上一篇博客中提到，当我们用delete删除数据时，因为无法降低高水位，对表做全表扫描时会带来数据库性能问题，而truncate可以大大改善这一状况，下面来看测试
SQL> conn zlm/zlm
Connected.
--创建表之前先记录下剩余表空间大小
SQL> select * from dba_free_space where tablespace_name='ZLM';
TABLESPACE_NAME FILE_ID BLOCK_ID BYTES BLOCKS RELATIVE_FNO
------------------------------ ---------- ---------- ---------- ---------- ------------
ZLM 6 128 51380224 6272 6
--创建一个100W行的大表
SQL> create table test1(int number);
Table created.
SQL> analyze table test1 compute statistics;
Table analyzed.
SQL> select table_name,num_rows,blocks,empty_blocks,avg_row_len,last_analyzed from dba_tables where table_name like 'TEST%';
TABLE_NAME NUM_ROWS BLOCKS EMPTY_BLOCKS AVG_ROW_LEN LAST_ANALYZED
------------------------------ ---------- ---------- ------------ ----------- ------------------
TEST1 0 0 0 0 27-SEP-14
注意，创建完表结构而未插入数据时，表的高水位是0，并没有为该表分配任何数据块
SQL> declare
2 i number := 0;
3 begin
4 for i in 1 .. 1000000 loop
5 insert into test1 values (i);
6 end loop;
7 commit;
8 end;
9 /
PL/SQL procedure successfully completed.
SQL> select count(*) from test1;
COUNT(*)
----------
1000000
--查看高水位状况
SQL> select header_file,header_block,bytes,blocks,extents from dba_segments where segment_name like 'TEST%';
HEADER_FILE HEADER_BLOCK BYTES BLOCKS EXTENTS
----------- ------------ ---------- ---------- ----------
6 130 13631488 1664 28
SQL> select block_id,extent_id,bytes,blocks from dba_extents where segment_name like 'TEST%';
BLOCK_ID EXTENT_ID BYTES BLOCKS
---------- ---------- ---------- ----------
128 0 65536 8
136 1 65536 8
144 2 65536 8
152 3 65536 8
160 4 65536 8
168 5 65536 8
176 6 65536 8
184 7 65536 8
192 8 65536 8
200 9 65536 8
208 10 65536 8
216 11 65536 8
224 12 65536 8
232 13 65536 8
240 14 65536 8
248 15 65536 8
256 16 1048576 128
384 17 1048576 128
512 18 1048576 128
640 19 1048576 128
768 20 1048576 128
896 21 1048576 128
1024 22 1048576 128
1152 23 1048576 128
1280 24 1048576 128
1408 25 1048576 128
1536 26 1048576 128
1664 27 1048576 128
SQL> select * from dba_free_space where tablespace_name='ZLM';
TABLESPACE_NAME FILE_ID BLOCK_ID BYTES BLOCKS RELATIVE_FNO
------------------------------ ---------- ---------- ---------- ---------- ------------
ZLM 6 1792 37748736 4608 6
SQL> set autot trace
SQL> select count(*) from test1;
Execution Plan
----------------------------------------------------------
Plan hash value: 3896847026
--------------------------------------------------------------------
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
--------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 2 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | | |
| 2 | TABLE ACCESS FULL| TEST1 | 1 | 2 (0)| 00:00:01 |
--------------------------------------------------------------------
Statistics
----------------------------------------------------------
0 recursive calls
1 db block gets
1599 consistent gets
0 physical reads
0 redo size
526 bytes sent via SQL*Net to client
519 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
1 rows processed
SQL> delete from test1;
1000000 rows deleted.
