使用USE_HASH Hint调优一个SQL语句
开发的同事写了个很长的SQL， 执行效率低下，于是找到我：
SQL如下：
auto trace结果如下：
| 0 | SELECT STATEMENT | | 397
| 172K| 3934 (1)| 00:00:48 |
|* 1 | HASH JOIN | | 397
| 172K| 3934 (1)| 00:00:48 |
|* 2 | HASH JOIN | | 332
| 140K| 3780 (1)| 00:00:46 |
| 3 | NESTED LOOPS | | 298
| 122K| 3626 (1)| 00:00:44 |
| 4 | NESTED LOOPS | | 298
| 122K| 3626 (1)| 00:00:44 |
| 5 | NESTED LOOPS OUTER | | 255
| 101K| 3116 (1)| 00:00:38 |
|* 6 | HASH JOIN | | 255
| 99195 | 3116 (1)| 00:00:38 |
|* 7 | HASH JOIN | | 255
| 93585 | 3064 (1)| 00:00:37 |
|* 8 | HASH JOIN RIGHT OUTER | | 255
| 89760 | 2994 (1)| 00:00:36 |
|* 9 | TABLE ACCESS FULL | T_VENCUS | 558
| 6138 | 18 (0)| 00:00:01 |
| 10 | NESTED LOOPS OUTER | | 255
| 86955 | 2976 (1)| 00:00:36 |
|* 11 | HASH JOIN OUTER | | 255
| 83130 | 2540 (1)| 00:00:31 |
| 12 | NESTED LOOPS OUTER | | 255
| 53295 | 1018 (1)| 00:00:13 |
|* 13 | HASH JOIN | | 255
| 45645 | 253 (1)| 00:00:04 |
| 14 | TABLE ACCESS FULL | D_ABCFL | 5225
| 36575 | 8 (0)| 00:00:01 |
|* 15 | HASH JOIN RIGHT OUTER | | 1445
| 242K| 245 (1)| 00:00:03 |
| 16 | VIEW | | 1
| 52 | 0 (0)| 00:00:01 |
| 17 | HASH GROUP BY | | 1
| 61 | 0 (0)| 00:00:01 |
|* 18 | TABLE ACCESS BY INDEX ROWID| T_RPT_SALE | 1
| 61 | 0 (0)| 00:00:01 |
| 19 | INDEX FULL SCAN | I_U_SALE_TWO | 1
| | 0 (0)| 00:00:01 |
|* 20 | HASH JOIN | | 1445
| 169K| 244 (1)| 00:00:03 |
|* 21 | HASH JOIN | | 1445
| 33235 | 158 (1)| 00:00:02 |
| 22 | TABLE ACCESS FULL | D_BHXS | 30
| 330 | 4 (0)| 00:00:01 |
|* 23 | TABLE ACCESS FULL | T_WARE_CLASS_BASE | 5631
| 67572 | 154 (1)| 00:00:02 |
|* 24 | TABLE ACCESS FULL | T_WARE | 9860
| 934K| 86 (0)| 00:00:02 |
| 25 | VIEW PUSHED PREDICATE | | 1
| 30 | 3 (0)| 00:00:01 |
| 26 | SORT GROUP BY | | 1
| 19 | 3 (0)| 00:00:01 |
|* 27 | FILTER | |
| | | |
| 28 | TABLE ACCESS BY INDEX ROWID | T_STORE_H | 1
| 19 | 3 (0)| 00:00:01 |
|* 29 | INDEX UNIQUE SCAN | PK_T_STORE_H | 1
| | 2 (0)| 00:00:01 |
| 30 | VIEW | | 4942
| 564K| 1523 (2)| 00:00:19 |
| 31 | HASH GROUP BY | | 4942
| 115K| 1523 (2)| 00:00:19 |
|* 32 | TABLE ACCESS FULL | T_STORE_H | 487
K| 11M| 1510 (1)| 00:00:19 |
| 33 | TABLE ACCESS BY INDEX ROWID | T_STORE_H | 1
| 15 | 2 (0)| 00:00:01 |
|* 34 | INDEX UNIQUE SCAN | PK_T_STORE_H | 1
| | 1 (0)| 00:00:01 |
|* 35 | TABLE ACCESS FULL | T_WARE_EXT | 9823
| 143K| 69 (0)| 00:00:01 |
