Postgresql 表和索引的预期存储大小

我正在Oracle Linux Server 6.3版上使用postgresql 9.2.1

我正在努力获取表和索引的预期存储大小

多亏了这个网站的一些建议，我制作了如下表格的公式

--如果是表格

postgres=# \d test
           Table "public.test"
    Column     |         Type          | Modifiers 
---------------+-----------------------+-----------
 c1            | integer               | not null
 c2            | character varying(20) | not null
 c3            | character varying(8)  | not null
 c4            | character varying(6)  | not null
 c5            | character varying(15) | 
 c6            | character varying(20) | 
 c7            | character varying(20) | 
 c8            | character varying(20) | 
Indexes:
    "idx_test" PRIMARY KEY, btree (c1, c3, c4, c5)
Tablespace: "test"

postgres=# insert into test values(1, 
                                   '11111111111111111111', -- 20(exactly same with max length of each column)
                                   '11111111',             -- 8
                                   '111111',               -- 6
                                   '111111111111111',      -- 15
                                   '11111111111111111111', -- 20
                                   '11111111111111111111', -- 20
                                   '11111111111111111111');-- 20
INSERT 0 1

postgres=# select * from pgstattuple('test');           
 table_len | tuple_count | tuple_len | tuple_percent | dead_tuple_count | dead_tuple_len | dead_tuple_percent | free_space | free_percent                          
-----------+-------------+-----------+---------------+------------------+----------------+--------------------+------------+--------------
      8192 |           1 |        81 |          0.99 |                0 |              0 |                  0 |       8072 |        98.54 

postgres=# insert into test values(2, 
                                   '11111111111111111111', -- 20(exactly same with max length of each column)
                                   '11111111',             -- 8
                                   '111111',               -- 6
                                   '111111111111111',      -- 15
                                   '11111111111111111111', -- 20
                                   '11111111111111111111', -- 20
                                   '11111111111111111111');-- 20      
INSERT 0 1

postgres=# select * from pgstattuple('test');
 table_len | tuple_count | tuple_len | tuple_percent | dead_tuple_count | dead_tuple_len | dead_tuple_percent | free_space | free_percent 
-----------+-------------+-----------+---------------+------------------+----------------+--------------------+------------+--------------
      8192 |           2 |       162 |          1.98 |                0 |              0 |                  0 |       7980 |        97.41                    

所以，我发现每个页面可以容纳88（X）个元组

• 实际存储中的元组大小：8072（第一次插入后的空闲空间）-7980（第二次插入后的空闲空间）=92
• 页面默认值=8192-8072（第一次插入后的可用空间）-92（元组大小）=28
• 9182-28=92*X（每页最大元组数）

如果是索引

postgres=# \d test_pkey
           Index "public.idx_test"
    Column     |         Type          |  Definition   
---------------+-----------------------+---------------
 c1            | integer               | c1
 c2            | character varying(20) | c2
 c3            | character varying(8)  | c3
 c4            | character varying(6)  | c4
primary key, btree, for table "public.test"

postgres=# truncate table test;
postgres=# vacuum;
postgres=# analyze;

postgres=# insert into test values(1, 
                                   '11111111111111111111', -- 20(exactly same with max length of each column)
                                   '11111111',             -- 8
                                   '111111',               -- 6
                                   '111111111111111',      -- 15
                                   '11111111111111111111', -- 20
                                   '11111111111111111111', -- 20
                                   '11111111111111111111');-- 20
INSERT 0 1

postgres=# select * from pgstattuple('idx_test');
 table_len | tuple_count | tuple_len | tuple_percent | dead_tuple_count | dead_tuple_len | dead_tuple_percent | free_space | free_percent 
-----------+-------------+-----------+---------------+------------------+----------------+--------------------+------------+--------------
     16384 |           1 |        56 |          0.34 |                0 |              0 |                  0 |       8088 |        49.37
(1 row)

postgres=# select * from pgstatindex('idx_test');
 version | tree_level | index_size | root_block_no | internal_pages | leaf_pages | empty_pages | deleted_pages | avg_leaf_density | leaf_fragmentation 
---------+------------+------------+---------------+----------------+------------+-------------+---------------+------------------+--------------------
       2 |          0 |       8192 |             1 |              0 |          1 |           0 |             0 |             0.79 |                  0
(1 row)

postgres=# insert into test values(1, 
                                   '11111111111111111111', -- 20(exactly same with max length of each column)
                                   '11111111',             -- 8
                                   '111111',               -- 6
                                   '111111111111111',      -- 15
                                   '11111111111111111111', -- 20
                                   '11111111111111111111', -- 20
                                   '11111111111111111111');-- 20

