在展开下面的original sql 和 execution plan之前，要知道这个SQL的问题就在于占用大量的TEMP space
orignal SQL
SELECT roster.IC_N AS icN, roster.WORK_SHIFT_C AS workShiftC, roster.EXTRA_SHIFT_C AS extraShiftC, roster.GENERATED_SHIFT_C AS generatedShiftC, roster.RESERVE_SHIFT_C AS reserveShiftCode, num.STAFF_N AS staffNo, grp.scheme_n AS schemeN, deploy.deploy_department_c AS dept, code.deployment_c AS deploymentC FROM STAFF_ROSTER roster, STAFF_WORK WORK1, STAFF_CATEGORY category, staff_roster_group grp, STAFF staff, staff_number num, staff_cross_deploy deploy, staff_deployment_code code WHERE WORK1.IC_N = roster.IC_N AND category.IC_N = roster.IC_N AND CATEGORY.CATEGORY_C in (‘SQC‘,‘RQC‘,‘WOS‘,‘CES‘,‘WCES‘,‘NTCES‘,‘HEP‘,‘SOSOS‘,‘SVCOS‘,‘CCOS‘,‘CCSOS‘,‘IGCMO‘,‘SOCMO‘,‘YOCMO‘,‘LS‘,‘LSUP‘,‘LSO‘,‘IGCMO2‘,‘SOCMO2‘,‘YOCMO2‘,‘IGCMO3‘,‘SOCMO3‘,‘YOCMO3‘) AND staff.IC_N = roster.IC_N AND staff.IC_N = num.IC_N AND staff.IC_N = deploy.IC_N AND staff.ic_n = grp.ic_n AND staff.ic_n = code.ic_n(+) AND code.DEPARTMENT_C(+) = ‘C‘ AND TO_CHAR(code.SHIFT_D(+), ‘DD/MM/YYYY‘) = ‘11/03/2014‘ AND deploy.WORK_DEPARTMENT_C = ‘C‘ AND work1.home_department_c = deploy.work_department_c AND work1.home_department_c = roster.department_c AND TO_CHAR(roster.SHIFT_D, ‘DD/MM/YYYY‘) = ‘11/03/2014‘ AND WORK1.START_DT <= roster.shift_d AND ((WORK1.END_DT > roster.shift_d) OR (WORK1.END_DT IS NULL)) AND category.EFFECTIVE_D = (SELECT MAX(SC.EFFECTIVE_D) FROM STAFF_CATEGORY SC WHERE SC.IC_N = roster.IC_N AND SC.EFFECTIVE_D <= roster.SHIFT_D) AND num.EFFECTIVE_D = (SELECT MAX(SN.EFFECTIVE_D) FROM STAFF_NUMBER SN WHERE SN.IC_N = roster.IC_N AND SN.EFFECTIVE_D <= roster.SHIFT_D) AND TO_DATE(‘11/03/2014‘, ‘dd/mm/yyyy‘) BETWEEN deploy.start_dt AND deploy.end_dt AND grp.EFFECTIVE_D = (SELECT MAX(SG.EFFECTIVE_D) FROM STAFF_ROSTER_GROUP SG WHERE SG.IC_N = roster.IC_N AND SG.EFFECTIVE_D <= roster.SHIFT_D)
origanl EXECUTION plan
PLAN_TABLE_OUTPUT -------------------------------------------------------------------------------------------------------------------------------- Plan hash value: 1872515827 --------------------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | --------------------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 267 | | 64452 (12)| 00:12:54 | | 1 | NESTED LOOPS | | 1 | 267 | | 64452 (12)| 00:12:54 | | 2 | NESTED LOOPS | | 1 | 245 | | 64433 (12)| 00:12:54 | | 3 | NESTED LOOPS | | 1 | 224 | | 64432 (12)| 00:12:54 | | 4 | NESTED LOOPS OUTER | | 1 | 201 | | 64431 (12)| 00:12:54 | | 5 | NESTED LOOPS | | 1 | 177 | | 64428 (12)| 00:12:54 | |* 6 | HASH JOIN | | 1 | 150 | | 64427 (12)| 00:12:54 | |* 7 | HASH JOIN | | 1 | 123 | | 43227 (12)| 00:08:39 | |* 8 | HASH