当表随着数据的增加，会使用越来越多的块，HWM会升高，当删除一些数据后，可能会产生很多空块（不包含数据的块），但它们仍在HWM之下。为了降低高水位线，有必要对表进行重组 （reorganization）:
ALTER TABLE T ENABLE ROW MOVEMENT;
ALTER TABLE T SHRINK SPACE;
CREATE TABLE t ( id NUMBER, n1 NUMBER, n2 NUMBER, pad VARCHAR2(4000) ); execute dbms_random.seed(0) INSERT INTO t SELECT rownum AS id, 1+mod(rownum,251) AS n1, 1+mod(rownum,251) AS n2, dbms_random.string('p',255) AS pad FROM dual CONNECT BY level <= 10000 ORDER BY dbms_random.value; REM REM 收集表t的统计信息 REM BEGIN dbms_stats.gather_table_stats( ownname => user, tabname => 'T', estimate_percent => 100, method_opt => 'for all columns size skewonly', cascade => TRUE ); END; / REM REM 将当前session的statistics_level设置为all,以使用v$sql_plan_statistics查看逻辑读信息 REM ALTER SESSION SET statistics_level = all; REM REM 全表扫描t,显示的逻辑读信息为439 REM SELECT /*+ full(t) */ * FROM t WHERE n2 = 19; SELECT last_output_rows, last_cr_buffer_gets, last_cu_buffer_gets FROM v$session s, v$sql_plan_statistics p WHERE s.prev_sql_id = p.sql_id AND s.prev_child_number = p.child_number AND s.sid = sys_context('userenv','sid') AND p.operation_id = 1; LAST_OUTPUT_ROWS LAST_CR_BUFFER_GETS LAST_CU_BUFFER_GETS ---------------- ------------------- ------------------- 40 439 0 REM REM 删除大部分行并重新测试，显示逻辑读信息仍然为439，因为许多无用的空块被访问 REM DELETE t WHERE n2 <> 19; SELECT /*+ full(t) */ * FROM t WHERE n2 = 19; SELECT last_output_rows, last_cr_buffer_gets, last_cu_buffer_gets FROM v$session s, v$sql_plan_statistics p WHERE s.prev_sql_id = p.sql_id AND s.prev_child_number = p.child_number AND s.sid = sys_context('userenv','sid') AND p.operation_id = 1; LAST_OUTPUT_ROWS LAST_CR_BUFFER_GETS LAST_CU_BUFFER_GETS ---------------- ------------------- ------------------- 40 439 0 REM REM 重组织表t,并重新测试，显示的逻辑读为7 REM (this works in Oracle Database 10g only) REM ALTER TABLE t ENABLE ROW MOVEMENT; ALTER TABLE t SHRINK SPACE; SELECT /*+ full(t) */ * FROM t WHERE n2 = 19; SELECT last_output_rows, last_cr_buffer_gets, last_cu_buffer_gets FROM v$session s, v$sql_plan_statistics p WHERE s.prev_sql_id = p.sql_id AND s.prev_child_number = p.child_number AND s.sid = sys_context('userenv','sid') AND p.operation_id = 1; LAST_OUTPUT_ROWS LAST_CR_BUFFER_GETS LAST_CU_BUFFER_GETS ---------------- ------------------- ------------------- 40 7 0 DROP TABLE t;
什么时候应该收缩表空间呢？
从统计信息表中dba_tables，可以计算出表共使用的空间（AVG_ROW_LEN * NUM_ROWS），和占用的块数（BLOCKS）（高水位之下的块），据此大概可知高水位之下空闲块的数量，以决定是否收缩表。
SQL> select dbms_metadata.get_ddl('TABLE','TEST') from dual; DBMS_METADATA.GET_DDL('TABLE','TEST') -------------------------------------------------------------------------- CREATE TABLE "SYS"."TEST" ( "TEXT" CHAR(2000) ) PCTFREE 10 PCTUSED 40 INITRANS 1 MAXTRANS 255 NOCOMPRESS LOGGING STORAGE(INITIAL 65536 NEXT 1048576 MINEXTENTS 1 MAXEXTENTS 2147483645 PCTINCREASE 0 FREELISTS 1 FREELIST GROUPS 1 BUFFER_POOL DEFAULT) TABLESPACE "USERS" SQL> insert into test select object_name from dba_objects where rownum < 3000; 已创建2999行。 SQL> commit; 提交完成。 SQL> exec dbms_stats.gather_table_stats(ownname=>'SYS',tabname=>'TEST',estimate_percent=>30); PL/SQL 过程已成功完成。 SQL> select TABLE_NAME,BLOCKS,EMPTY_BLOCKS,avg_row_len,num_rows from dba_tables where table_name='TE ST' and owner='SYS'; TABLE_NAME BLOCKS EMPTY_BLOCKS AVG_ROW_LEN NUM_ROWS ------------------------------ ---------- ------------ ----------- ---------- TEST 1000 0 2001 2999 SQL> delete from test where rownum <2000; 已删除1999行。 SQL> commit; 提交完成。 SQL> exec dbms_stats.gather_table_stats(ownname=>'SYS',tabname=>'TEST',estimate_percent=>30); PL/SQL 过程已成功完成。 SQL> select TABLE_NAME,BLOCKS,EMPTY_BLOCKS,avg_row_len,num_rows from dba_tables where table_name='TE ST' and owner='SYS'; TABLE_NAME BLOCKS EMPTY_BLOCKS AVG_ROW_LEN NUM_ROWS ------------------------------ ---------- ------------ ----------- ---------- TEST 1000 0 2001 1000 --BLOCKS没有改变，同样是1000。 SQL> set autot traceonly SQL> select * from test; 已选择1000行。 执行计划 ---------------------------------------------------------- Plan hash value: 1357081020 -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1000 | 1954K| 221 (1)| 00:00:03 | | 1 | TABLE ACCESS FULL| TEST | 1000 | 1954K| 221 (1)| 00:00:03 | -------------------------------------------------------------------------- 统计信息 ---------------------------------------------------------- 135 recursive calls 0 db block gets 1076 consistent gets --扫描的块数 1076 0 physical reads 0 redo size 2026014 bytes sent via SQL*Net to client 1111 bytes received via SQL*Net from client 68 SQL*Net roundtrips to/from client 2 sorts (memory) 0 sorts (disk) 1000 rows processed SQL> set autot off SQL> alter table test enable row movement; 表已更改。 