INSERT 0 1       

postgres=# select * from pgstattuple('idx_test');                                                                                                                                                                                                                                                   
 table_len | tuple_count | tuple_len | tuple_percent | dead_tuple_count | dead_tuple_len | dead_tuple_percent | free_space | free_percent
-----------+-------------+-----------+---------------+------------------+----------------+--------------------+------------+--------------             
     16384 |           2 |       112 |          0.68 |                0 |              0 |                  0 |       8028 |           49              
(1 row)                                                                                                                                                

postgres=# select * from pgstatindex('idx_test');                                                                                            
 version | tree_level | index_size | root_block_no | internal_pages | leaf_pages | empty_pages | deleted_pages | avg_leaf_density | leaf_fragmentation 
---------+------------+------------+---------------+----------------+------------+-------------+---------------+------------------+--------------------
       2 |          0 |       8192 |             1 |              0 |          1 |           0 |             0 |             1.52 |                  0 
(1 row)        

同样，我发现每个页面可以容纳135（Y）个元组

• 实际存储中的元组大小：8088（第一次插入后的空闲空间）-8028（第二次插入后的空闲空间）=60
• 页面默认值=8192-8088（第一次插入后的可用空间）-60（元组大小）=44
• 8192-44=60*Y（每页最大元组数）

当我在表中插入1350行时。。。。我有这个

postgres=# select * from pgstattuple('test');
 table_len | tuple_count | tuple_len | tuple_percent | dead_tuple_count | dead_tuple_len | dead_tuple_percent | free_space | free_percent 
-----------+-------------+-----------+---------------+------------------+----------------+--------------------+------------+--------------
    131072 |        1350 |    109350 |         83.43 |                0 |              0 |                  0 |       6424 |          4.9
(1 row)


postgres=# select * from pgstattuple('idx_test');
 table_len | tuple_count | tuple_len | tuple_percent | dead_tuple_count | dead_tuple_len | dead_tuple_percent | free_space | free_percent 
-----------+-------------+-----------+---------------+------------------+----------------+--------------------+------------+--------------
     90112 |        1350 |     54000 |         59.93 |                0 |              0 |                  0 |      13580 |        15.07
(1 row)

postgres=# select * from pgstatindex('idx_test');
 version | tree_level | index_size | root_block_no | internal_pages | leaf_pages | empty_pages | deleted_pages | avg_leaf_density | leaf_fragmentation 
---------+------------+------------+---------------+----------------+------------+-------------+---------------+------------------+--------------------
       2 |          1 |      81920 |             3 |              0 |          9 |           0 |             0 |            81.49 |                  0
(1 row)     

表的文件大小

1350（行数）/88（X）=15.34-->这意味着需要16页，这意味着文件大小为16*8192=131072。看起来不错

然而，指数sise是不同的

1350（行数）/135（Y）=10正好。。。文件大小需要10*8192，但需要90112

再插入一行，除此之外，索引大小应该扩展（如果我是对的），所以我尝试了，但没有更改

postgres=# insert into test values(27108,'sanjuk1052','20121022','233338','172,20,30,177','win7','IE','9,0');
INSERT 0 1
postgres=# select * from pgstattuple('test');
 table_len | tuple_count | tuple_len | tuple_percent | dead_tuple_count | dead_tuple_len | dead_tuple_percent | free_space | free_percent 
-----------+-------------+-----------+---------------+------------------+----------------+--------------------+------------+--------------
    131072 |        1351 |    109431 |         83.49 |                0 |              0 |                  0 |       6332 |         4.83
(1 row)

postgres=# select * from pgstattuple('idx_test');
 table_len | tuple_count | tuple_len | tuple_percent | dead_tuple_count | dead_tuple_len | dead_tuple_percent | free_space | free_percent 
-----------+-------------+-----------+---------------+------------------+----------------+--------------------+------------+--------------
     90112 |        1351 |     54040 |         59.97 |                0 |              0 |                  0 |      13536 |        15.02
(1 row)

postgres=# select * from pgstatindex('idx_test');
 version | tree_level | index_size | root_block_no | internal_pages | leaf_pages | empty_pages | deleted_pages | avg_leaf_density | leaf_fragmentation 
---------+------------+------------+---------------+----------------+------------+-------------+---------------+------------------+--------------------
       2 |          1 |      81920 |             3 |              0 |          9 |           0 |             0 |            81.55 |                  0
(1 row)

我甚至不能确定这种方法是否合理，但这项工作必须完成，即使不是完美的

我需要我自己的公式，特别是为了获得预期的索引大小，包括由TOAST重新筛选的附加文件

---

如果您有任何建议，我们将不胜感激。

您在这里遇到的真正问题之一是varchar没有使用恒定的空间量。随着时间的推移，这可能会使您的页面估计值偏离（您的估计值基本上是一个最大值，可能更小）。以你的桌子结构，我认为任何东西都不会被烤

这也会影响索引，因为您有可变长度的字段。因此，您可以预期您的估计值代表了一个最大的、未知的准确金额。实际大小将取决于实际数据，而不仅仅是模式和行数