| 36 | TABLE ACCESS FULL | T_WARE_BASE | 9860
| 211K| 52 (0)| 00:00:01 |
| 37 | VIEW PUSHED PREDICATE | | 1
| 17 | 0 (0)| 00:00:01 |
| 38 | SORT GROUP BY | | 1
| 48 | 0 (0)| 00:00:01 |
|* 39 | FILTER | |
| | | |
|* 40 | TABLE ACCESS BY INDEX ROWID | T_RPT_SALE | 1
| 48 | 0 (0)| 00:00:01 |
|* 41 | INDEX RANGE SCAN | I_U_SALE_TWO | 1
| | 0 (0)| 00:00:01 |
|* 42 | INDEX UNIQUE SCAN | PK_T_WARE_SALEPRICE | 1
| | 1 (0)| 00:00:01 |
| 43 | TABLE ACCESS BY INDEX ROWID | T_WARE_SALEPRICE | 1
| 15 | 2 (0)| 00:00:01 |
|* 44 | TABLE ACCESS FULL | T_WARE_CLASS_BASE | 9447
| 110K| 154 (1)| 00:00:02 |
|* 45 | TABLE ACCESS FULL | T_WARE_CLASS_BASE | 10191
| 119K| 154 (1)| 00:00:02 |
--------------------------------------------------------------------------------
----------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - access("W"."WAREID"="WC2"."WAREID")
2 - access("W"."WAREID"="WC3"."WAREID")
6 - access("W"."WAREID"="WB"."WAREID")
7 - access("W"."WAREID"="WE"."WAREID" AND "W"."COMPID"="WE"."COMPID")
8 - access("W"."COMPID"="V"."COMPID"(+) AND "W"."LASTVENCUSNO"="V"."VENCUSNO"
(+))
9 - filter("V"."COMPID"(+)=2)
11 - access("W"."COMPID"="ZKC"."COMPID"(+) AND "W"."WAREID"="ZKC"."WAREID"(+))
13 - access("XS"."ABC"="FL"."FL" AND "W"."WAREID"="FL"."WAREID")
15 - access("W"."WAREID"="S6"."WAREID"(+) AND "W"."COMPID"="S6"."COMPID"(+))
18 - filter("RS"."COMPID"=2 AND "RS"."ACCDATE">=TO_DATE(' 2016-08-08 00:00:00'
, 'syyyy-mm-dd
hh24:mi:ss') AND "RS"."ACCDATE"<=TO_DATE(' 2016-09-07 00:00:00', '
syyyy-mm-dd hh24:mi:ss'))
20 - access("W"."WAREID"="WC"."WAREID")
21 - access("WC"."CLASSCODE"="XS"."CLASSCODE")
23 - filter("WC"."CLASSCODE" IS NOT NULL AND "WC"."CLASSGROUPNO"='87')
24 - filter("W"."COMPID"=2)
27 - filter("W"."COMPID"=2)
29 - access("H"."COMPID"=2 AND "H"."BUSNO"=1000 AND "H"."WAREID"="W"."WAREID")
32 - filter("H"."COMPID"=2)
34 - access("H"."COMPID"(+)=2 AND "H"."BUSNO"(+)=1000 AND "W"."WAREID"="H"."WA
REID"(+))
35 - filter("WE"."COMPID"=2)
39 - filter(TRUNC(ADD_MONTHS(SYSDATE@!,-12)+60)>=TRUNC(ADD_MONTHS(SYSDATE@!,-1
2)+31) AND "W"."COMPID"=2)
40 - filter("RS"."COMPID"=2 AND "RS"."ACCDATE">=TRUNC(ADD_MONTHS(SYSDATE@!,-12
)+31) AND
"RS"."ACCDATE"<=TRUNC(ADD_MONTHS(SYSDATE@!,-12)+60))
41 - access("RS"."WAREID"="W"."WAREID")
42 - access("WS"."COMPID"=2 AND "WS"."SALEGROUPID"=1 AND "W"."WAREID"="WS"."WA
