Dynamic Rule based implementation in PL/SQL

Question

Hi,
We are trying to implement a dynamic rule based application in Oracle 9i. Its simple logic where we store expressions as case statments and actions seperated by commas as follows.
Rule: 'Age > 18 and Age <65'
True Action: 'Status = ''Valid'' , description = ''age in range'''
False Action: 'Status =''Invalid'', Description=''Age not in range'''
Where Age,Status, description are all part of one table.
One way of implementing this is fire rule for each record in the table and then based on true or false call action as update.
i.e
select (case when 'Age > 18 and Age <65' then 1 else 0 end) age_rule from tableX
(above query will in in a cursor xcur)
Then we search for
if age_rule = 1 then
update tablex set Status = ''Valid'' , description = ''age in range'' where id=xcur.id;
else
update tablex set Status =''Invalid'', Description=''Age not in range'' where id=xcur.id;
end if;
This method will result in very slow performance due to high i/o. We want to implement this in collection based method.
Any ideas on how to dynamically check rules and apply actions to collection without impact on performance. (we have nearly 3million rows and 80 rules to be applied)
Thanks in advance

Answer 1

Returning to your original question, first of all, there is a small flaw in the requirements, because if you apply all the rules to the same table/cols, than the table will have results of only last rule that was processed.
Suppose rule#1:
Rule: 'Age > 18 and Age <65'
True Action: 'Status = ''Valid'' , description = ''age in range'''
False Action: 'Status =''Invalid'', Description=''Age not in range'''
and Rule#2:
Rule: 'Name like ''A%'''
True Action: 'Status = 'Invalid'' , description = ''name begins with A'''
False Action: 'Status =''Invalid'', Description=''name not begins with A'''
Then after applying of rule#1 and rule#2, results of the rule#1 will be lost, because second rule will modify the results of the first rule.
Regarding to using collections instead of row by row processing, I think that a better approach would be to move that evaluating cursor inside an update statement, in my tests this considerably reduced processed block count and response time.
Regarding to the expression filter, even so, that you are not going to move to 10g, you still can test this feature and see how it is implemented, to get some ideas of how to better implement your solution. There is a nice paper http://www-db.cs.wisc.edu/cidr2003/program/p27.pdf that describes expression filter implementation.
Here is my example of two different methods for expression evaluation that I've benchmarked, first is similar to your original example and second is with expression evaluation moved inside an update clause.
-- fist create two tables rules and data.
drop table rules;
drop table data;
create table rules( id number not null primary key, rule varchar(255), true_action varchar(255), false_action varchar(255) );
create table data( id integer not null primary key, name varchar(255), age number, status varchar(255), description varchar(255) );
-- populate this tables with information.
insert into rules
select rownum id
, 'Age > '||least(a,b)||' and Age < '||greatest(a,b) rule
, 'Status = ''Valid'', description = ''Age in Range''' true_action
, 'Status = ''Invalid'', description = ''Age not in Range''' false_action
from (
select mod(abs(dbms_random.random),60)+10 a, mod(abs(dbms_random.random),60)+10 b
from all_objects
where rownum <= 2
insert into data
select rownum, object_name, mod(abs(dbms_random.random),60)+10 age, null, null
from all_objects
commit;
-- this is method #1, evaluate rule against every record in the data and do the action
declare
eval number;
id number;
data_cursor sys_refcursor;
begin
execute immediate 'alter session set cursor_sharing=force';
for rules in ( select * from rules ) loop
open data_cursor for 'select case when '||rules.rule||' then 1 else 0 end eval, id from data';
loop
fetch data_cursor into eval, id;
exit when data_cursor%notfound;
if eval = 1 then
execute immediate 'update data set '||rules.true_action|| ' where id = :id' using id;
else
execute immediate 'update data set '||rules.false_action|| ' where id = :id' using id;
end if;
end loop;
end loop;
end;
-- this is method #2, evaluate rule against every record in the data and do the action in update, not in select
begin
execute immediate 'alter session set cursor_sharing=force';
for rules in ( select * from rules ) loop
execute immediate 'update data set '||rules.true_action|| ' where id in (
select id
from (
select case when '||rules.rule||' then 1 else 0 end eval, id
from data
where eval = 1 )';
execute immediate 'update data set '||rules.false_action|| ' where id in (
select id
from (
select case when '||rules.rule||' then 1 else 0 end eval, id
from data
where eval = 0 )';
end loop;
end;
Here are SQL_TRACE results for method#1:
call count cpu elapsed disk query current rows
Parse 37 0.01 0.04 0 0 0 0
Execute 78862 16.60 17.50 0 187512 230896 78810
Fetch 78884 3.84 3.94 2 82887 1 78913
total 157783 20.46 21.49 2 270399 230897 157723
and this is results for method#2:
call count cpu elapsed disk query current rows
Parse 6 0.00 0.00 0 0 0 0
Execute 6 1.93 12.77 0 3488 170204 78806
Fetch 1 0.00 0.00 0 7 0 2
total 13 1.93 12.77 0 3495 170204 78808
You can compare this two methods using SQL_TRACE.