SELECT .. .. ..
FROM AWP.TTT_TTT_TRANSACTION_2018 INCT
LEFT JOIN AWP.TTT_TTT_TELM TELM
ON INCT.TELLER_NO = TELM.TELLER_NO
LEFT JOIN SJB_CD_01.TMP_TT_BAN_EDP ED1P
ON INCT.TRAN_CODE = ED1P.交易代碼
LEFT JOIN AWP.TTT_TTT_CB_CIF CUSM
ON INCT.CUSTOMER_NO = CUSM.CUSTOMER_NO_MICM
LEFT JOIN AWP.TTT_TTT_CB_ACCT INVM
ON INCT.ACCT_NO = INVM.ACCT_NO
LEFT JOIN AWP.TTT_TTT_BRHM BRHM
ON INVM.BRANCH_NO = BRHM.BRANCH_NO
LEFT JOIN AWP.TTT_TTT_DEPP DEPP
ON INVM.ACCT_TYPE = DEPP.ACCT_TYPE || DEPP.INT_CAT
JOIN SJB_CD_01.TMP_TT_LZ_POS_XYK_ALL XYK
ON XYK.持卡人證件號碼 = CUSM.ID_NO_MICM
WHERE 1 = 1
AND TO_DATE(INCT.TRAN_DATE, 'yyyy-mm-dd') BETWEEN
TO_DATE(XYK.交易日期, 'yyyy-mm-dd') AND
TO_DATE(XYK.交易日期, 'yyyy-mm-dd') + 5
AND INCT.INCT_01_AMOUNT BETWEEN XYK.刷卡金額 * 0.9 AND XYK.刷卡金額
SQL執行6分鐘以後報錯：ORA-12805:並行查詢服務器意外停止
1.查看執行計劃，確認SQL是否進行的並行操作
PLAN HASH VALUE: 1375319919
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
| ID | OPERATION | NAME | ROWS | BYTES | COST (%CPU)| TIME | PSTART| PSTOP | TQ |IN-OUT| PQ DISTRIB |
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 1480 | 14871 (1)| 00:02:59 | | | | | |
| 1 | PX COORDINATOR | | | | | | | | | | |
| 2 | PX SEND QC (RANDOM) | :TQ10009 | 1 | 1480 | 14871 (1)| 00:02:59 | | | Q1,09 | P->S | QC (RAND) |
|* 3 | HASH JOIN OUTER | | 1 | 1480 | 14871 (1)| 00:02:59 | | | Q1,09 | PCWP | |
| 4 | PX RECEIVE | | 1 | 1231 | 14858 (1)| 00:02:59 | | | Q1,09 | PCWP | |
| 5 | PX SEND HASH | :TQ10008 | 1 | 1231 | 14858 (1)| 00:02:59 | | | Q1,08 | P->P | HASH |
|* 6 | HASH JOIN OUTER | | 1 | 1231 | 14858 (1)| 00:02:59 | | | Q1,08 | PCWP | |
| 7 | PX RECEIVE | | 1 | 1103 | 14845 (1)| 00:02:59 | | | Q1,08 | PCWP | |
| 8 | PX SEND HASH | :TQ10007 | 1 | 1103 | 14845 (1)| 00:02:59 | | | Q1,07 | P->P | HASH |
| 9 | NESTED LOOPS OUTER | | 1 | 1103 | 14845 (1)| 00:02:59 | | | Q1,07 | PCWP | |
|* 10 | HASH JOIN OUTER | | 1 | 976 | 14841 (1)| 00:02:59 | | | Q1,07 | PCWP | |
| 11 | PX RECEIVE | | 1 | 854 | 14532 (1)| 00:02:55 | | | Q1,07 | PCWP | |
| 12 | PX SEND HASH | :TQ10006 | 1 | 854 | 14532 (1)| 00:02:55 | | | Q1,06 | P->P | HASH |
|* 13 | HASH JOIN OUTER | | 1 | 854 | 14532 (1)| 00:02:55 | | | Q1,06 | PCWP | |
| 14 | PX RECEIVE | | 1 | 803 | 14517 (1)| 00:02:55 | | | Q1,06 | PCWP | |
| 15 | PX SEND HASH | :TQ10005 | 1 | 803 | 14517 (1)| 00:02:55 | | | Q1,05 | P->P | HASH |