REID")
44 - filter("WC3"."CLASSGROUPNO"='01')
45 - filter("WC2"."CLASSGROUPNO"='33')
统计信息
----------------------------------------------------------
1 recursive calls
0 db block gets
91720 consistent gets
0 physical reads
0 redo size
784821 bytes sent via SQL*Net to client
4348 bytes received via SQL*Net from client
350 SQL*Net roundtrips to/from client
10450 sorts (memory)
0 sorts (disk)
5224 rows processed
这个SQL几个表都是全表扫描，由于是查询的所有商品的信息（零售行业），所以调整表扫描方式等途径想提升效率提升会很有限。
从这个执行计划来看，子视图zkc实际有几万条记录，T_WARE数据也有几千条。CBO选择了NESTED LOOP进行表连接。由于数据量较大，这里采用HASH JOIN应当效率较高
强制CBO选择HASH JOIN
SELECT /*+ USE_HASH(zkc) */ ****
hint加入之后，效果很显著，原来需要6分多才能执行完成，采用HASH JOIN之后，15秒不到就执行完成
前后数据记录有差别是因为写这个博客的时候，环境已经变化。
但是可以看到，逻辑读的数量已经不是一个数量级的。
SQL如下：
点击(此处)折叠或打开
- SELECT w.wareid,
- w.warename,
- w.warespec,
- w.wareunit,
- w.factoryid,
- w.bandcode,
- w.lastpurprice,
- ws.saleprice,
- CASE
- WHEN ws.saleprice = 0 THEN
- 0
- ELSE
- round((ws.saleprice - w.lastpurprice) / ws.saleprice, 4)
- END mll,
- fl.fl,
- zkc.xl1,
- s2.xl2,
- zkc.xl5,
- zkc.xl3,
- zkc.xl4,
- zbkc.zbkc,
- CASE
- WHEN nvl(xl4, 0) = 0 THEN
- 0
- ELSE
- round(zbkc.zbkc / xl4, 4)
- END cxb,
- zkc.zkc,
- CASE
- WHEN xl4 = 0 THEN
- 0
- ELSE
- round(zkc.zkc / xl4, 4)
- END zcxb,
- round(CASE
- WHEN zkc.zkc = 0 THEN
- 0
- ELSE
- (zkc.zkc - zbkc.zbkc) / zkc.zkc
- END,
- 4) fdkcbfb,
- nvl(xl4, 0) * xs.xs - zbkc.zbkc bhl,
- w.midqty,
- v.cashtype,
- nvl(h.sumpendingqty, 0) sumpendingqty,
- s6.zdxl,
- s6.zdxsje,
- w.purstatus,
- w.salestatus,
- zkc.xsje,
- zkc.zkcje,
- zbkc.zbkcje,
- w.purtax,
- wc.classcode jx,
- wb.fileno,
- w.inittime,
- we.we_num03,
- we.we_lchar03,
- w.store_req,
- we.we_lchar02,
- w.warningdays,
- CASE
- WHEN wc2.classcode = '3302' THEN
- '是'
- ELSE
- '否'
- END otc,
- wc3.classcode,
- w.waremark,
- fl.vendorno1 gysbh,
- fl.vendorno1,
- w.lastvencusno,
- fl.vendorno2,
- 1000 fdbh
- FROM t_ware w
- JOIN d_abcfl fl
- ON w.wareid = fl.wareid
- JOIN t_ware_class_base wc
- ON w.wareid = wc.wareid
- AND wc.classgroupno = '87'
- JOIN d_bhxs xs
- ON wc.classcode = xs.classcode
- AND xs.abc = fl.fl
- LEFT JOIN (SELECT h.compid,
- h.wareid,
- SUM(h.sumqty) zkc,
- SUM(h.sumqty * h.storepurprice) zkcje,
- SUM(h.lastyqty) xl1,
- SUM(h.lastm3qty) xl5,
- SUM(h.lastm2qty) xl3,
- SUM(h.lastmqty) xl4,
- SUM(h.lastmqty * h.storepurprice) xsje