JOIN | | 1 | 96 | | 21602 (12)| 00:04:20 | | 9 | TABLE ACCESS BY INDEX ROWID | STAFF_ROSTER | 1 | 28 | | 3 (0)| 00:00:01 | | 10 | NESTED LOOPS | | 11 | 759 | | 138 (1)| 00:00:02 | | 11 | NESTED LOOPS | | 44 | 1804 | | 19 (0)| 00:00:01 | |* 12 | TABLE ACCESS BY INDEX ROWID| STAFF_CROSS_DEPLOY | 44 | 1364 | | 19 (0)| 00:00:01 | |* 13 | INDEX RANGE SCAN | STAFF_CROSS_DEPLOY_NNDX | 73 | | | 9 (0)| 00:00:01 | |* 14 | INDEX UNIQUE SCAN | STAFF_PK | 1 | 10 | | 0 (0)| 00:00:01 | |* 15 | INDEX RANGE SCAN | STAFF_ROSTER_PK | 1 | | | 2 (0)| 00:00:01 | | 16 | VIEW | VW_SQ_2 | 7227 | 190K| | 21463 (12)| 00:04:18 | | 17 | HASH GROUP BY | | 7227 | 268K| 234M| 21463 (12)| 00:04:18 | | 18 | MERGE JOIN | | 5097K| 184M| | 1944 (27)| 00:00:24 | | 19 | SORT JOIN | | 7227 | 127K| | 39 (8)| 00:00:01 | | 20 | INDEX FAST FULL SCAN | STAFF_NUMBER_PK | 7227 | 127K| | 37 (3)| 00:00:01 | |* 21 | SORT JOIN | | 14107 | 275K| 792K| 1731 (20)| 00:00:21 | |* 22 | INDEX FAST FULL SCAN | STAFF_ROSTER_IDX1 | 14107 | 275K| | 1638 (21)| 00:00:20 | | 23 | VIEW | VW_SQ_3 | 7283 | 192K| | 21624 (12)| 00:04:20 | | 24 | HASH GROUP BY | | 7283 | 270K| 236M| 21624 (12)| 00:04:20 | | 25 | MERGE JOIN | | 5136K| 186M| | 1954 (27)| 00:00:24 | | 26 | SORT JOIN | | 7283 | 128K| | 47 (7)| 00:00:01 | | 27 | INDEX FAST FULL SCAN | STAFF_ROSTER_GROUP_PK | 7283 | 128K| | 45 (3)| 00:00:01 | |* 28 | SORT JOIN | | 14107 | 275K| 792K| 1731 (20)| 00:00:21 | |* 29 | INDEX FAST FULL SCAN | STAFF_ROSTER_IDX1 | 14107 | 275K| | 1638 (21)| 00:00:20 | | 30 | VIEW | VW_SQ_1 | 7137 | 188K| | 21200 (12)| 00:04:15 | | 31 | HASH GROUP BY | | 7137 | 264K| 231M| 21200 (12)| 00:04:15 | | 32 | MERGE JOIN | | 5033K| 182M| | 1927 (27)| 00:00:24 | | 33 | SORT JOIN | | 7137 | 125K| | 24 (13)| 00:00:01 | | 34 | TABLE ACCESS FULL | STAFF_CATEGORY | 7137 | 125K| | 22 (5)| 00:00:01 | |* 35 | SORT JOIN | | 14107 | 275K| 792K| 1731 (20)| 00:00:21 | |* 36 | INDEX FAST FULL SCAN | STAFF_ROSTER_IDX1 | 14107 | 275K| | 1638 (21)| 00:00:20 | |* 37 | TABLE ACCESS BY INDEX ROWID | STAFF_WORK | 1 | 27 | | 1 (0)| 00:00:01 | |* 38 | INDEX UNIQUE SCAN | STAFF_WORK_UNDX | 1 | | | 0 (0)| 00:00:01 | | 39 | TABLE ACCESS BY INDEX ROWID | STAFF_DEPLOYMENT_CODE | 1 | 24 | | 3 (0)| 00:00:01 | |* 40 | INDEX RANGE SCAN | STAFF_DEPLOYMENT_CODE_PK | 1 | | | 2 (0)| 00:00:01 | |* 41 | INDEX RANGE SCAN | STAFF_NUMBER_PK | 1 | 23 | | 1 (0)| 00:00:01 | |* 42 | INDEX RANGE SCAN | STAFF_ROSTER_GROUP_PK | 1 | 21 | | 1 (0)| 00:00:01 | | 43 | INLIST ITERATOR | | | | | | | |* 44 | INDEX UNIQUE SCAN | STAFF_CATEGORY_PK | 1 | 22 | | 19 (0)| 00:00:01 | --------------------------------------------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 