Execution Plan
----------------------------------------------------------
Plan hash value: 2642947686
--------------------------------------------------------------------
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
--------------------------------------------------------------------
| 0 | DELETE STATEMENT | | 1 | 2 (0)| 00:00:01 |
| 1 | DELETE | TEST1 | | | |
| 2 | TABLE ACCESS FULL| TEST1 | 1 | 2 (0)| 00:00:01 |
--------------------------------------------------------------------
Statistics
----------------------------------------------------------
814 recursive calls
1038983 db block gets
1953 consistent gets
8 physical reads
245334988 redo size
847 bytes sent via SQL*Net to client
769 bytes received via SQL*Net from client
3 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
1000000 rows processed
用delete删除100W行数据的大表TEST1时，产生了大量的redo（2亿多），另外还有很多一致性读（1953），读取了100多W个数据块，814次递归调用，可以看到，用delete删除表记录，对数据库的性能消耗是很大的，尤其是当delete大量行的时候
SQL> set autot off
SQL> select count(*) from test1;
COUNT(*)
----------
0
SQL> select header_file,header_block,bytes,blocks,extents from dba_segments where segment_name like 'TEST%';
HEADER_FILE HEADER_BLOCK BYTES BLOCKS EXTENTS
----------- ------------ ---------- ---------- ----------
6 130 13631488 1664 28
--用delete删除数据后查看执行计划
SQL> set autot trace
SQL> select count(*) from test1;
Execution Plan
----------------------------------------------------------
Plan hash value: 3896847026
--------------------------------------------------------------------
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
--------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 2 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | | |
| 2 | TABLE ACCESS FULL| TEST1 | 1 | 2 (0)| 00:00:01 |
--------------------------------------------------------------------
Statistics
----------------------------------------------------------
0 recursive calls
1 db block gets
1599 consistent gets
0 physical reads
0 redo size
525 bytes sent via SQL*Net to client
519 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
1 rows processed
此时虽然表中已经没有一条记录了，但是由于TEST1表并没有索引，走的是全表扫描，全表扫描就是要从hearder_block的值一致扫描到blocks得值，即从block 130一致到block 1664，可以看到，即使去访问一个空表，也要消耗1599次逻辑读，这就是高水位没有下降的缘故，之前说过，delete操作并不会降低表的高水位，带来的副作用就是访问该表时带来的性能下降（产生大量逻辑读）
--truncate表后再次查看统计信息
SQL> truncate table test1;
Table truncated.
SQL> set autot trace
SQL> select count(*) from test1;
Execution Plan
----------------------------------------------------------
Plan hash value: 3896847026
--------------------------------------------------------------------
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
--------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 2 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | | |
| 2 | TABLE ACCESS FULL| TEST1 | 1 | 2 (0)| 00:00:01 |
--------------------------------------------------------------------
Statistics
----------------------------------------------------------
2 recursive calls
1 db block gets
7 consistent gets
0 physical reads
96 redo size
525 bytes sent via SQL*Net to client
519 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
1 rows processed
此时再去全表扫描TEST1表，只有7次一致性读了，大大降低了需要扫描的数据库块，只有1个，
--查看此时TEST1表的表信息
SQL> select table_name,num_rows,blocks,empty_blocks,avg_row_len,last_analyzed from dba_tables where table_name like 'TEST%';
TABLE_NAME NUM_ROWS BLOCKS EMPTY_BLOCKS AVG_ROW_LEN LAST_ANALYZED
------------------------------ ---------- ---------- ------------ ----------- ------------------
TEST1 0 0 0 0 27-SEP-14
--似乎不太准确，重新收集一下统计信息
SQL> analyze table test1 compute statistics;
Table analyzed.
--再次查看
SQL> select table_name,num_rows,blocks,empty_blocks,avg_row_len,last_analyzed from dba_tables where table_name like 'TEST%';
TABLE_NAME NUM_ROWS BLOCKS EMPTY_BLOCKS AVG_ROW_LEN LAST_ANALYZED
------------------------------ ---------- ---------- ------------ ----------- ------------------
TEST1 0 0 8 0 27-SEP-14
可以看到，truncate已经把高水位降低到8了，而且是8个空块（EMPTY_BLOCKS），表示高水位以下未使用的空间，即第一个EXTENT为TEST1表分配的BLOCK空间，并且是最低值了，这与我们刚才是创建的只有表结构的空表不一样，同样是没有行数据的表，刚才创建表结构时的高水位为0。因此可以这么说，truncate降低高水位的作用也是有限的，还是剩余了1个extent的blocks的高水位，并没有完全消除。如果默认1个extent的block要大于8，那么高水位也要超过8
SQL> select header_file,header_block,bytes,blocks,extents from dba_segments where segment_name like 'TEST%';
HEADER_FILE HEADER_BLOCK BYTES BLOCKS EXTENTS
----------- ------------ ---------- ---------- ----------
6 130 65536 8 1
dba_segment视图也可以反映这一高水位的情况，如果对表做全表扫描，就是从130开始，扫描8个数据块，而并非之前的从130扫描到1664了
--查看剩余表空间容量
SQL> select * from dba_free_space where tablespace_name='ZLM';
TABLESPACE_NAME FILE_ID BLOCK_ID BYTES BLOCKS RELATIVE_FNO
------------------------------ ---------- ---------- ---------- ---------- ------------
ZLM 6 136 51314688 6264 6
truncate表数据后，ZLM剩余表空间又变大了，但是要注意，又多了8个block的消耗，建表之前查看的BLOCK_ID值是128
总结：
delete删除数据会读取大量的数据块，并产生大量的redo，对数据库产生性能影响，尤其是对大表操作时，删除大量的行数据时，而truncate虽然可以有效降低高水位，但其也有一定的局限性：
1. 并不能完全消除空表的高水位，仍然会有一定的空间浪费（8K）
2. 无法用于非清空全表数据的删除的场景
2498
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