- FROM t_store_h h
- GROUP BY h.compid, h.wareid) zkc
- ON w.wareid = zkc.wareid
- AND w.compid = zkc.compid
- LEFT JOIN (SELECT h.compid,
- h.wareid,
- SUM(h.sumqty) zbkc,
- SUM(h.sumqty * h.storepurprice) zbkcje
- FROM t_store_h h
- WHERE h.busno = 1000
- GROUP BY h.compid, h.wareid) zbkc
- ON w.compid = zbkc.compid
- AND w.wareid = zbkc.wareid
- LEFT JOIN (SELECT rs.compid, rs.wareid, SUM(rs.wareqty) xl2
- FROM t_rpt_sale rs
- WHERE rs.accdate >= trunc(add_months(SYSDATE, -12) + 31)
- AND rs.accdate <= trunc(add_months(SYSDATE, -12) + 60)
- GROUP BY rs.compid, rs.wareid) s2
- ON w.compid = s2.compid
- AND w.wareid = s2.wareid
- LEFT JOIN (SELECT rs.compid,
- rs.wareid,
- SUM(rs.wareqty) zdxl,
- SUM(rs.netsum) zdxsje
- FROM t_rpt_sale rs
- WHERE rs.accdate >= to_date('2016/8/8', 'yyyy-mm-dd')
- AND rs.accdate <= to_date('2016/9/7', 'yyyy-mm-dd')
- GROUP BY rs.compid, rs.wareid) s6
- ON w.compid = s6.compid
- AND w.wareid = s6.wareid
- LEFT JOIN t_vencus v
- ON w.lastvencusno = v.vencusno
- AND w.compid = v.compid
- LEFT JOIN t_store_h h
- ON w.compid = h.compid
- AND w.wareid = h.wareid
- AND h.busno = 1000, t_ware_saleprice ws, t_ware_base wb,
- t_ware_ext we, t_ware_class_base wc2, t_ware_class_base wc3
- WHERE ws.salegroupid = 1
- AND w.compid = ws.compid
- AND w.wareid = ws.wareid
- AND w.wareid = wb.wareid
- AND w.wareid = we.wareid
- AND w.compid = we.compid
- AND w.wareid = wc2.wareid
- AND wc2.classgroupno = '33'
- AND w.wareid = wc3.wareid
- AND wc3.classgroupno = '01'
- AND w.compid = 2
| 0 | SELECT STATEMENT | | 397
| 172K| 3934 (1)| 00:00:48 |
|* 1 | HASH JOIN | | 397
| 172K| 3934 (1)| 00:00:48 |
|* 2 | HASH JOIN | | 332
| 140K| 3780 (1)| 00:00:46 |
| 3 | NESTED LOOPS | | 298
| 122K| 3626 (1)| 00:00:44 |
| 4 | NESTED LOOPS | | 298
| 122K| 3626 (1)| 00:00:44 |
| 5 | NESTED LOOPS OUTER | | 255
| 101K| 3116 (1)| 00:00:38 |
|* 6 | HASH JOIN | | 255
| 99195 | 3116 (1)| 00:00:38 |
|* 7 | HASH JOIN | | 255
| 93585 | 3064 (1)| 00:00:37 |
|* 8 | HASH JOIN RIGHT OUTER | | 255
| 89760 | 2994 (1)| 00:00:36 |
|* 9 | TABLE ACCESS FULL | T_VENCUS | 558
| 6138 | 18 (0)| 00:00:01 |
| 10 | NESTED LOOPS OUTER | | 255
| 86955 | 2976 (1)| 00:00:36 |
|* 11 | HASH JOIN OUTER | | 255
| 83130 | 2540 (1)| 00:00:31 |
| 12 | NESTED LOOPS OUTER | | 255
| 53295 | 1018 (1)| 00:00:13 |
