6.索引的重建

2013-08-05

－－－－－－－－－－－－－－－－索引的重建－－－－－－－－－－－－－－－－－－－－

什么情况下需要重建索引？

alter index index_name rebuild;  ——当索引很大的时候，会执行很长时间，严重影响性能。

1、如果B树索引失效了是必须要重建的。

2、如果没有失效，怎样判断需要重建：

当索引占用空间超过了一个数据块，且符合下面任何一个条件的时候：

A、 B－TREE的高度大于3

B、 数据删除率大于20％

C、 索引可使用的百分比小于75％

SQL> create index ind_obj_id on tt1(object_id);

Index created.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

207928          2         87           0%

SQL> insert into tt1 select * from all_objects;

41173 rows created.

SQL> commit;

Commit complete.

SQL> insert into tt1 select * from all_objects;

41173 rows created.

SQL> commit;

Commit complete.

SQL> alter index ind_obj_id rebuild;

Index altered.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

   1671196          2         90 0%

删除数据：

SQL> delete from tt1 where rownum<=100000;

93519 rows deleted.

SQL> rollback;

Rollback complete.

SQL> delete from tt1 where rownum<=60000;

60000 rows deleted.

SQL> delete from tt1 where rownum<=10000;

10000 rows deleted.

SQL> commit;

Commit complete.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

1671196         2         90    74.8510997765160020958307937424480586833%

重建索引：

SQL> alter index ind_obj_id rebuild;

Index altered.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

431816             2         88     0%

pct free = 10％

插入数据：

SQL> insert into tt1 select * from all_objects;

41173 rows created.

SQL> insert into tt1 select * from all_objects;

41173 rows created.

SQL> insert into tt1 select * from all_objects;

41173 rows created.

SQL> insert into tt1 select * from all_objects;

41173 rows created.

SQL> insert into tt1 select * from all_objects;

41173 rows created.

SQL> commit;

Commit complete.

SQL> insert into tt1 select * from tt1;

229384 rows created.

SQL> commit;

Commit complete.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

11106604           3       66          0%

说明刚才插入的值有24％的索引空间是没有生效的，索引键值没有生效。

SQL> alter index ind_obj_id rebuild;

Index altered.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

   8164012     3         90          0%

少量插入数据：

SQL> insert into tt1 select * from tt1 where rownum<=500;

500 rows created.

SQL> insert into tt1 select * from tt1 where rownum<=500;

500 rows created.

SQL> commit;

Commit complete.

SQL> insert into tt1 select * from tt1 where rownum<=500;

500 rows created.

SQL> insert into tt1 select * from tt1 where rownum<=500;

500 rows created.

SQL> commit;

Commit complete.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

   8252000       3         89         0%

SQL> insert into tt1 select * from tt1 where rownum<=500;

500 rows created.

SQL> commit;

Commit complete.

SQL> insert into tt1 select * from tt1 where rownum<=500;

500 rows created.

SQL> commit;

Commit complete.

SQL> insert into tt1 select * from tt1 where rownum<=1000;

1000 rows created.

SQL> commit;

Commit complete.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

   8299976          3         89 0%

插入少量数据的时候，索引会自动维护的。

update操作是否也会影响索引：

SQL> update tt1 set object_id=object_id+1;

461768 rows updated.

SQL> commit;

Commit complete.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

  16599888       3        67      33.4890835606691306314221762589409944777%

SQL> alter index ind_obj_id rebuild;

Index altered.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

   8219984      3         90          0%

update少量的数据：

SQL> update tt1 set object_id=object_id+1 where rownum<=500;

500 rows updated.

SQL> commit;

Commit complete.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

   8219984          3         90 .108162364688881774208900464665518703436%

SQL> update tt1 set object_id=object_id+1 where rownum<=1000;

1000 rows updated.

SQL> commit;

Commit complete.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

   8219984          3         90 .216091000242021920271064550703592296788%

---结果：说明update操作队索引基本无影响，可以自动去维护。--

------------------------------------------------------

问题：

按照需求要求表上要创建索引，但是表是空表，先插入数据在创建索引性能好高？

是先创建索引再插入数据性能高？（先插入数据再建索引好）

实验测试：

先创建索引：

SQL> truncate table tt1;

Table truncated.

