Query performance using distinct

Question

Greetings! We're on Oracle 8.1.7, Solaris 2.8.
I have a query that utilizes a different access path if I use the word distinct in the select from this view. Here is our query:
SELECT
DISTINCT SETID,
VENDOR_ID,
VENDOR_NAME_SHORT,
AR_NUM,
NAME1,
ADDRESS1,
ADDRESS2,
CITY
FROM
PS_VENDOR_VW
WHERE
SETID LIKE 'MNSA' AND
NAME1='FUN' ORDER BY NAME1, SETID, VENDOR_ID
The view SQL is:
SELECT /*+ FIRST_ROWS */ a.setid
, a.vendor_name_short
, a.name1
, c.address1
, c.address2
, c.city
, a.vendor_id
FROM SYSADM.ps_vendor a
SELECT /*+ INDEX_ASC(B PSAVENDOR_ADDR) */
FROM SYSADM.ps_vendor_addr b
WHERE b.effdt = (
SELECT MAX(effdt)
FROM SYSADM.ps_vendor_addr
WHERE setid = b.setid
AND vendor_id = b.vendor_id
AND address_seq_num = b.address_seq_num
AND effdt <= sysdate)) c
WHERE a.setid = c.setid (+)
AND a.vendor_id = c.vendor_id (+)
AND a.prim_addr_seq_num=c.address_seq_num (+)
This query does an index full scan on an index for ps_vendor_addr. It takes 2+ minutes to run. Now, if I remove the word distinct, it uses an index range scan and "view pushed predicate". It runs in 2 seconds.
I've tried with and without the first_rows hint in the view. If I leave off the INDEX_ASC hint then it does a full table scan of table ps_vendor_addr. It refuses to do a range scan with the hint. Can anybody tell me how I can get the 'distinct' query tuned?
2 minutes may not seem like a lot but when online users run the query many times a day, it is very frustrating.
Thanks! I hope I provided enough info.

Answer 1

Thomas:
The different behaviours you are seeing are a result of the DISTINCT in the query. This causes a sort to be performed, and will influence the way that the CBO will execute the query. (You do know that you are using the Cost Based Optimizer because of the hints, and that you should analyze the tables and indexes?) You need to be able to re-write the view to avoid the need for the DISTINCT in the query.
Without knowing the meaning of the fields, it is really hard to say anything meaningful, but my guess is that it is the correlated sub-query that is ultimately causing the need for the DISTINCT. Is the combination of set_id,vendor_id and address_seq_num truly unique, or is the address_seq_num just a sequence.
For example in one of my databases, I have a table with INDV_ID, EFF_DT, EMPSTAT_SEQ. The empstat_seq field is just there to allow for more than one thing happening on the same day. The way we query this table is:
SELECT *
FROM empstat_t a
WHERE indv_id = :emp_id and
      TO_CHAR(eff_dt,'yyyymmdd')||TO_CHAR(empstat_seq,'009') =
              (SELECT MAX(TO_CHAR(eff_dt,'yyyymmdd')||TO_CHAR(empstat_seq,'009')
               FROM empstat_t
               WHERE a.indv_id = indv_id);Could something similar work in your case?
If not, assuming your statistics are up to date, I would also look at creating the view without hints to see what the optimizer comes up with on its own. It may be better than you think.
TTFN
John