Too many subselects in Oracle Query? - sql

Question

I'd like to know how I could improve the performance of the below query, since it is taking way too long to run, after all, it returns millions of rows... I'm a dummy when it comes to SQL...
SELECT CIAM.EXTERNAL_ID,
(SELECT NEW_CHARGES / 100
FROM BI_OWNER.CMF_BALANCE
WHERE ( ACCOUNT_NO, BILL_REF_NO ) = (SELECT ACCOUNT_NO,
MAX(BILL_REF_NO)
FROM BI_OWNER.CMF_BALANCE
WHERE
ACCOUNT_NO = CIAM.ACCOUNT_NO
GROUP BY ACCOUNT_NO))
"AMOUNT LAST BILL",
(SELECT 'ACTIVE DISCOUNT'
|| ' '
|| CCK.AVAIL_PERIODS
|| '/'
|| CC.TOTAL_PERIODS
FROM BI_OWNER.CUSTOMER_CONTRACT_KEY CCK,
BI_OWNER.CUSTOMER_CONTRACT CC
WHERE CC.PARENT_ACCOUNT_NO = CIAM.ACCOUNT_NO
AND CC.END_DT IS NULL
AND EXISTS (SELECT 1
FROM CONTRACT_TYPES
WHERE CONTRACT_TYPE = CC.CONTRACT_TYPE
AND PLAN_ID_DISCOUNT IS NOT NULL
AND DURATION_UNITS = -3)
AND ROWNUM = 1
AND CCK.TRACKING_ID = CC.TRACKING_ID
AND CCK.TRACKING_ID_SERV = CC.TRACKING_ID_SERV) "DISCOUNT",
(SELECT CC.TOTAL_PERIODS
FROM BI_OWNER.CUSTOMER_CONTRACT_KEY CCK,
BI_OWNER.CUSTOMER_CONTRACT CC
WHERE CC.PARENT_ACCOUNT_NO = CIAM.ACCOUNT_NO
AND CC.END_DT IS NULL
AND EXISTS (SELECT 1
FROM CONTRACT_TYPES
WHERE CONTRACT_TYPE = CC.CONTRACT_TYPE
AND PLAN_ID_DISCOUNT IS NOT NULL
AND DURATION_UNITS = -3)
AND ROWNUM = 1
AND CCK.TRACKING_ID = CC.TRACKING_ID
AND CCK.TRACKING_ID_SERV = CC.TRACKING_ID_SERV) "CYCLE"
,
(SELECT SUM(BALANCE_DUE)
FROM BI_OWNER.CMF_BALANCE
WHERE ACCOUNT_NO = CIAM.ACCOUNT_NO
AND PPDD_DATE < TRUNC(SYSDATE))
"DEBT"
FROM BI_OWNER.CUSTOMER_ID_ACCT_MAP CIAM
WHERE EXTERNAL_ID_TYPE = 1
AND EXISTS (SELECT 1
FROM BI_OWNER.CMF
WHERE ACCOUNT_NO = CIAM.ACCOUNT_NO
AND PREV_CUTOFF_DATE > SYSDATE - 30)

Answer 1

I would recommend identifying the SQL id for the query then using the SQL Monitor Report as it will tell you exactly what the execution plan is and where the SQL is spending most of it's time.
A simple way to get the SQL Monitor Report from SQL*Plus follows:
spool c:\temp\SQL_Monitor_rpt.html
SET LONG 1000000
SET LONGCHUNKSIZE 1000000
SET LINESIZE 1000
SET PAGESIZE 0
SET TRIM ON
SET TRIMSPOOL ON
SET ECHO OFF
SET FEEDBACK OFF
alter session set "_with_subquery" = optimizer;
SELECT DBMS_SQLTUNE.report_sql_monitor(
sql_id => '&SQLID' ,
type => 'HTML',
report_level => 'ALL') AS report
FROM dual;
spool off
Basically, you need to know your table sizes and how to get the large tables to have data access via an index (e.g. index on columns found in the where clause).

