公众号：杨建荣的学习笔记 · 作者：r2笔记41天 · 发布：2014-07-22 18:52:46 · 原文链接

今天在生产环境中发现一条sql语句尽管走了主键索引，但是查询还是很慢。
sql语句类似下面的形式：
SELECT /*+ index (bl1_cyc_payer_pop BL1_CYC_PAYER_POP_PK) */
T_TAX.BA_NO,
T_TAX.TOTAL_TAX_AMT,
T_TAX.TAX_RELATION,
T_TAX_ITEM.TAX_ITEM_SEQ_NO,
T_TAX_ITEM.TAX_SEQ_NO,
T_TAX_ITEM.TAX_AUTHORITY,
T_TAX_ITEM.TAX_TYPE,
T_TAX_ITEM.TAX_RATE,
T_TAX_ITEM.TAX_AMOUNT,
T_TAX_ITEM.TAXABLE_AMOUNT,
…….
FROM T_TAX, T_TAX_ITEM, T_DOCUMENT, T_CYC_PAYER_POP –这几张都是大表，少则500万左右，多则1000多万。
WHERE T_TAX.TAX_ITEM_PERIOD_KEY = T_TAX_ITEM.PERIOD_KEY
AND T_TAX.CUSTOMER_KEY = T_CYC_PAYER_POP.CUSTOMER_KEY
AND T_TAX.BA_NO = T_CYC_PAYER_POP.BA_NO
AND T_TAX.CYCLE_SEQ_NO = T_CYC_PAYER_POP.CYCLE_SEQ_NO
AND T_TAX.CYCLE_SEQ_RUN = T_CYC_PAYER_POP.CYCLE_SEQ_RUN
AND T_TAX.PERIOD_KEY = T_CYC_PAYER_POP.PERIOD_KEY
AND T_TAX_ITEM.CUSTOMER_KEY = T_CYC_PAYER_POP.CUSTOMER_KEY
AND T_TAX_ITEM.TAX_SEQ_NO = T_TAX.TAX_SEQ_NO
AND T_DOCUMENT.PERIOD_KEY = T_CYC_PAYER_POP.PERIOD_KEY
AND T_DOCUMENT.CUSTOMER_KEY = T_CYC_PAYER_POP.CUSTOMER_KEY
AND T_DOCUMENT.BA_NO = T_CYC_PAYER_POP.BA_NO
AND T_DOCUMENT.CYCLE_SEQ_NO = T_CYC_PAYER_POP.CYCLE_SEQ_NO
AND T_DOCUMENT.CYCLE_SEQ_RUN = T_CYC_PAYER_POP.CYCLE_SEQ_RUN
AND T_DOCUMENT.DOC_PRODUCE_IND IN (‘Y’, ‘E’)
AND T_CYC_PAYER_POP.CUSTOMER_KEY = 78
AND T_CYC_PAYER_POP.PERIOD_KEY = 55
AND T_CYC_PAYER_POP.QA_GROUP = 3
AND T_CYC_PAYER_POP.CYCLE_SEQ_NO = 2925

查看该sql的执行计划
发现有严重的io问题，瓶颈就在于使用的primary key对应的index

这个问题很值得深究，完全可以使用如下的方式来验证。我尝试使用pk的Hint，另外不加任何hint，看表查询的时候会不会使用index
–>使用hint强制走主键查询
Plan hash value: 3105767292
—————————————————————————————————————————
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop |
—————————————————————————————————————————
| 0 | SELECT STATEMENT | | 1 | 12 | 2501 (1)| 00:00:31 | | |
| 1 | PARTITION RANGE SINGLE | | 1 | 12 | 2501 (1)| 00:00:31 | 171 | 171 |
|* 2 | TABLE ACCESS BY LOCAL INDEX ROWID| T_CYC_PAYER_POP | 1 | 12 | 2501 (1)| 00:00:31 | 171 | 171 |
|* 3 | INDEX FULL SCAN | T_CYC_PAYER_POP_PK | 541 | | 2444 (1)| 00:00:30 | 171 | 171 |
—————————————————————————————————————————

Predicate Information (identified by operation id):
—————————————————
2 – filter(“T_CYC_PAYER_POP”.”QA_GROUP”=3)
3 – access(“T_CYC_PAYER_POP”.”CYCLE_SEQ_NO”=2925 AND “T_CYC_PAYER_POP”.”PERIOD_KEY”=55 AND
“T_CYC_PAYER_POP”.”CUSTOMER_KEY”=78)
filter(“T_CYC_PAYER_POP”.”CUSTOMER_KEY”=78 AND “T_CYC_PAYER_POP”.”CYCLE_SEQ_NO”=2925 AND
“T_CYC_PAYER_POP”.”PERIOD_KEY”=55)
Statistics
———————————————————-
1 recursive calls
0 db block gets
31395 consistent gets
0 physical reads
0 redo size
910 bytes sent via SQL*Net to client
520 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
8 rows processed

–>来看看不使用hint之后，发生了什么
select T_CYC_PAYER_POP.CUSTOMER_KEY,
T_CYC_PAYER_POP.PERIOD_KEY,
T_CYC_PAYER_POP.QA_GROUP ,
T_CYC_PAYER_POP.CYCLE_SEQ_NO
from T_CYC_PAYER_POP where T_CYC_PAYER_POP.CUSTOMER_KEY = 78
AND T_CYC_PAYER_POP.PERIOD_KEY = 55
AND T_CYC_PAYER_POP.QA_GROUP = 3
AND T_CYC_PAYER_POP.CYCLE_SEQ_NO = 2925
/

Execution Plan
———————————————————-
Plan hash value: 23637115
—————————————————————————————————————————-
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop |
—————————————————————————————————————————-
| 0 | SELECT STATEMENT | | 1 | 12 | 2 (0)| 00:00:01 | | |
| 1 | PARTITION RANGE SINGLE | | 1 | 12 | 2 (0)| 00:00:01 | 171 | 171 |
|* 2 | TABLE ACCESS BY LOCAL INDEX ROWID| T_CYC_PAYER_POP | 1 | 12 | 2 (0)| 00:00:01 | 171 | 171 |
|* 3 | INDEX RANGE SCAN | T_CYC_PAYER_POP_5IX | 535 | | 1 (0)| 00:00:01 | 171 | 171 |
—————————————————————————————————————————-
Predicate Information (identified by operation id):
—————————————————
2 – filter(“T_CYC_PAYER_POP”.”CUSTOMER_KEY”=78 AND “T_CYC_PAYER_POP”.”PERIOD_KEY”=55)
3 – access(“T_CYC_PAYER_POP”.”QA_GROUP”=3 AND “T_CYC_PAYER_POP”.”CYCLE_SEQ_NO”=2925)

Statistics
———————————————————-
1 recursive calls
0 db block gets
951 consistent gets
644 physical reads
80 redo size
910 bytes sent via SQL*Net to client
520 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
8 rows processed
启用了index range scan,而且从执行计划和统计信息来看，明显要比全索引扫描效率高得多。
可以看到使用index range scan之后，先查询了索引列的信息，然后无法走索引过滤了其他的条件。根据目前的数据情况，这个效率要比全索引效率还高的多。

以下是做了hint的改动之后，统计信息的情况，可以看到明显的改善。对于这个Hint的细节需要和客户做更多的确认，毕竟对于调优不能越调越差，稳定和高效才是关键。
Statistics
———————————————————-
1 recursive calls
0 db block gets
1178 consistent gets
756 physical reads
0 redo size
3229 bytes sent via SQL*Net to client
553 bytes received via SQL*Net from client
5 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
58 rows processed