一:准备测试数据,使用HR用户,创建T1表,插入约30W的数据,并根据object_id创建普通索引,表占存储空间34M左右,索引占6M左右的存储空间
- SQL> conn /as sysdba
- 已连接。
- SQL> select default_tablespace from dba_users where username='HR';
- DEFAULT_TABLESPACE
- ------------------------------------------------------------
- USERS
- SQL> conn hr/hr
- 已连接。
- SQL> insert into t1 select * from t1;
- 已创建 74812 行。
- SQL> insert into t1 select * from t1;
- 已创建 149624 行。
- SQL> commit;
- 提交完成。
- SQL> create index idx_t1_id on t1(object_id);
- 索引已创建。
- SQL> exec dbms_stats.gather_table_stats('HR','T1',CASCADE=>TRUE);
- PL/SQL 过程已成功完成。
- SQL> select count(1) from t1;
- COUNT(1)
- ----------
- 299248
- SQL> select sum(bytes)/1024/1024 from dba_segments where segment_name='T1';
- SUM(BYTES)/1024/1024
- --------------------
- 34.0625
- SQL> select sum(bytes)/1024/1024 from dba_segments where segment_name='IDX_T1_ID';
- SUM(BYTES)/1024/1024
- --------------------
- 6
二:估算表在高水位线下还有多少空间可用,这个值应当越低越好,表使用率越接近高水位线,全表扫描所做的无用功也就越少!
DBMS_STATS包无法获取EMPTY_BLOCKS统计信息,所以需要用analyze命令再收集一次统计信息
- SQL> SELECT blocks, empty_blocks, num_rows FROM user_tables WHERE table_name ='T1';
- BLOCKS EMPTY_BLOCKS NUM_ROWS
- ---------- ------------ ----------
- 4302 0 299248
- SQL> analyze table t1 compute statistics;
- 表已分析。
- SQL> SELECT blocks, empty_blocks, num_rows FROM user_tables WHERE table_name ='T1';
- BLOCKS EMPTY_BLOCKS NUM_ROWS
- ---------- ------------ ----------
- 4302 50 299248
- SQL> col table_name for a20
- SQL> SELECT TABLE_NAME,
- 2 (BLOCKS * 8192 / 1024 / 1024) -
- 3 (NUM_ROWS * AVG_ROW_LEN / 1024 / 1024) "Data lower than HWM in MB"
- 4 FROM USER_TABLES
- 5 WHERE table_name = 'T1';
- TABLE_NAME Data lower than HWM in MB
- -------------------- -------------------------
- T1 5.07086182
三: 查看执行计划,全表扫描大概需要消耗CPU 1175
- SQL> explain plan for select * from t1;
- 已解释。
- SQL> select * from table(dbms_xplan.display);
- PLAN_TABLE_OUTPUT
- --------------------------------------------------------------------------------
- Plan hash value: 3617692013
- --------------------------------------------------------------------------
- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
- --------------------------------------------------------------------------
- | 0 | SELECT STATEMENT | | 299K| 28M| 1175 (1)| 00:00:15 |
- | 1 | TABLE ACCESS FULL| T1 | 299K| 28M| 1175 (1)| 00:00:15 |
- --------------------------------------------------------------------------
四:删除大部分数据,收集统计信息,全表扫描依然需要消耗CPU 1168
- SQL> delete from t1 where object_id>100;
- 已删除298852行。
- SQL> commit;
- 提交完成。
- SQL> select count(*) from t1;
- COUNT(*)
- ----------
- 396
- SQL> exec dbms_stats.gather_table_stats('HR','T1',CASCADE=>TRUE);
- PL/SQL 过程已成功完成。
- SQL> analyze table t1 compute statistics;
- 表已分析。
- SQL> SELECT blocks, empty_blocks, num_rows FROM user_tables WHERE table_name ='T1';
- BLOCKS EMPTY_BLOCKS NUM_ROWS
- ---------- ------------ ----------
- 4302 50 396
- SQL> explain plan for select * from t1;
- 已解释。
- SQL> select * from table(dbms_xplan.display);
- PLAN_TABLE_OUTPUT
- ------------------------------------------------------------------------------