| 16 | NESTED LOOPS | | 1 | 803 | 14517 (1)| 00:02:55 | | | Q1,05 | PCWP | |
| 17 | NESTED LOOPS | | 1131 | 803 | 14517 (1)| 00:02:55 | | | Q1,05 | PCWP | |
| 18 | BUFFER SORT | | | | | | | | Q1,05 | PCWC | |
| 19 | PX RECEIVE | | | | | | | | Q1,05 | PCWP | |
| 20 | PX SEND BROADCAST | :TQ10000 | | | | | | | | S->P | BROADCAST |
| 21 | NESTED LOOPS | | 13 | 6162 | 59 (0)| 00:00:01 | | | | | |
| 22 | TABLE ACCESS FULL | TMP_TT_LZ_POS_XYK_ALL | 14 | 2898 | 3 (0)| 00:00:01 | | | | | |
| 23 | TABLE ACCESS BY INDEX ROWID | TTT_TTT_CB_CIF | 1 | 267 | 4 (0)| 00:00:01 | | | | | |
|* 24 | INDEX RANGE SCAN | IDX1_TTT_TTT_CB_CIF | 1 | | 2 (0)| 00:00:01 | | | | | |
| 25 | PX PARTITION LIST ALL | | 87 | | 1092 (0)| 00:00:14 | 1 | 364 | Q1,05 | PCWC | |
|* 26 | INDEX RANGE SCAN | IND2_BANTRANSACTION_CCDCX_2018 | 87 | | 1092 (0)| 00:00:14 | 1 | 364 | Q1,05 | PCWP | |
|* 27 | TABLE ACCESS BY LOCAL INDEX ROWID| TTT_TTT_TRANSACTION_2018 | 1 | 329 | 1132 (0)| 00:00:14 | 1 | 1 | Q1,05 | PCWP | |
| 28 | BUFFER SORT | | | | | | | | Q1,06 | PCWC | |
| 29 | PX RECEIVE | | 5350 | 266K| 14 (0)| 00:00:01 | | | Q1,06 | PCWP | |
| 30 | PX SEND HASH | :TQ10001 | 5350 | 266K| 14 (0)| 00:00:01 | | | | S->P | HASH |
| 31 | TABLE ACCESS FULL | TMP_TT_BAN_EDP | 5350 | 266K| 14 (0)| 00:00:01 | | | | | |
| 32 | BUFFER SORT | | | | | | | | Q1,07 | PCWC | |
| 33 | PX RECEIVE | | 57656 | 6869K| 309 (1)| 00:00:04 | | | Q1,07 | PCWP | |
| 34 | PX SEND HASH | :TQ10002 | 57656 | 6869K| 309 (1)| 00:00:04 | | | | S->P | HASH |
| 35 | TABLE ACCESS FULL | TTT_TTT_TELM | 57656 | 6869K| 309 (1)| 00:00:04 | | | | | |
| 36 | TABLE ACCESS BY INDEX ROWID | TTT_TTT_CB_ACCT | 1 | 127 | 4 (0)| 00:00:01 | | | Q1,07 | PCWP | |
|* 37 | INDEX RANGE SCAN | IND1_TTT_TTT_CB_ACCT | 1 | | 3 (0)| 00:00:01 | | | Q1,07 | PCWP | |
| 38 | BUFFER SORT | | | | | | | | Q1,08 | PCWC | |
| 39 | PX RECEIVE | | 1937 | 242K| 13 (0)| 00:00:01 | | | Q1,08 | PCWP | |
| 40 | PX SEND HASH | :TQ10003 | 1937 | 242K| 13 (0)| 00:00:01 | | | | S->P | HASH |
| 41 | TABLE ACCESS FULL | TTT_TTT_BRHM | 1937 | 242K| 13 (0)| 00:00:01 | | | | | |
| 42 | BUFFER SORT | | | | | | | | Q1,09 | PCWC | |
| 43 | PX RECEIVE | | 1134 | 275K| 13 (0)| 00:00:01 | | | Q1,09 | PCWP | |
| 44 | PX SEND HASH | :TQ10004 | 1134 | 275K| 13 (0)| 00:00:01 | | | | S->P | HASH |
| 45 | TABLE ACCESS FULL | TTT_TTT_DEPP | 1134 | 275K| 13 (0)| 00:00:01 | | | | | |
