PostgreSQL数据库B-Tree索引的物理存储结构是怎样的

这篇文章主要讲解了“PostgreSQL数据库B-Tree索引的物理存储结构是怎样的”，文中的讲解内容简单清晰，易于学习与理解，下面请大家跟着小编的思路慢慢深入，一起来研究和学习“PostgreSQL数据库B-Tree索引的物理存储结构是怎样的”吧！

创新互联服务项目包括辽阳网站建设、辽阳网站制作、辽阳网页制作以及辽阳网络营销策划等。多年来，我们专注于互联网行业，利用自身积累的技术优势、行业经验、深度合作伙伴关系等，向广大中小型企业、政府机构等提供互联网行业的解决方案，辽阳网站推广取得了明显的社会效益与经济效益。目前，我们服务的客户以成都为中心已经辐射到辽阳省份的部分城市，未来相信会继续扩大服务区域并继续获得客户的支持与信任！

一、测试数据

我们继续使用上一节使用的测试数据，这一次我们追加插入>1000行的数据。

-- 为方便对比，插入数据前先查看索引元数据页
testdb=# select * from bt_metap('pk_t_index');
 magic  | version | root | level | fastroot | fastlevel | oldest_xact | last_cleanup_num_tuples 
--------+---------+------+-------+----------+-----------+-------------+-------------------------
 340322 |       3 |    1 |     0 |        1 |         0 |           0 |                      -1
(1 row)

testdb=# do $$
testdb$# begin
testdb$#   for i in 19..1020 loop
testdb$#   insert into t_index (id, c1, c2) values (i, '#'||i||'#', '#'||i||'#');
testdb$#   end loop;
testdb$# end $$;
DO
testdb=# select count(*) from t_index;
 count 
-------
  1008
(1 row)

二、索引存储结构

插入数据后，重新查看索引元数据页信息：

testdb=# select * from bt_metap('pk_t_index');
 magic  | version | root | level | fastroot | fastlevel | oldest_xact | last_cleanup_num_tuples 
--------+---------+------+-------+----------+-----------+-------------+-------------------------
 340322 |       3 |    3 |     1 |        3 |         1 |           0 |                      -1
(1 row)

root block从原来的block 1变为block 3，查看block 3的的Special space：

testdb=# select * from bt_page_stats('pk_t_index',3);
 blkno | type | live_items | dead_items | avg_item_size | page_size | free_size | btpo_prev | btpo_next | btpo | btpo_flags 
-------+------+------------+------------+---------------+-----------+-----------+-----------+-----------+------+------------
     3 | r    |          3 |          0 |            13 |      8192 |      8096 |         0 |         0 |    1 |          2
(1 row)

type=r，表示root index block，这个block有3个index entries（live_items=3，该index block只是root block（btpo_flags=BTP_ROOT）。下面我们来看看这个block中的index entries：

testdb=# select * from bt_page_items('pk_t_index',3);
 itemoffset |  ctid   | itemlen | nulls | vars |          data           
------------+---------+---------+-------+------+-------------------------
          1 | (1,0)   |       8 | f     | f    | 
          2 | (2,53)  |      16 | f     | f    | 7b 01 00 00 00 00 00 00
          3 | (4,105) |      16 | f     | f    | e9 02 00 00 00 00 00 00
(3 rows)

root/branch index block存储的是指向其他index block的指针。第1行，index entries指向第1个index block，由于该block没有left block，因此，itemlen只有8个字节，数据范围为1-\x0000017b（十进制值为379）；第2行，index entries指向第2个index block，数据范围为380-\x000002e9（745）；第3行，index entries指向第4个index block，数据范围为大于745的值。
这里有个疑惑，正常来说，root index block中的entries应指向index block，但ctid的值（2,53）和（4,105）指向的却是Heap Table Block，PG11 Beta2的Bug？