SQL> alter table test shrink space; --对表收缩 表已更改。 SQL> exec dbms_stats.gather_table_stats(ownname=>'SYS',tabname=>'TEST',estimate_percent=>30); PL/SQL 过程已成功完成。 SQL> select TABLE_NAME,BLOCKS,EMPTY_BLOCKS,avg_row_len,num_rows from dba_tables where table_name='TE ST' and owner='SYS'; TABLE_NAME BLOCKS EMPTY_BLOCKS AVG_ROW_LEN NUM_ROWS ------------------------------ ---------- ------------ ----------- ---------- TEST 250 0 2001 1000 --对表分析后，BLOCKS为250。 SQL> set autot traceonly SQL> select * from test; 已选择1000行。 执行计划 ---------------------------------------------------------- Plan hash value: 1357081020 -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1000 | 1954K| 56 (0)| 00:00:01 | | 1 | TABLE ACCESS FULL| TEST | 1000 | 1954K| 56 (0)| 00:00:01 | -------------------------------------------------------------------------- 统计信息 ---------------------------------------------------------- 135 recursive calls 0 db block gets 317 consistent gets --扫描的块数变为317 0 physical reads 0 redo size 2026014 bytes sent via SQL*Net to client 1111 bytes received via SQL*Net from client 68 SQL*Net roundtrips to/from client 2 sorts (memory) 0 sorts (disk) 1000 rows processed SQL> set autot offALTER TABLE
shrink_clause
The shrink clause lets you manually shrink space in a table, index-organized table or its overflow segment, index, partition, subpartition, LOB segment, materialized view, or materialized view log. This clause is valid only for segments in tablespaces with automatic segment management. By default, Oracle Database compacts the segment, adjusts the high water mark, and releases the recuperated space immediately.
Compacting the segment requires row movement. Therefore, you must enable row movement for the object you want to shrink before specifying this clause. Further, if your application has any rowid-based triggers, you should disable them before issuing this clause.
COMPACT If you specify COMPACT, then Oracle Database only defragments the segment space and compacts the table rows for subsequent release. The database does not readjust the high water mark and does not release the space immediately. You must issue another ALTER TABLE ... SHRINK SPACE statement later to complete the operation. This clause is useful if you want to accomplish the shrink operation in two shorter steps rather than one longer step.
For an index or index-organized table, specifying ALTER [INDEX | TABLE] ... SHRINK SPACE COMPACT is equivalent to specifying ALTER [INDEX | TABLE ... COALESCE. The shrink_clause can be cascaded (please refer to the CASCADE clause, which follows) and compacts the segment more densely than does a coalesce operation, which can improve performance. However, if you do not want to release the unused space, then you can use the appropriate COALESCE clause.
CASCADE If you specify CASCADE, then Oracle Database performs the same operations on all dependent objects of table, including secondary indexes on index-organized tables.
Restrictions on the shrink_clause The shrink_clause is subject to the following restrictions:
You cannot specify this clause for a cluster, a clustered table, or any object with a LONG column.
Segment shrink is not supported for tables with function-based indexes or bitmap join indexes.
This clause does not shrink mapping tables of index-organized tables, even if you specify CASCADE.
You cannot specify this clause for a compressed table.
You cannot shrink a table that is the master table of an ON COMMIT materialized view. Rowid materialized views must be rebuilt after the shrink operation.
明白以下语句的作用：
ALTER TABLE
ALTER TABLE
ALTER TABLE
ALTER TABLE
ALTER TABLE
ALTER TABLE
ALTER INDEX IDXNAME SHRINK SPACE; -->索引段的收缩，同表段
使用 show_space (来自tom)区分段管理为manual,还是auto
DBMS_SPACE.UNUSED_SPACE Returns information about unused space in an object (table, index, or cluster).
DBMS_SPACE.FREE_BLOCKS Returns information about free data blocks in an object (table, index, or cluster) whose segment free space is managed by free lists (segment space management is MANUAL).
DBMS_SPACE.SPACE_USAGE Returns information about free data blocks in an object (table, index, or cluster) whose segment space management is AUTO.
Total Blocks --总块数
Full Blocks --填满的块数
FS4 Blocks
FS3 Blocks
FS2 Blocks
从以上几个输出也可以看出是否需要对表进行收缩。
http://download.oracle.com/docs/cd/B19306_01/server.102/b14231/schema.htm#sthref2156
参考：
http://blog.csdn.net/robinson_0612/article/details/6630673