Tricky query optimisation

Question

Hi all,
our task is to develop a new application based on the data
model of an old application. As the old application must remain
available and running, no changes are possible to the tune the
database! (currently running on 9.2.0.7)
We've put together this example code describing the problem:
-------------------- snip --------------------
-- clean up last run
drop table shares;
drop table files;
drop table families;
-- families table
create table families (
family_id integer,
primary key (family_id)
-- files table: each file belongs to one family
create table files (
file_id integer,
family_id integer not null,
primary key (file_id),
foreign key (family_id) references families (family_id)
-- shares table: shares can be defined on family basis or on file basis.
-- If defined on family and file, only the file shares are relevant!
create table shares (
file_id integer,
family_id integer not null,
person varchar2(100) not null,
percentage number(4,2) not null,
unique (file_id, family_id, person, percentage),
foreign key (file_id) references files (file_id),
foreign key (family_id) references families (family_id)
-- insert one family and two files with shares for demonstration purposes
insert into families values (1);
insert into files values (1,1);
insert into files values (2,1);
insert into shares values (null, 1, 'Thomas',33.4);
insert into shares values (null, 1, 'Marcus',66.6);
insert into shares values (2, 1, 'Thomas',50);
insert into shares values (2, 1, 'Marcus',50);
commit;
-- Goal: determine the shares for each file
-- good query:
select s.family_id, s.file_id, s.person, s.percentage
from shares s
where s.file_id is not null
union
select p.family_id, p.file_id, s.person, s.percentage
from shares s, files p
where p.family_id = p.family_id
and s.file_id is null
and not exists (select 1
from shares p2
where p2.file_id = p.file_id);
-- poor query:
select distinct p.family_id, p.file_id, s.person, s.percentage
from files p, shares s
where p.family_id = s.family_id
and (s.file_id is null or s.file_id = p.file_id)
order by p.family_id, p.file_id;
-------------------- /snip --------------------
Now here's the hook!
Amount of data (rows)
- families: 116.000
- files: 633.000
- shares: 423.000
- avg persons per family: 2.1
We were using the queries above in views for hiding the
complexity to our newly developed application.
A test case containing the view and several other tables performed
like this:
- execution time when view used the poor query: ~ 0.3 seconds
- execution time when view used the good query: ~ 37 seconds
We know the design has it flaws but it evolved with time and
proofed to be useful for the old application.
So my question is: does anyone have a clue how we can combine
performance and correctness within one statement? Any comment
is appreciated!

Answer 1

I'm with shoblock. If the "good" query returns the wrong resutls, then its unuable and not really good.
I'm afraid you're stuck with tuning the long-running query which returns the right results.