巧用rowid简化sql查询(r2笔记47天)
巧用rowid简化sql查询(r2笔记47天)
生产系统中有一条sql语句，目前执行的时间有点长了,而且看起来有些臃肿，客户问能不能改进一下。得到的sql语句如下：
SELECT COUNT(1)
FROM (
SELECT /*+ leading (payment_temp_table payment PAYMENT_DETAILS account memo) use_nl (payment_temp_table ar1_payment PAYMENT_DETAILS account memo) index (payment payment_pk) index (PAYMENT_DETAILS PAYMENT_DETAILS_pk) */
PAYMENT_DETAILS.PAYMENT_ID,
PAYMENT.CREDIT_ID,
PAYMENT.ACCOUNT_ID,
PAYMENT.AMOUNT,
PAYMENT.ORIGINAL_AMOUNT,
PAYMENT.ORIGINAL_CONVERTED_AMOUNT,
PAYMENT_DETAILS.PAYMENT_METHOD,
PAYMENT_DETAILS.DEPOSIT_DATE,
PAYMENT_DETAILS.CURRENCY,
PAYMENT_DETAILS.CHECK_NO,
PAYMENT_DETAILS.CREDIT_CARD_NUMBER,
PAYMENT_DETAILS.BANK_CODE,
PAYMENT_DETAILS.BANK_ACCOUNT_NUMBER,
ACCOUNT.ACCOUNT_TIMESTAMP,
ACCOUNT.CURRENCY,
PAYMENT_DETAILS.PAYMENT_SUB_METHOD,
ACCOUNT.BE,
MEMO.MEMO_TEXT,
PAYMENT_DETAILS.PERIOD_KEY,
PAYMENT_DETAILS.PARTITION_ID,
PAYMENT_DETAILS.AMOUNT,
PAYMENT.ACTIVITY_DATE,
PAYMENT_DETAILS.ACCOUNT_ID,
PAYMENT.L9_RT_ID,
PAYMENT_DETAILS.L9_CONV_INV_NUMBER,
PAYMENT_DETAILS.CC_EXPIRY_DATE,
PAYMENT_DETAILS.DIRECT_DEBIT_VOUCHER,
PAYMENT_DETAILS.PAYMENT_SOURCE_TYPE,
PAYMENT.L9_WHT_AMT,
PAYMENT.L9_WHT_CERT_NO,
PAYMENT.L9_WHT_RATE,
PAYMENT.L9_VAT_AMOUNT,
PAYMENT.L9_PRINT_RT_IND,
PAYMENT.L9_USER_ID,
PAYMENT.L9_RT_EXTRACT_IND,
PAYMENT.L9_ECA_REASON_CODE,
PAYMENT.L9_RT_GENERATE_MODE,
PAYMENT_DETAILS.L9_VAT_AMOUNT,
PAYMENT.BILL_SEQ_NO,
PAYMENT.ACTIVITY_INDICATOR,
ACCOUNT.L9_COMPANY_CODE,
PAYMENT_DETAILS.PAYMENT_SOURCE_ID,
PAYMENT.OPERATOR_ID
FROM PAYMENT_DETAILS,
PAYMENT,
ACCOUNT,
MEMO,
(SELECT row_number,
PAYMENT_ID,
CREDIT_ID,
ACCOUNT_ID,
PD_PARTITION_ID,
PD_PERIOD_KEY,
PT_PARTITION_ID,
PT_PERIOD_KEY
FROM (SELECT rownum as row_number,
PAYMENT_ID,
CREDIT_ID,
ACCOUNT_ID,
PD_PARTITION_ID,
PD_PERIOD_KEY,
PT_PARTITION_ID,
PT_PERIOD_KEY
FROM (SELECT /*+ leading (pt a pd) use_nl (pt a pd) index (pd PAYMENT_DETAILS_pk) */
DISTINCT PD.PAYMENT_ID,
PT.CREDIT_ID,
A.ACCOUNT_ID,
pd.partition_id pd_partition_id,
pd.period_key pd_period_key,
pt.partition_id pt_partition_id,
pt.period_key pt_period_key
FROM PAYMENT_DETAILS PD,
PAYMENT PT,
ACCOUNT A
WHERE PT.PARTITION_ID = A.PARTITION_ID
AND PT.ACCOUNT_ID = A.ACCOUNT_ID
AND PD.PARTITION_ID = PT.PYMDT_PARTITION_ID
AND PD.PERIOD_KEY = PT.PYMDT_PERIOD_KEY