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
PREDICATE INFORMATION (IDENTIFIED BY OPERATION ID):
---------------------------------------------------
3 - ACCESS("INVM"."ACCT_TYPE"="DEPP"."ACCT_TYPE"(+)||"DEPP"."INT_CAT"(+))
6 - ACCESS("INVM"."BRANCH_NO"="BRHM"."BRANCH_NO"(+))
10 - ACCESS("INCT"."TELLER_NO"="TELM"."TELLER_NO"(+))
13 - ACCESS("INCT"."TRAN_CODE"="ED1P"."交易代碼"(+))
24 - ACCESS("XYK"."持卡人證件號碼"="CUSM"."ID_NO_MICM")
26 - ACCESS("INCT"."CUSTOMER_NO"="CUSM"."CUSTOMER_NO_MICM")
27 - FILTER(TO_DATE("INCT"."TRAN_DATE",'yyyy-mm-dd')>=TO_DATE("XYK"."交易日期",'yyyy-mm-dd') AND
TO_DATE("INCT"."TRAN_DATE",'yyyy-mm-dd')<=TO_DATE("XYK"."交易日期",'yyyy-mm-dd')+5 AND "INCT"."INCT_01_AMOUNT">="XYK"."刷卡金額"*0.9 AND "INCT"."INCT_01_AMOUNT"<="XYK"."刷卡金額")
37 - ACCESS("INCT"."ACCT_NO"="INVM"."ACCT_NO"(+))
2.查看產生並行操作的原因，SQL語句本身沒有並行相關HINT，查看SQL表和索引信息以及並行信息如下：
OWNER OBJECT_TYPE OBJECT_NAME PARTITIONED SIZE_MB NUM_ROWS LAST_ANALYZED STATUS
-------------- ------------- ----------------------------- ----------- ---------- ---------- ------------- --------------
AWP TABLE TTT_TTT_BRHM NO 0.375 1937 2019/3/4 22:1 統計信息過期
AWP TABLE TTT_TTT_CB_ACCT NO 7011 48184163 2019/3/1 22:2 統計信息過期
AWP TABLE TTT_TTT_CB_CIF NO 9338 30525855 2019/3/2 1:29 統計信息過期
AWP TABLE TTT_TTT_DEPP NO 0.375 1134 2019/3/4 22:1 統計信息過期
AWP TABLE TTT_TTT_TELM NO 9 57656 2019/3/5 22:0 統計信息未過期
AWP TABLE TTT_TTT_TRANSACTION_2018 YES 499774.187 1303013540 2018/3/20 10: 統計信息未過期
SJB_CD_01 TABLE TMP_TT_BAN_EDP NO 0.375 5350 2017/7/5 22:0 統計信息未過期
SJB_CD_01 TABLE TMP_TT_LZ_POS_XYK_ALL NO 0.0625 14 2019/2/22 22: 統計信息未過期
TABLE_NAME DEGREE
------------------------------ --------
SJB_CD_01 TMP_TT_BAN_EDP 1
SJB_CD_01 TMP_TT_LZ_POS_XYK_ALL 1
AWP TTT_TTT_TRANSACTION_2018 1
AWP TTT_TTT_CB_CIF 1
AWP TTT_TTT_BRHM 1
AWP TTT_TTT_TELM 1
AWP TTT_TTT_DEPP 1
AWP TTT_TTT_CB_ACCT 1
INDEX_NAME TABLE_NAME DEGREE
------------------------------ ------------------------------ ----------------------------------------
IND1_TTT_TTT_CB_ACCT TTT_TTT_CB_ACCT 1
IND2_TTT_TTT_CB_ACCT TTT_TTT_CB_ACCT 1
IND3_TTT_TTT_CB_ACCT TTT_TTT_CB_ACCT 8
PK_TTT_TTT_CB_ACCT TTT_TTT_CB_ACCT 1
PK_TTT_TTT_CB_CIF TTT_TTT_CB_CIF 1
IDX1_TTT_TTT_CB_CIF TTT_TTT_CB_CIF 1
IDX2_TTT_TTT_CB_CIF TTT_TTT_CB_CIF 1
IDX3_TTT_TTT_CB_CIF TTT_TTT_CB_CIF 1
IDX1_TTT_TTT_DEPP TTT_TTT_DEPP 1
PK_TTT_TTT_TELM TTT_TTT_TELM 1