- Plan hash value: 3617692013
- --------------------------------------------------------------------------
- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
- --------------------------------------------------------------------------
- | 0 | SELECT STATEMENT | | 396 | 29700 | 1168 (1)| 00:00:15 |
- | 1 | TABLE ACCESS FULL| T1 | 396 | 29700 | 1168 (1)| 00:00:15 |
- --------------------------------------------------------------------------
五:估算表在高水位线下还有多少空间是无数据的,但在全表扫描时又需要做无用功的数据
- SQL> SELECT TABLE_NAME,
- 2 (BLOCKS * 8192 / 1024 / 1024) -
- 3 (NUM_ROWS * AVG_ROW_LEN / 1024 / 1024) "Data lower than HWM in MB"
- 4 FROM USER_TABLES
- 5 WHERE table_name = 'T1';
- TABLE_NAME Data lower than HWM in MB
- -------------------- -------------------------
- T1 33.5791626
六:对表进行碎片整理,重新收集统计信息
- SQL> alter table t1 enable row movement;
- 表已更改。
- SQL> alter table t1 shrink space cascade;
- 表已更改。
- SQL> select sum(bytes)/1024/1024 from dba_segments where segment_name='T1';
- SUM(BYTES)/1024/1024
- --------------------
- .125
- SQL> select sum(bytes)/1024/1024 from dba_segments where segment_name='IDX_T1_ID
- ';
- SUM(BYTES)/1024/1024
- --------------------
- .0625
- SQL> SELECT TABLE_NAME,
- 2 (BLOCKS * 8192 / 1024 / 1024) -
- 3 (NUM_ROWS * AVG_ROW_LEN / 1024 / 1024) "Data lower than HWM in MB"
- 4 FROM USER_TABLES
- 5 WHERE table_name = 'T1';
- TABLE_NAME Data lower than HWM in MB
- -------------------- -------------------------
- T1 33.5791626
- SQL> exec dbms_stats.gather_table_stats('HR','T1',CASCADE=>TRUE);
- PL/SQL 过程已成功完成。
- 这个时候,只剩下0.1M的无用功了,执行计划中,全表扫描也只需要消耗CPU 3
- SQL> SELECT TABLE_NAME,
- 2 (BLOCKS * 8192 / 1024 / 1024) -
- 3 (NUM_ROWS * AVG_ROW_LEN / 1024 / 1024) "Data lower than HWM in MB"
- 4 FROM USER_TABLES
- 5 WHERE table_name = 'T1';
- TABLE_NAME Data lower than HWM in MB
- -------------------- -------------------------
- T1 .010738373
- SQL> select * from table(dbms_xplan.display);
- PLAN_TABLE_OUTPUT
- --------------------------------------------------------------------------------
- Plan hash value: 3617692013
- --------------------------------------------------------------------------
- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
- --------------------------------------------------------------------------
- | 0 | SELECT STATEMENT | | 396 | 29700 | 3 (0)| 00:00:01 |
- | 1 | TABLE ACCESS FULL| T1 | 396 | 29700 | 3 (0)| 00:00:01 |
- --------------------------------------------------------------------------
- 总共只有5个块,空块却有50个,明显empty_blocks信息过期
- SQL> select blocks,empty_blocks,num_rows from user_tables where table_name='T1';
- BLOCKS EMPTY_BLOCKS NUM_ROWS
- ---------- ------------ ----------
- 5 50 396
- SQL> analyze table t1 compute statistics;
- 表已分析。
- SQL> select blocks,empty_blocks,num_rows from user_tables where table_name='T1';
- BLOCKS EMPTY_BLOCKS NUM_ROWS
- ---------- ------------ ----------
- 5 3 396
参考:http://surachartopun.com/2011/08/determine-hwm-and-reduce-it-by-shrink.html
关于如何确定哪些表需要进行碎片整理,可以使用附件中的脚本去查询,具体请参考:http://www.toadworld.com/KNOWLEDGE/KnowledgeXpertforOracle/tabid/648/TopicID/OTNN18/Default.aspx