SQL 调优——解释计划中的大量“字节”答案

【问题标题】：SQL tuning -- high number of "Bytes" in explain planSQL 调优——解释计划中的大量“字节”
【发布时间】：2011-12-10 05:54:59
【问题描述】：

您能否就可能导致 8.2 具有大量“字节”的原因提供任何想法？提前致谢。

   8.2 INDEX UNIQUE SCAN S_ORG_EXT_P1 -T146208  (Cost = 1 Rows = 9390061 Bytes = 112680732 )Acc=>"A1"."ROW_ID"="A4"."PR_DEPT_OU_ID"

完整的解释计划

 SELECT STATEMENT    Cost =37 (Cost = 37 Rows =  Bytes =  )
1.1 NESTED LOOPS     (Cost = 37 Rows = 1 Bytes = 361 )
 2.1 NESTED LOOPS     (Cost = 34 Rows = 1 Bytes = 317 )
  3.1 NESTED LOOPS     (Cost = 32 Rows = 1 Bytes = 273 )
   4.1 NESTED LOOPS     (Cost = 21 Rows = 1 Bytes = 245 )
    5.1 NESTED LOOPS     (Cost = 19 Rows = 1 Bytes = 201 )
     6.1 NESTED LOOPS     (Cost = 8 Rows = 1 Bytes = 179 )
      7.1 NESTED LOOPS     (Cost = 5 Rows = 1 Bytes = 135 )
       8.1 NESTED LOOPS     (Cost = 4 Rows = 1 Bytes = 123 )
        9.1 TABLE ACCESS BY INDEX ROWID S_OPTY -T203229  (Cost = 3 Rows = 1 Bytes = 29 ) Fil => "A4"."PR_DEPT_OU_ID" IS NOT NULL
         10.1 INDEX UNIQUE SCAN S_OPTY_P1 -T203252  (Cost = 2 Rows = 1 Bytes =  )Acc=>"A4"."ROW_ID"=:V1
        9.2 TABLE ACCESS BY INDEX ROWID S_SALES_METHOD -T134134  (Cost = 1 Rows = 327 Bytes = 30738 )
         10.1 INDEX UNIQUE SCAN S_SALES_METHOD_P1 -T143711  (Cost = 0 Rows = 1 Bytes =  )Acc=>"A5"."ROW_ID"="A4"."SALES_METHOD_ID"
       8.2 INDEX UNIQUE SCAN S_ORG_EXT_P1 -T146208  (Cost = 1 Rows = 9390061 Bytes = 112680732 )Acc=>"A1"."ROW_ID"="A4"."PR_DEPT_OU_ID"
      7.2 INDEX RANGE SCAN S_ORG_EXT_XM_BUCS1 -T246873  (Cost = 3 Rows = 1 Bytes = 44 )Acc=>"A2"."PAR_ROW_ID"="A1"."ROW_ID" AND "A2"."TYPE"='ACCOUNT_TIER' Fil => ("A2"."ATTRIB_03" IS NOT NULL AND "A2"."ATTRIB_04" IS NOT NULL)
     6.2 TABLE ACCESS BY INDEX ROWID S_POS_SKILL_IT -T133800  (Cost = 11 Rows = 10 Bytes = 220 )
      7.1 INDEX RANGE SCAN S_POS_SKILL_IT_F2 -T246871  (Cost = 2 Rows = 10 Bytes =  )Acc=>"A9"."LO_CHAR1"="A5"."NAME"
    5.2 TABLE ACCESS BY INDEX ROWID S_POS_SKILL -T133799  (Cost = 2 Rows = 1 Bytes = 44 ) Fil => "A7"."ASGN_IT_TYPE_NAME"='HPQ Opportunity Sales Method'
     6.1 INDEX UNIQUE SCAN S_POS_SKILL_P1 -T142149  (Cost = 1 Rows = 1 Bytes =  )Acc=>"A7"."ROW_ID"="A9"."POSTN_SKILL_ID"
   4.2 TABLE ACCESS BY INDEX ROWID S_POS_SKILL_IT -T133800  (Cost = 11 Rows = 1 Bytes = 28 ) Fil => ("A8"."LO_CHAR3" IS NOT NULL AND "A8"."LO_CHAR2" IS NOT NULL AND "A8"."LO_CHAR2"="A2"."ATTRIB_04")
    5.1 INDEX RANGE SCAN S_POS_SKILL_IT_F2 -T246871  (Cost = 2 Rows = 10 Bytes =  )Acc=>"A8"."LO_CHAR1"="A2"."ATTRIB_03"
  3.2 TABLE ACCESS BY INDEX ROWID S_POS_SKILL -T133799  (Cost = 2 Rows = 1 Bytes = 44 ) Fil => ("A6"."ASGN_IT_TYPE_NAME"='HPQ BU Tier BUCS' AND "A6"."POSTN_ID"="A7"."POSTN_ID")
   4.1 INDEX UNIQUE SCAN S_POS_SKILL_P1 -T142149  (Cost = 1 Rows = 1 Bytes =  )Acc=>"A6"."ROW_ID"="A8"."POSTN_SKILL_ID"
 2.2 INDEX RANGE SCAN S_ORG_EXT_XM_BUCS1 -T246873  (Cost = 3 Rows = 1 Bytes = 44 )Acc=>"A3"."PAR_ROW_ID"="A1"."ROW_ID" AND "A3"."TYPE"='BU_CUSTOMER_SEGMENT' AND "A8"."LO_CHAR3"."A3"."ATTR Fil" => ("A3"."ATTRIB_03" IS NOT NULL AND "A3"."ATTRIB_04" IS NOT NULL)

【问题讨论】：

标签： sql performance database-tuning

【解决方案1】：

它对超过 900 万行进行索引扫描，其中每行只有 12 个字节。

因此，高字节数是高行数造成的。如果您不希望它访问所有这些数据，则应尽量避免索引扫描。

【讨论】：

嗨 Guffa，感谢您分享您的见解。您能否帮助将您的答案与步骤 9.x 和 10.x 的结果联系起来？实际上，从 9.x 和 10.x 开始，它应该只返回 1 条记录，因此在第 8 步中——我希望它不会扫描整个表。