SQL> alter index ind_obj_id rebuild;

Index altered.

SQL> insert into tt1 select * from all_objects;  大批量插入数据

41173 rows created.

SQL> commit;

Commit complete.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

   1247408          2         53 0%    －－索引不会自动维护的。

SQL> alter index ind_obj_id rebuild;

Index altered.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

    743660          2         89 0%

插入少量数据也是这样的结果。而且数据量非常少的时候，即使重建了索引，pct_used也达不到90％，

也就是说数据量很小的话，没有必要作这种检查，数据量小即使全表扫表也问题不大。

检查这几个指标的时候要找到50000条数据以上的表才有意义。

结论：当空表中插入大量数据的时候，索引会失效的，所以应该是先插入数据再创建索引。

----------------------------------------------------------------------------

先删除20000条数据，再插入20000条数据，碎片空间会不会被重用？包括表和索引的，

如果能被重用，索引是否会有影响？

SQL> truncate table tt1;

Table truncated.

SQL> insert into tt1 select * from all_objects;

41173 rows created.

SQL> insert into tt1 select * from all_objects;

41173 rows created.

SQL> insert into tt1 select * from all_objects;

41173 rows created.

SQL> commit;

Commit complete.

SQL> alter index ind_obj_id rebuild;

Index altered.

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

   2198932          2         90   0%

SQL> select count(1) from user_extents where segment_name='IND_OBJ_ID';

 COUNT(1)

----------

       18

SQL> select count(1) from user_extents where segment_name='TT1';

 COUNT(1)

----------

       29

SQL> insert into tt1 select * from all_objects  where rownum<=20000;

20000 rows created.

SQL> commit;

Commit complete.

SQL> select count(1) from user_extents where segment_name='TT1';   －－都没有在HWM之上去开辟空间，碎片能被重用。

 COUNT(1)

----------

       29

SQL> select count(1) from user_extents where segment_name='IND_OBJ_ID';

 COUNT(1)

----------

18

SQL> analyze index ind_obj_id validate structure;

Index analyzed.

SQL> select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

BTREE_SPACE     HEIGHT   PCT_USED DELETE_PCT

----------- ---------- ---------- -----------------------------------------

   2262900      2         88      .740115718418514946962391513982642237223%

加大数据量操作也是这样的效果。

操作                          索引是否会失效         索引是否会自动维护

－－－－－－－－－－－－－－－－－－－－－－－－－－－－----------------

move操作                          会                   不会

shrink操作                       不会                  不会

少量的insert操作                 不会                   会

大量的insert操作                 不会                  不会

少量的update操作                 不会                   会

大量的update操作                 不会                  不会

delete操作                       不会                  不会

truncate操作                     不会                   会

空表中插入大量数据               不会                   不会

删除数据后插入同样的数据         不会                   会

－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－

索引的监控和维护是数据库最为难最为麻烦的工作，关键是要求不停的监控，而且修复，

修复的过程中rebuild对于大表时间会很久，历来DBA工作的难点。

index_stats——这个视图是创建在提示表空间上的会话级别的视图，查看之前必须要先分析索引和表数据的对应性，在当前会话查询才能查到值。

analyze index ind_obj_id validate structure;

select btree_space,height,pct_used,(del_lf_rows/decode(lf_rows,0,1,lf_rows))*100||'%' as delete_pct from index_stats;

查询索引的基本信息：

select * from user_indexes where index_name='IND_OBJ_ID'

如果上面的视图和select * from index_stats;查询的结果有些相同指标值不一样的话，说明这个索引没有分析，执行分析之后就可以了。

创建索引的时候指定表空间，比如将表数据和索引数据分两个磁盘存储，这样在读取的时候就能分摊IO的压力。企业部署用不同的磁盘存储索引，这样IO分摊，访问速度就会很快。

create index xxxx on tb_name(column_name) tablespace xxxxxx;

create index ind_obj_id on tt1(object_id) tablespace mytbs3;

转载于:https://blog.51cto.com/chenhuican/1279476