在rownum或with子句之间的Oracle 10g性能

Question

I am using Oracle Database 10g Enterprise Edition 10.2.0.4.0 64bit

and I would like to know, what is the best way to write the following query?

1. With rownum

SELECT * FROM
  (
  SELECT ID_DONNEE_H, DATE_DONNEE
  FROM DONNEE_H d
  WHERE d.DATE_DONNEE > sysdate -50000
  AND d.ID_SC = 38648
  ORDER BY DATE_DONNEE DESC
  )
  WHERE rownum=1;

2. With a WITH clause

with req as (
  select d.ID_DONNEE_H, row_number() over (order by DATE_DONNEE desc) as seqnum
  from DONNEE_H d
  where d.DATE_DONNEE > sysdate -50000
  AND d.ID_SC = 38648 ) 
select * from req where seqnum = 1;

3. With rank clause

select * from (select d.ID_DONNEE_H, row_number() over (order by DATE_DONNEE desc) as seqnum
  from DONNEE_H d
  where d.DATE_DONNEE > sysdate -50000
  AND d.ID_SC = 38648) test
  where seqnum = 1;

I think 2 and 3 are similar, but which is the fastest, 1, 2 or 3?

Answer 1

I don't think you can particularly generalise as to which query is the "best" in all situations. As with most IT questions, the answer is: "it depends"! You would need to investigate each query on its own merits.

As an aside, you've missed off another alternative - assuming that you're only after a single column, rather than the whole row for the highest column you're interested in:

with sample_data as (select 10 col1, 3 col2 from dual union all
                     select 20 col1, 3 col2 from dual union all
                     select 30 col1, 1 col2 from dual union all
                     select 40 col1, 2 col2 from dual)
select max(col1) keep (dense_rank first order by col2 desc) col1_val_of_max_col2
from   sample_data;

COL1_VAL_OF_MAX_COL2
--------------------
                  20