AND PD.PAYMENT_ID = PT.PAYMENT_ID
AND (a.be IN
(SELECT child_be_id
FROM gn1_boh_relation
START WITH parent_be_id = NVL(0, 0)
CONNECT BY PRIOR
child_be_id = parent_be_id) OR
a.be = NVL(0, 0))
AND ROWNUM <= 2000
ORDER BY pt.ACTIVITY_DATE DESC,
pt.CREDIT_ID DESC))
WHERE row_number BETWEEN 1 AND 100) PAYMENT_TEMP_TABLE
WHERE PAYMENT_DETAILS.PAYMENT_ID = PAYMENT_TEMP_TABLE.PAYMENT_ID
AND PAYMENT_DETAILS.partition_id =
payment_temp_table.pd_partition_id
AND PAYMENT_DETAILS.period_key =
payment_temp_table.pd_period_key
AND ACCOUNT.account_id = payment_temp_table.account_id
AND PAYMENT.credit_id = payment_temp_table.credit_id
AND PAYMENT.partition_id = payment_temp_table.pt_partition_id
AND PAYMENT.period_key = payment_temp_table.pt_period_key
AND MEMO.memo_id(+) = PAYMENT_DETAILS.memo_id);
index的信息如下：
INDEX_NAME TABLESPACE INDEX_TYPE UNIQUENES PAR COLUMN_LIST TABLE_TYPE STATUS NUM_ROWS LAST_ANAL G
------------------------------ ---------- ---------- --------- --- ------------------------------ ---------- ------ ---------- --------- -
PAYMENT_DETAILS_PK NORMAL UNIQUE YES PAYMENT_ID,PARTITION_ID,PERIOD_KEY TABLE N/A 6718838 16-JUL-14 N ------------------------------ ---------- ---------- --------- --- ------------------------------ ---------- ------ ---------- --------- -
PAYMENT_PK NORMAL UNIQUE YES CREDIT_ID,PARTITION_ID,PERIOD_KEY TABLE N/A 6914026 16-JUL-14 N可以看到基本索引在查询条件中能用都用到了,而且使用了hint，根据sql内容来看，这个查询只是需要得到对应的记录条数而已，是在原有的查询语句的基础上直接加了select count(1) from xxx改进的来的。 所以对于这类查询，就需要摆脱思想的束缚，可以最大程度上简化sql，达到同样的效果。 改进后的sql语句如下：
SELECT COUNT(1)
FROM (
SELECT
PAYMENT_DETAILS.rowid,
PAYMENT.rowid,
ACCOUNT.rowid ,
--AR1_MEMO.rowid
FROM PAYMENT_DETAILS,
PAYMENT,
ACCOUNT,
-- AR1_MEMO,
(SELECT pdrowid, ptrowid, arowid
FROM (SELECT rownum as row_number, pdrowid, ptrowid, arowid
FROM (SELECT /*+ leading (pt a pd) */
DISTINCT PD.rowid pdrowid,
PT.rowid ptrowid,
A.rowid arowid
FROM PAYMENT_DETAILS PD,
PAYMENT PT,
ACCOUNT A
WHERE PT.PARTITION_ID = A.PARTITION_ID
AND PT.ACCOUNT_ID = A.ACCOUNT_ID
AND PD.PARTITION_ID = PT.PYMDT_PARTITION_ID
AND PD.PERIOD_KEY = PT.PYMDT_PERIOD_KEY
AND PD.PAYMENT_ID = PT.PAYMENT_ID
AND (a.be IN
(SELECT child_be_id
FROM gn1_boh_relation
START WITH parent_be_id = NVL(0, 0)
CONNECT BY PRIOR
