Queries with custom functions

Question

Hi, Oracle gurus,
I experienced recently an issue with SQL performance when playing with queries using PL/SQL functions.
Let’s look at the following query:
SELECT my_package.my_function(x)
FROM Xvalues
This query returns values calculated by a PL/SQL function “my_function” in “my_package” package. The function performs some CPU intensive computations so that it takes it approx. 1 second to return a value. If the “Xvalues” table contains 100 rows then it takes Oracle 100 seconds to complete the query. OK.
Now I want to filter only non null results so that my query may look as:
SELECT y
FROM (
SELECT my_package.my_function(x) y
FROM Xvalues
WHERE y IS NOT NULL
It was surprising for me that it took Oracle ca. 200 seconds to complete this query, twice as long as the original one. I believe that the optimizer resolves this query to something like this:
SELECT my_package.my_function(x) y
FROM Xvalues
WHERE my_package.my_function(x) IS NOT NULL
When processing such a query Oracle first evaluates the WHERE clause to decide to return output values specified in the SELECT clause or not. That’s logical in most situations, but in this particular case it leads to longer execution time.
What I do to workaround this problem is splitting the query. First I put all values into a temporary table using
INSERT INTO TempTable
(SELECT my_package.my_function(x)
FROM Xvalues)
and then select non null results with another query
SELECT y
FROM TempTable
WHERE y IS NOT NULL
The whole execution time shrinks to original ca. 100 seconds what I wanted to achieve. Nevertheless I am not satisfied with this solution because it introduces some intermediate steps. This is only a simplified example but in fact my query is much more complex, consists of many nested subqueries, etc. and this approach – although effective – would require dozens of temporary tables and make the whole query unclear.
I wonder if there is a simpler solution possible, for example to have one single hierarchical query like above one and force Oracle first to fully process the nested subquery and then the embracing one? The goal is to decrease waiting for filtered results down to original 100 seconds (or a bit longer) …
Any suggestions will be greatly appreciated!
Greg

Answer 1

Maybe Tom Kyte's "trick" of the "select pl_sql_function () from dual" to invoke scalar subquery caching can help you:
select ename
  from emp, dept
where (select dname_lookup(emp.deptno) from dual) = 'SALES'
   and dept.deptno = 30
   and emp.deptno = dept.deptno;http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:1547006324238
As Tom says, in 11g you'll be able to cache that function (new 11g feature) which removes the need for the "trick":
Oracle® Database PL/SQL Language Reference 11g Release 1 (11.1)
- 8 Using PL/SQL Subprograms
-- Using the Cross-Session PL/SQL Function Result Cache
http://download.oracle.com/docs/cd/B28359_01/appdev.111/b28370/subprograms.htm#BABFHACJ
"On the PL/SQL Function Result Cache" by Steven Feuerstein
http://www.oracle.com/technology/oramag/oracle/07-sep/o57plsql.html
Regards,
Zlatko