IND1_BANTRANSACTION_AITI_2018 TTT_TTT_TRANSACTION_2018 4
IND2_BANTRANSACTION_CCDCX_2018 TTT_TTT_TRANSACTION_2018 4
IND3_BANTRANSACTION_E_2018 TTT_TTT_TRANSACTION_2018 4
IND4_BANTRANSACTION_I_2018 TTT_TTT_TRANSACTION_2018 4
其實腳本可以優化一下：添加一列DEGREE
3.由上可知，因爲索引的並行，唯一的大表TTT_TTT_TRANSACTION_2018使用並行的方式訪問導致以上的ORA報錯。
4.索引並行度產生的原因
a)創建索引的時候，爲了縮短create時間，使用了並行的方式創建
b)分區表對分區操作的時候需要維護(rebuild)索引，使用了並行的方式
解決：只需要關閉索引的並行度即可：alter index xxxxxx noparallel
5.因是線上環境，無權限進行直接alter操作，故使用HINT從SQL語句上消除並行
6./*+ noparallel USE_HASH(INCT TELM ED1P CUSM INVM BRHM DEPP XYK) */ 執行計劃如下
PLAN HASH VALUE: 374343828
---------------------------------------------------------------------------------------------------------------------
| ID | OPERATION | NAME | ROWS | BYTES | COST (%CPU)| TIME | PSTART| PSTOP |
---------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 1480 | 17M (1)| 58:53:47 | | |
|* 1 | HASH JOIN OUTER | | 1 | 1480 | 17M (1)| 58:53:47 | | |
|* 2 | HASH JOIN OUTER | | 1 | 1231 | 17M (1)| 58:53:47 | | |
|* 3 | HASH JOIN OUTER | | 1 | 1103 | 17M (1)| 58:53:46 | | |
|* 4 | HASH JOIN OUTER | | 1 | 976 | 17M (1)| 58:05:07 | | |
|* 5 | HASH JOIN OUTER | | 1 | 854 | 17M (1)| 58:05:03 | | |
|* 6 | HASH JOIN | | 1 | 803 | 17M (1)| 58:05:03 | | |
|* 7 | HASH JOIN | | 13 | 6162 | 325K (1)| 01:05:07 | | |
| 8 | TABLE ACCESS FULL| TMP_TT_LZ_POS_XYK_ALL | 14 | 2898 | 3 (0)| 00:00:01 | | |
| 9 | TABLE ACCESS FULL| TTT_TTT_CB_CIF | 30M| 7772M| 325K (1)| 01:05:05 | | |
| 10 | PARTITION LIST ALL| | 1303M| 399G| 17M (1)| 56:58:32 | 1 | 364 |
| 11 | TABLE ACCESS FULL| TTT_TTT_TRANSACTION_2018 | 1303M| 399G| 17M (1)| 56:58:32 | 1 | 364 |
| 12 | TABLE ACCESS FULL | TMP_TT_BAN_EDP | 5350 | 266K| 14 (0)| 00:00:01 | | |
| 13 | TABLE ACCESS FULL | TTT_TTT_TELM | 57656 | 6869K| 309 (1)| 00:00:04 | | |
| 14 | TABLE ACCESS FULL | TTT_TTT_CB_ACCT | 48M| 5835M| 243K (1)| 00:48:37 | | |
| 15 | TABLE ACCESS FULL | TTT_TTT_BRHM | 1937 | 242K| 13 (0)| 00:00:01 | | |
| 16 | TABLE ACCESS FULL | TTT_TTT_DEPP | 1134 | 275K| 13 (0)| 00:00:01 | | |
---------------------------------------------------------------------------------------------------------------------
PREDICATE INFORMATION (IDENTIFIED BY OPERATION ID):
---------------------------------------------------
1 - ACCESS("INVM"."ACCT_TYPE"="DEPP"."ACCT_TYPE"(+)||"DEPP"."INT_CAT"(+))
