Odd WHERE NOT EXISTS performance on DB2
I am experiencing very odd performance on DB2 version 9.1 when running the query below:
select a.CYCL_NUM
, a.AC_NUM
, a.AUTHS_DTE
, a.PL_ID
, a.APRVD_RSPN_CDE
, a.AUTHS_AMT
, a.AUTHS_STS_CDE
, a.TRAN_CTGR_CDE
, a.MRCHN_CTGR_CDE
, d.out_pu_au_amt
from nwhd12.chldr_auths a, nwhd12.w_chldr_ac d
where cycl_num = 200911
and a.ac_num = d.ac_num
and APRVD_RSPN_CDE = 'APV'
and not exists (
select 1 from auths_rev_hist b
where a.cycl_num = b.cycl_num
and a.auths_dte = b.auths_dte
and a.TRAN_CTGR_CDE = b.TRAN_CTGR_CDE
and a.PL_ID = b.pl_id
and a.APRVD_RSPN_CDE = b.APRVD_RSPN_CDE
and a.AUTHS_AMT = b.auths_amt
and a.TRAN_CTGR_CDE = b.TRAN_CTGR_CDE
and a.MRCHN_CTGR_CDE = MRCHN_CTGR_CDE
)
;
What is supposed to happen is that the query accesses partion 97 of nwhd12.chldr_auths, since that is the partition corresponding to cycle 200911. Instead, after accessing partition 97, it starts accessing every other partition in nwhd12.chldr_auths. Now, I was told that this is because of the "WHERE NOT EXISTS", but there is still the res开发者_运维技巧triction on cycles in this statement (a.cycl_num = b.cycl_num), so why is it scanning all the partitions?
If I hard code the cycle in the where not exists, then the query performs as expected.
Thanks, Dave
if the planner is this easily confused, you need to try a few different formulations. this untested (I don't even have DB2, but CTEs originated there):
WITH hist AS (
cycl_num
, ac_num
, auths_dte
, pl_id
, aprvd_rspn_cde
, auths_amt
, auths_sts_cde
, tran_ctgr_cde
, mrchn_ctgr_cde
FROM auths_rev_hist b
)
, auths AS (
SELECT
cycl_num
, ac_num
, auths_dte
, pl_id
, aprvd_rspn_cde
, auths_amt
, auths_sts_cde
, tran_ctgr_cde
, mrchn_ctgr_cde
FROM nwhd12.chldr_auths
WHERE cycl_num = 200911
AND aprvd_rspn_cde = 'APV'
EXCEPT
SELECT ... FROM hist
)
SELECT a.*, d.out_pu_au_amt
FROM auths a, nwhd12.w_chldr_ac d
WHERE a.ac_num = d.ac_num
精彩评论