Oracle 10g Connect By Prior - Performance Issues

Question

I have the following SQL statement:

SELECT 
    CONNECT_BY_ROOT ANIMAL_ID "ORIGINAL_ANIMAL" ,
    ANIMAL_ID, LINE_ID, SIRE_ANIMAL_ID, DAM_ANIMAL_ID,
    LEVEL -1 "LEVEL" FROM ANIMALS 
START WITH ANIMAL_ID IN( '2360000002558' )
CONNECT BY
  ((PRIOR SIRE_ANIMAL_ID = ANIMAL_ID and LEVEL < 5) OR (PRIOR DAM_ANIMAL_ID = ANIMAL_ID AND LEVEL < 5))

This in in a table with about 1.6 Million animals. Each record has Animal_Id, Sire_Animal_Id, and Dam_Animal_Id (Sire = Father, Dam = Mother).

I use this sql to display the full animal pedigree. Results Will show Animal, 2 Parent, 4 GrandParents, etc.

My issue is that this statement takes 15 seconds, for one animal. There has got to be a way to optimize this. Any thoughts?

Answer 1

I haven't had a long time to test this so there is a bit of DYOR in the answer but would using an inline view help?

As you haven't posted an explain plan I can't help too much i'm afraid and in the solution below, you may find that the union in the WITH clause causes you performance issues but it might help you on your way to a solution.

WITH ani
  AS (SELECT animal_id, 
             line_id, 
             sire_animal_id, 
             dam_animal_id, 
             sire_animal_id AS generic_id
        FROM animals
      UNION
      SELECT animal_id, 
             line_id, 
             sire_animal_id, 
             dam_animal_id, 
             dam_animal_id AS generic_id
        FROM animals)
SELECT CONNECT_BY_ROOT animal_id "ORIGINAL_ANIMAL",
       animal_id,
       line_id,
       sire_animal_id,
       dam_animal_id,
       LEVEL - 1 "LEVEL"
  FROM ani
 START WITH animal_id = '2360000002558'
 CONNECT BY (PRIOR generic_id = animal_id AND LEVEL < 5 )

Answer 2

I tried recreating your situation and I wasn't able to get Oracle to use the indexes wisely. I'm sure there's some smart way to do it. But if no one else here can figure it out, below is the dumb, ugly way.

Since you're only getting a certain number of levels you can manually create a connect by. Get the first level, union that to the second level (which gets results from a copy of the first query), union that to the third level (which gets results from a copy of the second query), etc. I only did three levels here, but you can copy and paste to make the fourth. It's harder to use since the original id is repeated so many times, but it's super fast (0.005 seconds on my machine with 1.6 million records.)

--Original animal
select '101' original_animal, animal_id, line_id, sire_animal_id, dam_animal_id, 0 "level" from animals where animal_id = '101'
union all
--Parents
select '101' original_animal, animal_id, line_id, sire_animal_id, dam_animal_id, 1 "level" from animals
where animal_id = (select sire_animal_id from animals where animal_id = '101')
union all
select '101' original_animal, animal_id, line_id, sire_animal_id, dam_animal_id, 1 "level" from animals
where animal_id = (select dam_animal_id from animals where animal_id = '101')
union all
--Grand parents
select '101' original_animal, animal_id, line_id, sire_animal_id, dam_animal_id, 2 "level" from animals
where animal_id =
(
  select sire_animal_id from animals
  where animal_id = (select sire_animal_id from animals where animal_id = '101')
)
union all
select '101' original_animal, animal_id, line_id, sire_animal_id, dam_animal_id, 2 "level" from animals
where animal_id =
(
  select dam_animal_id from animals
  where animal_id = (select sire_animal_id from animals where animal_id = '101')
)
union all
select '101' original_animal, animal_id, line_id, sire_animal_id, dam_animal_id, 2 "level" from animals
where animal_id =
(
  select sire_animal_id from animals
  where animal_id = (select dam_animal_id from animals where animal_id = '101')
)
union all
select '101' original_animal, animal_id, line_id, sire_animal_id, dam_animal_id, 2 "level" from animals
where animal_id =
(
  select dam_animal_id from animals
  where animal_id = (select dam_animal_id from animals where animal_id = '101')
);

Answer 3

Yes, there are indexes.

Answer 4

Are there indexes on sire_animal_id and dam_animal_id? It might be doing full table scans.