6 - access("ITEM_1"="ROSTER"."IC_N" AND "ITEM_2"=ROWID) 7 - access("ITEM_5"="ROSTER"."IC_N" AND "ITEM_6"=ROWID) 8 - access("ITEM_3"="ROSTER"."IC_N" AND "ITEM_4"=ROWID) 12 - filter("DEPLOY"."END_DT">=TO_DATE(‘ 2014-03-11 00:00:00‘, ‘syyyy-mm-dd hh24:mi:ss‘) AND "DEPLOY"."START_DT"<=TO_DATE(‘ 2014-03-11 00:00:00‘, ‘syyyy-mm-dd hh24:mi:ss‘)) 13 - access("DEPLOY"."WORK_DEPARTMENT_C"=‘C‘) 14 - access("STAFF"."IC_N"="DEPLOY"."IC_N") 15 - access("STAFF"."IC_N"="ROSTER"."IC_N" AND "ROSTER"."DEPARTMENT_C"=‘C‘) filter(TO_CHAR(INTERNAL_FUNCTION("ROSTER"."SHIFT_D"),‘DD/MM/YYYY‘)=‘11/03/2014‘) 21 - access("SN"."EFFECTIVE_D"<="ROSTER"."SHIFT_D") filter("SN"."EFFECTIVE_D"<="ROSTER"."SHIFT_D") 22 - filter(TO_CHAR(INTERNAL_FUNCTION("ROSTER"."SHIFT_D"),‘DD/MM/YYYY‘)=‘11/03/2014‘) 28 - access("SG"."EFFECTIVE_D"<="ROSTER"."SHIFT_D") filter("SG"."EFFECTIVE_D"<="ROSTER"."SHIFT_D") 29 - filter(TO_CHAR(INTERNAL_FUNCTION("ROSTER"."SHIFT_D"),‘DD/MM/YYYY‘)=‘11/03/2014‘) 35 - access("SC"."EFFECTIVE_D"<="ROSTER"."SHIFT_D") filter("SC"."EFFECTIVE_D"<="ROSTER"."SHIFT_D") 36 - filter(TO_CHAR(INTERNAL_FUNCTION("ROSTER"."SHIFT_D"),‘DD/MM/YYYY‘)=‘11/03/2014‘) 37 - filter("WORK1"."START_DT"<="ROSTER"."SHIFT_D" AND ("WORK1"."END_DT">"ROSTER"."SHIFT_D" OR "WORK1"."END_DT" IS NULL)) 38 - access("WORK1"."IC_N"="ROSTER"."IC_N" AND "WORK1"."HOME_DEPARTMENT_C"=‘C‘) 40 - access("STAFF"."IC_N"="CODE"."IC_N"(+) AND "CODE"."DEPARTMENT_C"(+)=‘C‘) filter(TO_CHAR(INTERNAL_FUNCTION("CODE"."SHIFT_D"(+)),‘DD/MM/YYYY‘)=‘11/03/2014‘) 41 - access("STAFF"."IC_N"="NUM"."IC_N" AND "NUM"."EFFECTIVE_D"="VW_COL_1") filter("NUM"."EFFECTIVE_D"="VW_COL_1") 42 - access("STAFF"."IC_N"="GRP"."IC_N" AND "GRP"."EFFECTIVE_D"="VW_COL_1") filter("GRP"."EFFECTIVE_D"="VW_COL_1") 44 - access("CATEGORY"."IC_N"="ROSTER"."IC_N" AND ("CATEGORY"."CATEGORY_C"=‘CCOS‘ OR "CATEGORY"."CATEGORY_C"=‘CCSOS‘ OR "CATEGORY"."CATEGORY_C"=‘CES‘ OR "CATEGORY"."CATEGORY_C"=‘HEP‘ OR "CATEGORY"."CATEGORY_C"=‘IGCMO‘ OR "CATEGORY"."CATEGORY_C"=‘IGCMO2‘ OR "CATEGORY"."CATEGORY_C"=‘IGCMO3‘ OR "CATEGORY"."CATEGORY_C"=‘LS‘ OR "CATEGORY"."CATEGORY_C"=‘LSO‘ OR "CATEGORY"."CATEGORY_C"=‘LSUP‘ OR "CATEGORY"."CATEGORY_C"=‘NTCES‘ OR "CATEGORY"."CATEGORY_C"=‘RQC‘ OR "CATEGORY"."CATEGORY_C"=‘SOCMO‘ OR "CATEGORY"."CATEGORY_C"=‘SOCMO2‘ OR "CATEGORY"."CATEGORY_C"=‘SOCMO3‘ OR "CATEGORY"."CATEGORY_C"=‘SOSOS‘ OR "CATEGORY"."CATEGORY_C"=‘SQC‘ OR "CATEGORY"."CATEGORY_C"=‘SVCOS‘ OR "CATEGORY"."CATEGORY_C"=‘WCES‘ OR "CATEGORY"."CATEGORY_C"=‘WOS‘ OR "CATEGORY"."CATEGORY_C"=‘YOCMO‘ OR "CATEGORY"."CATEGORY_C"=‘YOCMO2‘ OR "CATEGORY"."CATEGORY_C"=‘YOCMO3‘) AND "CATEGORY"."EFFECTIVE_D"="VW_COL_1") 91 rows selected.