|* 13 | HASH JOIN | | 255
| 45645 | 253 (1)| 00:00:04 |
| 14 | TABLE ACCESS FULL | D_ABCFL | 5225
| 36575 | 8 (0)| 00:00:01 |
|* 15 | HASH JOIN RIGHT OUTER | | 1445
| 242K| 245 (1)| 00:00:03 |
| 16 | VIEW | | 1
| 52 | 0 (0)| 00:00:01 |
| 17 | HASH GROUP BY | | 1
| 61 | 0 (0)| 00:00:01 |
|* 18 | TABLE ACCESS BY INDEX ROWID| T_RPT_SALE | 1
| 61 | 0 (0)| 00:00:01 |
| 19 | INDEX FULL SCAN | I_U_SALE_TWO | 1
| | 0 (0)| 00:00:01 |
|* 20 | HASH JOIN | | 1445
| 169K| 244 (1)| 00:00:03 |
|* 21 | HASH JOIN | | 1445
| 33235 | 158 (1)| 00:00:02 |
| 22 | TABLE ACCESS FULL | D_BHXS | 30
| 330 | 4 (0)| 00:00:01 |
|* 23 | TABLE ACCESS FULL | T_WARE_CLASS_BASE | 5631
| 67572 | 154 (1)| 00:00:02 |
|* 24 | TABLE ACCESS FULL | T_WARE | 9860
| 934K| 86 (0)| 00:00:02 |
| 25 | VIEW PUSHED PREDICATE | | 1
| 30 | 3 (0)| 00:00:01 |
| 26 | SORT GROUP BY | | 1
| 19 | 3 (0)| 00:00:01 |
|* 27 | FILTER | |
| | | |
| 28 | TABLE ACCESS BY INDEX ROWID | T_STORE_H | 1
| 19 | 3 (0)| 00:00:01 |
|* 29 | INDEX UNIQUE SCAN | PK_T_STORE_H | 1
| | 2 (0)| 00:00:01 |
| 30 | VIEW | | 4942
| 564K| 1523 (2)| 00:00:19 |
| 31 | HASH GROUP BY | | 4942
| 115K| 1523 (2)| 00:00:19 |
|* 32 | TABLE ACCESS FULL | T_STORE_H | 487
K| 11M| 1510 (1)| 00:00:19 |
| 33 | TABLE ACCESS BY INDEX ROWID | T_STORE_H | 1
| 15 | 2 (0)| 00:00:01 |
|* 34 | INDEX UNIQUE SCAN | PK_T_STORE_H | 1
| | 1 (0)| 00:00:01 |
|* 35 | TABLE ACCESS FULL | T_WARE_EXT | 9823
| 143K| 69 (0)| 00:00:01 |
| 36 | TABLE ACCESS FULL | T_WARE_BASE | 9860
| 211K| 52 (0)| 00:00:01 |
| 37 | VIEW PUSHED PREDICATE | | 1
| 17 | 0 (0)| 00:00:01 |
| 38 | SORT GROUP BY | | 1
| 48 | 0 (0)| 00:00:01 |
|* 39 | FILTER | |
| | | |
|* 40 | TABLE ACCESS BY INDEX ROWID | T_RPT_SALE | 1
| 48 | 0 (0)| 00:00:01 |
|* 41 | INDEX RANGE SCAN | I_U_SALE_TWO | 1
| | 0 (0)| 00:00:01 |
|* 42 | INDEX UNIQUE SCAN | PK_T_WARE_SALEPRICE | 1
| | 1 (0)| 00:00:01 |
| 43 | TABLE ACCESS BY INDEX ROWID | T_WARE_SALEPRICE | 1
| 15 | 2 (0)| 00:00:01 |
|* 44 | TABLE ACCESS FULL | T_WARE_CLASS_BASE | 9447
| 110K| 154 (1)| 00:00:02 |
|* 45 | TABLE ACCESS FULL | T_WARE_CLASS_BASE | 10191
| 119K| 154 (1)| 00:00:02 |
--------------------------------------------------------------------------------
----------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - access("W"."WAREID"="WC2"."WAREID")
2 - access("W"."WAREID"="WC3"."WAREID")