In a B-tree leaf page, ctid points to a heap tuple. In an internal page, the block number part of ctid points to another page in the index itself, while the offset part (the second number) is ignored and is usually 1.

testdb=# select * from heap_page_items(get_raw_page('t_index',2)) where t_ctid = '(2,53)';
 lp | lp_off | lp_flags | lp_len | t_xmin  | t_xmax | t_field3 | t_ctid | t_infomask2 | t_infomask | t_hoff | t_bits | t_oid |                  t_data                  
----+--------+----------+--------+---------+--------+----------+--------+-------------+------------+--------+--------+-------+------------------------------------------
 53 |   5648 |        1 |     43 | 1612755 |      0 |      360 | (2,53) |           3 |       2306 |     24 |        |       | \x7b0100001323333739232020200d2333373923
(1 row)

testdb=# select * from heap_page_items(get_raw_page('t_index',4)) where t_ctid = '(4,105)';
 lp  | lp_off | lp_flags | lp_len | t_xmin  | t_xmax | t_field3 | t_ctid  | t_infomask2 | t_infomask | t_hoff | t_bits | t_oid |                  t_data                  
-----+--------+----------+--------+---------+--------+----------+---------+-------------+------------+--------+--------+-------+------------------------------------------
 105 |   3152 |        1 |     43 | 1612755 |      0 |      726 | (4,105) |           3 |       2306 |     24 |        |       | \xe90200001323373435232020200d2337343523
(1 row)

回到正题，我们首先看看index block 1的相关数据：

testdb=# select * from bt_page_stats('pk_t_index',1);
 blkno | type | live_items | dead_items | avg_item_size | page_size | free_size | btpo_prev | btpo_next | btpo | btpo_flags 
-------+------+------------+------------+---------------+-----------+-----------+-----------+-----------+------+------------
     1 | l    |        367 |          0 |            16 |      8192 |       808 |         0 |         2 |    0 |          1
(1 row)

testdb=# select * from bt_page_items('pk_t_index',1) limit 10;
 itemoffset |  ctid  | itemlen | nulls | vars |          data           
------------+--------+---------+-------+------+-------------------------
          1 | (2,53) |      16 | f     | f    | 7b 01 00 00 00 00 00 00
          2 | (0,1)  |      16 | f     | f    | 02 00 00 00 00 00 00 00
          3 | (0,2)  |      16 | f     | f    | 04 00 00 00 00 00 00 00
          4 | (0,3)  |      16 | f     | f    | 08 00 00 00 00 00 00 00
          5 | (0,4)  |      16 | f     | f    | 10 00 00 00 00 00 00 00
          6 | (0,6)  |      16 | f     | f    | 11 00 00 00 00 00 00 00
          7 | (0,5)  |      16 | f     | f    | 12 00 00 00 00 00 00 00
          8 | (0,8)  |      16 | f     | f    | 13 00 00 00 00 00 00 00
          9 | (0,9)  |      16 | f     | f    | 14 00 00 00 00 00 00 00
         10 | (0,10) |      16 | f     | f    | 15 00 00 00 00 00 00 00
(10 rows)

第1个block的Special space，其中type=l，表示leaf index block，btpo_flags=BTP_LEAF表示该block仅仅为leaf index block，block的index entries指向heap table。同时，这个block里面有367个items，右边兄弟block号是2（btpo_next）。
值得注意到，index entries的第1个条目，是最大值\x017b，第2个条目才是最小值，接下来的条目是按顺序存储的其他值。源码的README（src/backend/access/nbtree/README）里面有解释：

On a page that is not rightmost in its tree level, the "high key" is
kept in the page's first item, and real data items start at item 2.
The link portion of the "high key" item goes unused.  A page that is
rightmost has no "high key", so data items start with the first item.
Putting the high key at the left, rather than the right, may seem odd,
but it avoids moving the high key as we add data items.