OK. 不用纠结于复杂的SQL 逻辑，只要知道hash 占用大量的TEMP，解决办法就是避免HASH.
而存在hash的原因就是因为 subquery 被 unnest了，那么我们可以避免 unnest，方法是设置下面几个参数之一，或者用hint
SQL > alter session set "_optimizer_cost_based_transformation" = OFF ; SQL > alter session set "_gby_hash_aggregation_enabled" = FALSE ; SQL > alter session set "_unnest_subquery"=FALSE;
通过设置参数后得到的execution plan如下
PLAN_TABLE_OUTPUT -------------------------------------------------------------------------------------------------------------------------------- Plan hash value: 2603715670 --------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | --------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 186 | 202 (1)| 00:00:03 | | 1 | NESTED LOOPS | | 1 | 186 | 196 (1)| 00:00:03 | | 2 | NESTED LOOPS | | 1 | 164 | 177 (1)| 00:00:03 | | 3 | NESTED LOOPS | | 1 | 143 | 176 (1)| 00:00:03 | | 4 | NESTED LOOPS | | 1 | 120 | 175 (1)| 00:00:03 | | 5 | NESTED LOOPS OUTER | | 11 | 1023 | 164 (1)| 00:00:02 | | 6 | NESTED LOOPS | | 11 | 759 | 138 (1)| 00:00:02 | | 7 | NESTED LOOPS | | 44 | 1804 | 19 (0)| 00:00:01 | |* 8 | TABLE ACCESS BY INDEX ROWID| STAFF_CROSS_DEPLOY | 44 | 1364 | 19 (0)| 00:00:01 | |* 9 | INDEX RANGE SCAN | STAFF_CROSS_DEPLOY_NNDX | 73 | | 9 (0)| 00:00:01 | |* 10 | INDEX UNIQUE SCAN | STAFF_PK | 1 | 10 | 0 (0)| 00:00:01 | | 11 | TABLE ACCESS BY INDEX ROWID | STAFF_ROSTER | 1 | 28 | 3 (0)| 00:00:01 | |* 12 | INDEX RANGE SCAN | STAFF_ROSTER_PK | 1 | | 2 (0)| 00:00:01 | | 13 | TABLE ACCESS BY INDEX ROWID | STAFF_DEPLOYMENT_CODE | 1 | 24 | 3 (0)| 00:00:01 | |* 14 | INDEX RANGE SCAN | STAFF_DEPLOYMENT_CODE_PK | 1 | | 2 (0)| 00:00:01 | |* 15 | TABLE ACCESS BY INDEX ROWID | STAFF_WORK | 1 | 27 | 1 (0)| 00:00:01 | |* 16 | INDEX UNIQUE SCAN | STAFF_WORK_UNDX | 1 | | 0 (0)| 00:00:01 | |* 17 | INDEX RANGE SCAN | STAFF_NUMBER_PK | 1 | 23 | 1 (0)| 00:00:01 | | 18 | SORT AGGREGATE | | 1 | 18 | | | |* 19 | INDEX RANGE SCAN | STAFF_NUMBER_PK | 1 | 18 | 2 (0)| 00:00:01 | |* 20 | INDEX RANGE SCAN | STAFF_ROSTER_GROUP_PK | 1 | 21 | 1 (0)| 00:00:01 | | 21 | SORT AGGREGATE | | 1 | 18 | | | |* 22 | INDEX RANGE SCAN | STAFF_ROSTER_GROUP_PK | 1 | 18 | 2 (0)| 00:00:01 | | 23 | INLIST ITERATOR | | | | | | |* 24 | INDEX UNIQUE SCAN | STAFF_CATEGORY_PK | 1 | 22 | 19 (0)| 00:00:01 | | 25 | SORT AGGREGATE | | 1 | 18 | | | |* 26 | INDEX RANGE SCAN | STAFF_CATEGORY_PK | 1 | 18 | 2 (0)| 00:00:01 | ---------------------------------------------------------------------------------------------------------------
但是用hint的方式去避免unnest 则没有成功，需要测试。
另外疑惑的地方是，max 这种函数在subquery里应该是不unnest的，为什么这里unnest了？
总结一下， 这个case的问题就在于temp占用高，而hash是典型的占用空间多，所以应该自然而然的想到其它join。
优化实例- not use hash to avoid temp space issue,布布扣,bubuko.com
优化实例- not use hash to avoid temp space issue
原文：http://www.cnblogs.com/kramer/p/3608343.html