Answer 2

Here is an initial stab and may provide significant improvement. Many of your queries were correlated subqueries being executed for every record. Instead, I tried to build pre-query aggregates per account number in the select from/join section. Query first, then I'll explain logic after.
SELECT
CIAM.EXTERNAL_ID,
CMF_BALANCE.New_Charges / 100.0 "AMOUNT LAST BILL",
CCKs.Discount,
CCKs.Cycle,
AcntLast30.SumBalance "DEBT"
FROM
(SELECT
CMF.Account_No,
max( Bal.Bill_Ref_No ) MaxBillRef,
sum( case when Bal.PPDD_Date < TRUNC(SYSDATE )
then Bal.Balance_Due else 0 end ) SumBalance
from
BI_OWNER.CMF
JOIN BI_OWNER.CMF_BALANCE BAL
on CMF.Account_No = Bal.Account_No
where
CMF.PREV_CUTOFF_DATE > SYSDATE - 30
group by
CMF.Account_No ) AcntLast30
JOIN BI_OWNER.CUSTOMER_ID_ACCT_MAP CIAM
on AcntLast30.Account_No = CIAM.Account_No
AND CIAM.EXTERNAL_ID_TYPE = 1
JOIN BI_OWNER.CMF_BALANCE
on AcntLast30.Account_No = CMFBalance.Account_No
AND AcntLast30.MaxBillRef = CMFBalance.Bill_Ref_No
JOIN
(select
CC.Parent_Account_No,
CC.TOTAL_PERIODS "CYCLE",
'ACTIVE DISCOUNT' || ' ' || CCK.AVAIL_PERIODS || '/' || CC.TOTAL_PERIODS "DISCOUNT"
FROM
BI_OWNER.CUSTOMER_CONTRACT CC
JOIN BI_OWNER.CUSTOMER_CONTRACT_KEY CCK
ON CC.TRACKING_ID = CCK.TRACKING_ID
AND CC.TRACKING_ID_SERV = CCK.TRACKING_ID_SERV
AND ROWNUM = 1
JOIN ( select distinct Contract_Type
FROM CONTRACT_TYPES
WHERE PLAN_ID_DISCOUNT IS NOT NULL
AND DURATION_UNITS = -3) CT
on CC.Contract_Type = CT.Contract_Type
WHERE
CC.END_DT IS NULL ) CCKs
on AcntLast30.Account_No = CCKs.Parent_Account_No
The initial "FROM" clause, I have a subquery because you appear to be only interested in accounts within the last 30 days. So, while I'm there, I am joining to your CMF_Balance table and getting the maximum Bill_Ref_No per account AND the sum of the balance when the PPDD_Date is less than the TRUNC(sysdate) which is your "DEBT" result column. So now we have the finite list of accounts you are interested in with the account, max bill on file and the balance due summed up.
(SELECT
CMF.Account_No,
max( Bal.Bill_Ref_No ) MaxBillRef,
sum( case when Bal.PPDD_Date < TRUNC(SYSDATE )
then Bal.Balance_Due else 0 end ) SumBalance
from
BI_OWNER.CMF
JOIN BI_OWNER.CMF_BALANCE BAL
on CMF.Account_No = Bal.Account_No
where
CMF.PREV_CUTOFF_DATE > SYSDATE - 30
group by
CMF.Account_No ) AcntLast30
Next, simple join to the CIAM table to only get accounts for External_ID_Type = 1. This too could have been merged into the query above for "AcntLast30" alias result.
JOIN BI_OWNER.CUSTOMER_ID_ACCT_MAP CIAM
on AcntLast30.Account_No = CIAM.Account_No
AND CIAM.EXTERNAL_ID_TYPE = 1
Now, since the "AcntLast30" query has the account and max bill reference we then join back to the CMF_Balance on the account and bill reference # once thus giving us the CMF_BALANCE.New_Charges / 100.0 "AMOUNT LAST BILL"
JOIN BI_OWNER.CMF_BALANCE
on AcntLast30.Account_No = CMFBalance.Account_No
AND AcntLast30.MaxBillRef = CMFBalance.Bill_Ref_No
Finally the subquery alias result "CCKs". Since the Discount and Cycle use the same query/subquery/exists, I just ran it once that qualified on the discounts types and pulled the Account_No for the JOIN condition. Now we have the Discount and Cycle values per account.
If you are returning so many rows, I believe the performance gained by grabbing these pre-query aggregates once up-front and joining to by the account will be much faster than that of each time individually subquerying at every row.
There was a reference to ROWNUM without any table/alias reference so I am not sure the impact of that one within the query.
Final note. For things like the discount that may not be applicable, you may need to change it to a LEFT JOIN, of which those values would show as NULL. But without knowing the extent of data, Cartesian products of 1:many entries in given tables, I think this will work well for you. For the most part it looked like everything was resulting in only one record qualified per account where higher importance on join (such as the max bill reference).