官方文档也有相关解释：

Note that the first item on any non-rightmost page (any page with a non-zero value in the btpo_next field) is the page's “high key”, meaning its data serves as an upper bound on all items appearing on the page, while its ctid field is meaningless. Also, on non-leaf pages, the first real data item (the first item that is not a high key) is a “minus infinity” item, with no actual value in its data field. Such an item does have a valid downlink in its ctid field, however.

下面我们再来看看index block 2&4：

testdb=# select * from bt_page_stats('pk_t_index',2);
 blkno | type | live_items | dead_items | avg_item_size | page_size | free_size | btpo_prev | btpo_next | btpo | btpo_flags 
-------+------+------------+------------+---------------+-----------+-----------+-----------+-----------+------+------------
     2 | l    |        367 |          0 |            16 |      8192 |       808 |         1 |         4 |    0 |          1
(1 row)

testdb=# select * from bt_page_items('pk_t_index',2) limit 10;
 itemoffset |  ctid   | itemlen | nulls | vars |          data           
------------+---------+---------+-------+------+-------------------------
          1 | (4,105) |      16 | f     | f    | e9 02 00 00 00 00 00 00
          2 | (2,53)  |      16 | f     | f    | 7b 01 00 00 00 00 00 00
          3 | (2,54)  |      16 | f     | f    | 7c 01 00 00 00 00 00 00
          4 | (2,55)  |      16 | f     | f    | 7d 01 00 00 00 00 00 00
          5 | (2,56)  |      16 | f     | f    | 7e 01 00 00 00 00 00 00
          6 | (2,57)  |      16 | f     | f    | 7f 01 00 00 00 00 00 00
          7 | (2,58)  |      16 | f     | f    | 80 01 00 00 00 00 00 00
          8 | (2,59)  |      16 | f     | f    | 81 01 00 00 00 00 00 00
          9 | (2,60)  |      16 | f     | f    | 82 01 00 00 00 00 00 00
         10 | (2,61)  |      16 | f     | f    | 83 01 00 00 00 00 00 00
(10 rows)

testdb=# select * from bt_page_stats('pk_t_index',4);
 blkno | type | live_items | dead_items | avg_item_size | page_size | free_size | btpo_prev | btpo_next | btpo | btpo_flags 
-------+------+------------+------------+---------------+-----------+-----------+-----------+-----------+------+------------
     4 | l    |        276 |          0 |            16 |      8192 |      2628 |         2 |         0 |    0 |          1
(1 row)

testdb=# select * from bt_page_items('pk_t_index',4) limit 10;
 itemoffset |  ctid   | itemlen | nulls | vars |          data           
------------+---------+---------+-------+------+-------------------------
          1 | (4,105) |      16 | f     | f    | e9 02 00 00 00 00 00 00
          2 | (4,106) |      16 | f     | f    | ea 02 00 00 00 00 00 00
          3 | (4,107) |      16 | f     | f    | eb 02 00 00 00 00 00 00
          4 | (4,108) |      16 | f     | f    | ec 02 00 00 00 00 00 00
          5 | (4,109) |      16 | f     | f    | ed 02 00 00 00 00 00 00
          6 | (4,110) |      16 | f     | f    | ee 02 00 00 00 00 00 00
          7 | (4,111) |      16 | f     | f    | ef 02 00 00 00 00 00 00
          8 | (4,112) |      16 | f     | f    | f0 02 00 00 00 00 00 00
          9 | (4,113) |      16 | f     | f    | f1 02 00 00 00 00 00 00
         10 | (4,114) |      16 | f     | f    | f2 02 00 00 00 00 00 00
(10 rows)

感谢各位的阅读，以上就是“PostgreSQL数据库B-Tree索引的物理存储结构是怎样的”的内容了，经过本文的学习后，相信大家对PostgreSQL数据库B-Tree索引的物理存储结构是怎样的这一问题有了更深刻的体会，具体使用情况还需要大家实践验证。这里是创新互联，小编将为大家推送更多相关知识点的文章，欢迎关注！