child_be_id = parent_be_id) OR
a.be = NVL(0, 0))
AND ROWNUM <= 2147483627
ORDER BY pt.ACTIVITY_DATE DESC,
pt.CREDIT_ID DESC))
WHERE row_number BETWEEN 1 AND 9223372036854775807) PAYMENT_TEMP_TABLE
WHERE PAYMENT_DETAILS.rowid = PAYMENT_TEMP_TABLE.pdrowid
AND ACCOUNT.rowid = payment_temp_table.arowid
AND PAYMENT.rowid = payment_temp_table.ptrowid
AND MEMO.memo_id(+) = PAYMENT_DETAILS.memo_id直接通过rowid来做关联，因为不需要输出所有的数据，只要输出列出含有主键列，就可以考虑使用rowid来代替。 sql语句极大的简化了,不过还没有完，还可以考虑做点什么。因为memo这个表比较大，没有走主键，查询会走全表扫描，耗费不少时间，但是结果集中貌似也不是很需要，因为结果集只考虑最终的返回数据条数，可以考虑是否能够从查询中去掉这个表。 简单验证一下，模拟这种类似的外连接，看看结果集是否会有影响。
SQL>
insert into test_full values(1);
insert into test_full values(2);
insert into test_full values(3);
insert into test_partial values(1,'a');
insert into test_partial values(2,'b');
1 row created.
SQL>
1 row created.
SQL>
1 row created.
SQL>
1 row created.
SQL>
1 row created.
SQL>
SQL> select test_full.id, test_partial.name from test_full,test_partial
2 where test_partial.id(+) = test_full.id
3 /
ID NAME
---------- ------------------------------
1 a
2 b
3通过上面的简单测试，可以说明，如果表test_full中含有的记录，通过和test_partial做外连接，也是全量输出test_full的值。 这样，可以考虑把memo表去除，整理后的sql语句简化为：
SELECT COUNT(1)
FROM (
SELECT
PAYMENT_DETAILS.rowid,
PAYMENT.rowid,
ACCOUNT.rowid
FROM PAYMENT_DETAILS,
PAYMENT,
ACCOUNT,
(SELECT pdrowid, ptrowid, arowid
FROM (SELECT rownum as row_number, pdrowid, ptrowid, arowid
FROM (SELECT /*+ leading (pt a pd) */
DISTINCT PD.rowid pdrowid,
PT.rowid ptrowid,
A.rowid arowid
FROM PAYMENT_DETAILS PD,
PAYMENT PT,
ACCOUNT A
WHERE PT.PARTITION_ID = A.PARTITION_ID
AND PT.ACCOUNT_ID = A.ACCOUNT_ID
AND PD.PARTITION_ID = PT.PYMDT_PARTITION_ID
AND PD.PERIOD_KEY = PT.PYMDT_PERIOD_KEY
AND PD.PAYMENT_ID = PT.PAYMENT_ID
AND (a.be IN
(SELECT child_be_id
FROM gn1_boh_relation
START WITH parent_be_id = NVL(0, 0)
CONNECT BY PRIOR
child_be_id = parent_be_id) OR
a.be = NVL(0, 0))
AND ROWNUM <= 2147483627
ORDER BY pt.ACTIVITY_DATE DESC,
pt.CREDIT_ID DESC))
WHERE row_number BETWEEN 1 AND 9223372036854775807) PAYMENT_TEMP_TABLE
WHERE PAYMENT_DETAILS.rowid = PAYMENT_TEMP_TABLE.pdrowid
AND ACCOUNT.rowid = payment_temp_table.arowid
AND PAYMENT.rowid = payment_temp_table.ptrowid 对于一些Hint也做了删减，保证不必要的资源消耗。 总之，性能sql，对于sql的简化也是一种考验，如果能够最大程度的简化,也是sql调优的进步。