2 - ACCESS("INVM"."BRANCH_NO"="BRHM"."BRANCH_NO"(+))
3 - ACCESS("INCT"."ACCT_NO"="INVM"."ACCT_NO"(+))
4 - ACCESS("INCT"."TELLER_NO"="TELM"."TELLER_NO"(+))
5 - ACCESS("INCT"."TRAN_CODE"="ED1P"."交易代碼"(+))
6 - ACCESS("INCT"."CUSTOMER_NO"="CUSM"."CUSTOMER_NO_MICM")
FILTER(TO_DATE("INCT"."TRAN_DATE",'yyyy-mm-dd')>=TO_DATE("XYK"."交易日期",'yyyy-mm-dd') AND
TO_DATE("INCT"."TRAN_DATE",'yyyy-mm-dd')<=TO_DATE("XYK"."交易日期",'yyyy-mm-dd')+5 AND
"INCT"."INCT_01_AMOUNT">="XYK"."刷卡金額"*0.9 AND "INCT"."INCT_01_AMOUNT"<="XYK"."刷卡金額")
7 - ACCESS("XYK"."持卡人證件號碼"="CUSM"."ID_NO_MICM")
7.由第二步可知，SQL語句中只有一張大表，猜測其他都是參數表，是典型的星型架構的SQL語句。這一類的SQL的主要消耗在這一張大表的訪問方式。這裏是全表掃描TABLE ACCESS FULL| TTT_TTT_TRANSACTION_2018 | 1303M。避免出現其他幺蛾子，所以使用HASH固定SQL的執行計劃。這樣一來SQL確實沒有報錯，但是運行了2個小時纔出結果，2個小時，這顯然不能忍......
8.查看SQL語句可以看出大表有過濾條件TO_DATE(INCT.TRAN_DATE, 'yyyy-mm-dd') BETWEEN
TO_DATE(XYK.交易日期, 'yyyy-mm-dd') AND
TO_DATE(XYK.交易日期, 'yyyy-mm-dd') + 5 ，而TRAN_DATE正好是分區鍵，只是SQL是通過傳值的方式從表中獲取數據，順便查了以下表XYZ的數據量，還好只有幾條。這邊把值帶入，量化變量，得出如下SQL語句
select /*+ NOPARALLEL use_hash(inct telm ed1p cusm invm brhm depp xyk) */
......
from awp.t48_ban_transaction_2018 inct
left join awp.t48_ban_telm telm
on inct.teller_no = telm.teller_no
left join sjb_cd_01.tmp_lj_ban_edp ed1p
on inct.tran_code = ed1p.交易代碼
left join awp.t48_ban_cb_cif cusm
on inct.customer_no = cusm.customer_no_micm
left join awp.t48_ban_cb_acct invm
on inct.acct_no = invm.acct_no
left join awp.t48_ban_brhm brhm
on invm.branch_no = brhm.branch_no
left join awp.t48_ban_depp depp
on invm.acct_type = depp.acct_type || depp.int_cat
join sjb_cd_01.tmp_lj_lz_pos_xyk_all xyk
on xyk.持卡人證件號碼 = cusm.id_no_micm
where 1 = 1
and (inct.tran_date > '2018-12-13' AND inct.tran_date < '2018-12-18')
OR (inct.tran_date > '2018-06-28' AND inct.tran_date < '2018-07-04')
OR (inct.tran_date > '2018-11-02' AND inct.tran_date < '2018-11-07')
OR (inct.tran_date > '2018-08-26' AND inct.tran_date < '2018-09-03')
OR (inct.tran_date > '2018-07-25' AND inct.tran_date < '2018-07-30')
and inct.inct_01_amount between xyk.刷卡金額 * 0.9 and xyk.刷卡金額;
SQL使用了分區裁剪，12分鐘出結果~這個結果勉強能忍，但是仍不盡如人意。要求儘量優化到分鐘級
於是我發出提問：這個SQL最終返回結果是多少？
因爲這個數據量要做到分鐘級，顯然HASH不可能，只有NEST LOOP可行，如果返回結果少，說明有可能走NEST LOOP
得到的回覆是：很少最多不超過100條
這樣看來可以走NEST LOOP，那麼怎麼查看是哪張表起的作用呢？顯然我們需要找與大表直接關聯的表,然後驗證這個表是否能夠把與大表關聯的結果集變小，如果可以，嵌套循環驅動到底就行。於是加如下HINT /*+ leading(xyk) noparallel */ 原SQL 0.2秒出結果！！！！