整个过程基本上没有什么什么新意，仅为留一下一个脚印。
在一个非常空闲的系统中，用AWR发现一条语句的consistent gets比较高：
SELECT * FROM ( SELECT RESOURCES.CATALOGID, ARTICLE.TITLE AS TITLE, ARTICLE.ID AS ID, ARTICLE.HOTDESC AS BRIEF, ARTICLE.CREATORTIME AS TIME, ARTICLE.PIC AS PICTURE, ARTICLE.ISPICARTICLE AS ISPICARTICLE, RESOURCES.URL AS FILEURL, ARTICLE.TYPE AS TYPE, ARTICLE.LINKURL AS LINKURL, RESOURCES.SITE, RESOURCES.ORDERLINE AS L_ORDERS, TO_CHAR(ARTICLE.CREATORTIME, 'YYYY') AS YEAR, TO_CHAR(ARTICLE.CREATORTIME, 'MM') AS MONTH, TO_CHAR(ARTICLE.CREATORTIME, 'DD') AS DAY, TO_CHAR(ARTICLE.CREATORTIME, 'HH24') AS HOUR, TO_CHAR(ARTICLE.CREATORTIME, 'MI') AS MINUTE, TO_CHAR(ARTICLE.CREATORTIME, 'SS') AS SECOND, ARTICLE.ISNEW, ARTICLE.ISHOT, ARTICLE.TITLEFONTCOLOR, ARTICLE.TITLEFONTNAME, ARTICLE.TITLEBOLD , ARTICLE.ORDERS, ARTICLE.ISINSTANCY FROM ARTICLE, RESOURCES WHERE 1=1 AND RESOURCES.STATE=1 AND TO_CHAR(ARTICLE.ID)=TRIM(RESOURCES.SOURCEID) AND RESOURCES.CATALOGID=20794 ORDER BY ISINSTANCY DESC, L_ORDERS DESC ) WHERE ROWNUM<31
执行计划及统计信息如下：
--------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
--------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 30 | 38790 | 2972 (1)|
| 1 | COUNT STOPKEY | | | | |
| 2 | VIEW | | 126 | 159K| 2972 (1)|
| 3 | SORT ORDER BY STOPKEY | | 126 | 18018 | 2972 (1)|
| 4 | HASH JOIN | | 126 | 18018 | 2972 (1)|
| 5 | TABLE ACCESS BY INDEX ROWID| RESOURCES | 126 | 7560 | 48 (0)|
| 6 | INDEX RANGE SCAN | IDX_RESOURCES | 126 | | 1 (0)|
| 7 | TABLE ACCESS FULL | ARTICLE | 60606 | 4912K| 2923 (1)|
--------------------------------------------------------------------------------------
Statistics
----------------------------------------------------------
0 recursive calls
0 db block gets
13175 consistent gets
0 physical reads
0 redo size
2654 bytes sent via SQL*Net to client
469 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
6 rows processed
在内存查询where子句中的 RESOURCES.CATALOGID=20794 ，估计就是引起对 IDX_RESOURCES 的 RANGE SCAN ，而HASH JOIN 应该是由于 TO_CHAR(ARTICLE.ID)=TRIM(RESOURCES.SOURCEID) 这个条件引起的，由于ARTICLE.ID被TO_CHAR函数包裹着，所以，一般的索引是不会被使用的：
select i.index_name,i.index_type,c.column_name from user_indexes i,user_ind_columns c where i.table_name='ARTICLE' and i.index_name=c.index_name order by i.index_name INDEX_NAME INDEX_TYPE COLUMN_NAME ------------------------------ --------------------------- ------------------------------ PK_ARTICLE NORMAL ID
于是乎给ARTICLE.ID建一个TO_CHAR的函数索引：
create index fidx__to_char_ARTICLE_ID on ARTICLE(TO_CHAR(id));
执行计划改变了：
-------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
-------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 30 | 38790 | 1561 (1)|
| 1 | COUNT STOPKEY | | | | |
| 2 | VIEW | | 76140 | 93M| 1561 (1)|
| 3 | SORT ORDER BY STOPKEY | | 76140 | 10M| 1561 (1)|
| 4 | NESTED LOOPS | | 76140 | 10M| 1561 (1)|
| 5 | TABLE ACCESS BY INDEX ROWID| RESOURCES | 126 | 7560 | 48 (0)|
| 6 | INDEX RANGE SCAN | IDX_RESOURCES | 126 | | 1 (0)|
| 7 | TABLE ACCESS BY INDEX ROWID| ARTICLE | 606 | 50298 | 12 (0)|
| 8 | INDEX RANGE SCAN | FIDX__TO_CHAR_ARTICLE_ID | 12 | | 1 (0)|
-------------------------------------------------------------------------------------------------
Statistics
----------------------------------------------------------
0 recursive calls
0 db block gets
21 consistent gets
0 physical reads
0 redo size
2654 bytes sent via SQL*Net to client
469 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
6 rows processed
大大减少了 consistent gets ，这系统变得更加空闲了。