How To Efficiently Drop A Table With Many Extents

文件 ID:
68836.1
型別:
BULLETIN

上次修訂日期:
12-JUN-2008
狀態:
PUBLISHED

How to efficiently drop a table with many extents

PURPOSE
~~~~~~~

    This note describes why a user process can consume large amounts of CPU
    after dropping a table consisting of many extents, and a potential
    workaround to stop the problem occurring. Essentially the CPU is being
    used to manipulate the extents i.e. moving used extents (uet$) to free
    extents (fet$). In certain circumstances it may be possible to regulate
    this CPU activity.

SCOPE & APPLICATION
~~~~~~~~~~~~~~~~~~~
This article is intended to assist DBAs who may need to drop a table
consisting of many extents.

RELATED DOCUMENTS
~~~~~~~~~~~~~~~~~
Note 61997.1 SMON - Temporary Segment Cleanup and Free Space Coalescing

Permanent object cleanup
~~~~~~~~~~~~~~~~~~~~~~~~

   If a permanent object (table) is made up of many extents, and the object is
   to be dropped, the user process dropping the object will consume large
   amounts of CPU - this is an inescapable fact. However, with some forethought
   it is possible to mitigate the effects of CPU usage (and hence the knock-on
   effect on other users of system resources) thus:

   1. Identify, but do NOT drop the table
   2. Truncate the table, specifying the REUSE STORAGE clause. This will be
      quick as extents are not deallocated; the highwater mark is simply
      adjusted to the segment header block.
   3. Deallocate unused extents from the table, SPECIFYING THE KEEP CLAUSE.
      This is the crux - you can control how many extents are to be deallocated
      by specifying how much (in terms of Kb or Mb) of the table is NOT
      to be deallocated.

   Example:
   o. Table BIGTAB is 2Gb in size and consists of 262144 8Kb extents
   o. There is little CPU power available, and (from past experience) it is
      known that dropping an object of this number of extents can take days
   o. The system is quiet at night times (no other users or batch jobs)

   In the above example the table could be dropped in 'phases' over the period
   of a few nights as follows:
   1. Truncate the table, specifying the REUSE STORAGE clause:
      SQL> TRUNCATE TABLE BIGTAB REUSE STORAGE;
   2. If it takes 3 days (72 hours) to drop the table, spread this out over
      6 nights i.e. drop 1/3 Gb per night. This can be achieved in 6 (nightly)
      steps as follows:
      Night 1:
        SQL> ALTER TABLE BIGTAB DEALLOCATE UNUSED KEEP 1707M; (2Gb*5/6)
      Night 2:
        SQL> ALTER TABLE BIGTAB DEALLOCATE UNUSED KEEP 1365M; (2Gb*4/6)
      Night 3:
        SQL> ALTER TABLE BIGTAB DEALLOCATE UNUSED KEEP 1024M; (2Gb*3/6)
      Night 4:
        SQL> ALTER TABLE BIGTAB DEALLOCATE UNUSED KEEP 683M; (2Gb*2/6)
      Night 5:
        SQL> ALTER TABLE BIGTAB DEALLOCATE UNUSED KEEP 341M; (2Gb*1/6)
      Night 6:
        SQL> DROP TABLE BIGTAB;

The same method can be applied if LOB segments or indexes are involved.

        SQL> ALTER TABLE MODIFY LOB ()
             DEALLOCATE UNUSED KEEP M;

        SQL> ALTER INDEX DEALLOCATE UNUSED KEEP M;

Caveats
~~~~~~~

   o. If you have inadvertently tried to drop the table, this method will
      not work. This is because the drop will first convert the segment to
      a temporary segment, and only then start cleaning up the now temporary
      segment's extents. Thus, if the drop is interrupted, the temporary
      segment will now be cleaned up by SMON.

   o. This method will only work for table, lob and index segment types.

   o. This method will not work for segments bigger than 4gb in size due to
      unpublished bug:
      1190939 4G

-------------------------------------------------------------------------------
Oracle Support Services

以下是自己測試過程:

C:\Documents and Settings\yibin>sqlplus /nolog

SQL*Plus: Release 9.2.0.8.0 - Production on 星期三 1月 14 21:43:37 2009

SQL> conn as sysdba
已連線。

SQL> create table YIBIN_truncate as select * from dba_objects;

表已建立。

SQL> insert into YIBIN_truncate select * from YIBIN_truncate;

已建立6342行。

SQL> /

已建立12684行。

SQL> /

已建立25368行。

SQL> /

已建立50736行。

SQL>
SQL> insert into YIBIN_truncate select * from YIBIN_truncate;

已建立3247104行。

SQL> commit;

提交完成。

SQL>
SQL> insert into YIBIN_truncate select * from YIBIN_truncate;

已建立6494208行。

SQL> commit;

提交完成。

SQL>
SQL> insert into YIBIN_truncate select * from YIBIN_truncate;

已建立12988416行。

SQL> commit;

提交完成。

SQL> select bytes/1024/1024 from dba_segments where segment_name='YIBIN_TRUNCATE'
;

BYTES/1024/1024
---------------
2503

SQL> select extents,blocks from dba_segments where segment_name='YIBIN_TRUNCATE';

EXTENTS BLOCKS
---------- ----------
223 320384

SQL> select bytes/1024/1024 from dba_segments where segment_name='YIBIN_TRUNCATE'
;

BYTES/1024/1024
---------------
2503

SQL> truncate table YIBIN_truncate reuse storage;

表已截掉。

SQL> select extents,blocks from dba_segments where segment_name='YIBIN_TRUNCATE';

EXTENTS BLOCKS
---------- ----------
223 320384

SQL> select bytes/1024/1024 from dba_segments where segment_name='YIBIN_TRUNCATE'
;

BYTES/1024/1024
---------------
2503

SQL> alter table YIBIN_truncate deallocate unused keep 1000m;

表已更改。

SQL> select extents,blocks from dba_segments where segment_name='YIBIN_TRUNCATE';

EXTENTS BLOCKS
---------- ----------
197 128008

SQL> alter table YIBIN_truncate deallocate unused keep 1000m;

表已更改。

SQL> alter table YIBIN_truncate deallocate unused keep 0m;

表已更改。

SQL> select extents,blocks from dba_segments where segment_name='YIBIN_TRUNCATE';

EXTENTS BLOCKS
---------- ----------
1 8

SQL> select bytes/1024/1024 from dba_segments where segment_name='YIBIN_TRUNCATE'
;

BYTES/1024/1024
---------------
.0625

透過測試
在truncate大表時使用reuse storage，暫時不釋放extent ,再使用unused keep 分批釋放空間，這樣能減少truncate時對系

統資源的爭用。

How To Efficiently Drop A Table With Many Extents

相關文章