6 - access("W"."WAREID"="WB"."WAREID")
7 - access("W"."WAREID"="WE"."WAREID" AND "W"."COMPID"="WE"."COMPID")
8 - access("W"."COMPID"="V"."COMPID"(+) AND "W"."LASTVENCUSNO"="V"."VENCUSNO"
(+))
9 - filter("V"."COMPID"(+)=2)
11 - access("W"."COMPID"="ZKC"."COMPID"(+) AND "W"."WAREID"="ZKC"."WAREID"(+))
13 - access("XS"."ABC"="FL"."FL" AND "W"."WAREID"="FL"."WAREID")
15 - access("W"."WAREID"="S6"."WAREID"(+) AND "W"."COMPID"="S6"."COMPID"(+))
18 - filter("RS"."COMPID"=2 AND "RS"."ACCDATE">=TO_DATE(' 2016-08-08 00:00:00'
, 'syyyy-mm-dd
hh24:mi:ss') AND "RS"."ACCDATE"<=TO_DATE(' 2016-09-07 00:00:00', '
syyyy-mm-dd hh24:mi:ss'))
20 - access("W"."WAREID"="WC"."WAREID")
21 - access("WC"."CLASSCODE"="XS"."CLASSCODE")
23 - filter("WC"."CLASSCODE" IS NOT NULL AND "WC"."CLASSGROUPNO"='87')
24 - filter("W"."COMPID"=2)
27 - filter("W"."COMPID"=2)
29 - access("H"."COMPID"=2 AND "H"."BUSNO"=1000 AND "H"."WAREID"="W"."WAREID")
32 - filter("H"."COMPID"=2)
34 - access("H"."COMPID"(+)=2 AND "H"."BUSNO"(+)=1000 AND "W"."WAREID"="H"."WA
REID"(+))
35 - filter("WE"."COMPID"=2)
39 - filter(TRUNC(ADD_MONTHS(SYSDATE@!,-12)+60)>=TRUNC(ADD_MONTHS(SYSDATE@!,-1
2)+31) AND "W"."COMPID"=2)
40 - filter("RS"."COMPID"=2 AND "RS"."ACCDATE">=TRUNC(ADD_MONTHS(SYSDATE@!,-12
)+31) AND
"RS"."ACCDATE"<=TRUNC(ADD_MONTHS(SYSDATE@!,-12)+60))
41 - access("RS"."WAREID"="W"."WAREID")
42 - access("WS"."COMPID"=2 AND "WS"."SALEGROUPID"=1 AND "W"."WAREID"="WS"."WA
REID")
44 - filter("WC3"."CLASSGROUPNO"='01')
45 - filter("WC2"."CLASSGROUPNO"='33')
统计信息
----------------------------------------------------------
1 recursive calls
0 db block gets
91720 consistent gets
0 physical reads
0 redo size
784821 bytes sent via SQL*Net to client
4348 bytes received via SQL*Net from client
350 SQL*Net roundtrips to/from client
10450 sorts (memory)
0 sorts (disk)
5224 rows processed
这个SQL几个表都是全表扫描，由于是查询的所有商品的信息（零售行业），所以调整表扫描方式等途径想提升效率提升会很有限。
从这个执行计划来看，子视图zkc实际有几万条记录，T_WARE数据也有几千条。CBO选择了NESTED LOOP进行表连接。由于数据量较大，这里采用HASH JOIN应当效率较高
强制CBO选择HASH JOIN
SELECT /*+ USE_HASH(zkc) */ ****
hint加入之后，效果很显著，原来需要6分多才能执行完成，采用HASH JOIN之后，15秒不到就执行完成
前后数据记录有差别是因为写这个博客的时候，环境已经变化。
但是可以看到，逻辑读的数量已经不是一个数量级的。
来自 “ ITPUB博客 ” ，链接：http://blog.itpub.net/8520577/viewspace-2124846/，如需转载，请注明出处，否则将追究法律责任。
转载于:http://blog.itpub.net/8520577/viewspace-2124846/