it anyway, or verify that row through a SELECT statement to be sure you're not making an useless update?
Best regards,
Franklin Goncalves Jr.

Answer 1

Franklin:
I assume that you have a structure similar to:
BEGIN
   FOR rec in (SELECT oldval,newval FROM table) LOOP
      UPDATE t
      SET col = rec.newval
      WHERE col=rec.oldval
   END LOOP;
   COMMIT;
END;now, maybe oldval and newval are derived in some complex way somewhere else in your procedure, but the principle is the same.
If you are updating a column to a new value based on the curernt value of the same column, then the most efficient way to code the conditional would be something like:
BEGIN
   FOR rec in (SELECT oldval,newval FROM table) LOOP
      IF oldval <> newval THEN
           UPDATE t
           SET col = rec.newval
           WHERE col=rec.oldval
        END IF;
   END LOOP;
   COMMIT;
END;So, for every oldval, you will have one additional operation to check whether the update should be done. If the majority of the cases would result in the update being done, then you are essentially doing it for no purpose. There are two cases where it may save some time. If the majority of the cases would result in the update not being done, then it may save a little time. If the update is "expensive", that is, it takes a relatively long time to complete, then again, you may save some time.
If you are updating one column based on the value of another, then you need to do something like:
BEGIN
   FOR rec in (SELECT col1val,col2val FROM table) LOOP
          SELECT count(*) INTO l_count
          FROM t
          col1 = rec.col1val and
          col2 <> rec.col2val;
          IF l_count <> 0 THEN
           UPDATE t
           SET col2 = rec.col2val
           WHERE col1=rec.col1val and
                 col2 <> rec.col2val;
        END IF;
   END LOOP;
   COMMIT;
END;Now, you are doing a SELECT plus a conditional check for each iteration of the loop. Here, even if the UPDATE is expensive, the SELECT will be equally expensive. If you need to make the UPDATE, then you have done the expensive operation twice, and even if you do not need to update, you have done the operation once. Without the check, you have only done the operation once per iteration.
Generally, I would just do the update, and not worry about whether it needs to be done or not.
HTH
John