MySQL中dd::columns表結構轉table程序以及應用-有解無憂

GreatSQL社區原創內容未經授權不得隨意使用，轉載請聯系小編并注明來源，
GreatSQL是MySQL的國產分支版本，使用上與MySQL一致，

一、MySQL的dd表介紹

二、代碼跟蹤

三、知識應用

四、總結

一、MySQL的dd表介紹

MySQL的dd表是用來存放表結構和各種建表資訊的，客戶端建的表都存在mysql.table和mysql.columns表里，還有一個表mysql.column_type_elements比較特殊，用來存放SET和ENUM型別的欄位集合值資訊，看一下下面這張表的mysql.columns表和mysql.column_type_elements資訊，為了縮短顯示長度，這里只展示幾個重要的值，

#建表：
CREATE TABLE t1(id int  not null auto_increment primary key,col1 number,col2 VARCHAR(100),col3 pls_integer,
col4 enum('x','y') default 'x',col5 set('x1','y1'))  partition by hash(id) partitions 3;
SET SESSION debug='+d,skip_dd_table_access_check';
mysql> select name,ordinal_position,type,default_value_utf8,options,column_type_utf8 from mysql.columns where table_id=383;
+-------------+------------------+-----------------------+--------------------+-------------------+------------------+
| name        | ordinal_position | type                  | default_value_utf8 | options           | column_type_utf8 |
+-------------+------------------+-----------------------+--------------------+-------------------+------------------+
| col1        |                2 | MYSQL_TYPE_NEWDECIMAL | NULL               | interval_count=0; | decimal(65,0)    |
| col2        |                3 | MYSQL_TYPE_VARCHAR    | NULL               | interval_count=0; | varchar(100)     |
| col3        |                4 | MYSQL_TYPE_LONG       | NULL               | interval_count=0; | int              |
| col4        |                5 | MYSQL_TYPE_ENUM       | x                  | interval_count=2; | enum('x','y')    |
| col5        |                6 | MYSQL_TYPE_SET        | NULL               | interval_count=2; | set('x1','y1')   |
| DB_ROLL_PTR |                8 | MYSQL_TYPE_LONGLONG   | NULL               | NULL              |                  |
| DB_TRX_ID   |                7 | MYSQL_TYPE_INT24      | NULL               | NULL              |                  |
| id          |                1 | MYSQL_TYPE_LONG       | NULL               | interval_count=0; | int              |
+-------------+------------------+-----------------------+--------------------+-------------------+------------------+
8 rows in set (0.00 sec)

mysql.columns表說明如下：

ordinal_position是該欄位在表里的偏移量，這里多了3個欄位，DB_ROLL_PTR、DB_TRX_ID、id是用來執行undo的，記錄了欄位的版本資訊，

default_value_utf8是用來保存默認值的，options里面有interval_count用來保存集合型別的數值數的，columns表的options的key一共有如下幾種：

static const std::set<String_type> default_valid_option_keys = {
    "column_format", "geom_type",         "interval_count", "not_secondary",
    "storage",       "treat_bit_as_char", "zip_dict_id",    "is_array"};

mysql>  select * from mysql.column_type_elements where column_id=4286;
+-----------+---------------+------+
| column_id | element_index | name |
+-----------+---------------+------+
|      4286 |             1 | x    |
|      4286 |             2 | y    |
+-----------+---------------+------+
2 rows in set (0.01 sec)
#這里的column_id=4286是col4的id值，x和y分別對應了set定義時候的2個集合值，

二、代碼跟蹤

現在重新啟動資料庫，跟蹤一下這個columns表怎么轉為代碼里面的TABLE的field物件，首先找到表的dd資訊然后打開表獲取field資訊，

mysql> select * from t1;

輸入該命令后找到columns表轉為field的代碼：

#0  fill_column_from_dd (
    thd=0x555558b35a06 <std::char_traits<char>::compare(char const*, char const*, unsigned long)+61>, 
    share=0x7fffe83f1b60, 
    col_obj=0x555558bb0a5e <std::__cxx11::basic_string<char, std::char_traits<char>, Stateless_allocator<char, dd::String_type_alloc, My_free_functor> >::compare(std::__cxx11::basic_string<char, std::char_traits<char>, Stateless_allocator<char, dd::String_type_alloc, My_free_functor> > const&) const+142>, 
    null_pos=0x7fffe83f1880 "\005", null_bit_pos=32767, rec_pos=0x7fff2c05ac10 "explicit_encryption", 
    field_nr=0) at /mysql/sql/dd_table_share.cc:955
#1  0x00005555593c4c17 in fill_columns_from_dd (thd=0x7fff2c006890, share=0x7fff2cbf19e8, 
    tab_obj=0x7fff2cbb9b38) at /mysql/sql/dd_table_share.cc:1235
#2  0x00005555593c9e54 in open_table_def (thd=0x7fff2c006890, share=0x7fff2cbf19e8, table_def=...)
    at /mysql/sql/dd_table_share.cc:2408
#3  0x0000555558e76a13 in get_table_share (thd=0x7fff2c006890, db=0x7fff2cbeeff0 "db1", 
    table_name=0x7fff2cc03210 "t1", key=0x7fff2cbeed87 "db1", key_length=7, open_view=true, 
    open_secondary=false) at /mysql/sql/sql_base.cc:801
#4  0x0000555558e76f08 in get_table_share_with_discover (thd=0x7fff2c006890, table_list=0x7fff2cbee9b8, 
    key=0x7fff2cbeed87 "db1", key_length=7, open_secondary=false, error=0x7fffe83f1ea4)
    at /mysql/sql/sql_base.cc:889
#5  0x0000555558e7cd34 in open_table (thd=0x7fff2c006890, table_list=0x7fff2cbee9b8, 
    ot_ctx=0x7fffe83f2380) at /mysql/sql/sql_base.cc:3230
#6  0x0000555558e81769 in open_and_process_table (thd=0x7fff2c006890, lex=0x7fff2c01bdf0, 
    tables=0x7fff2cbee9b8, counter=0x7fff2c01be48, prelocking_strategy=0x7fffe83f2408, 
    has_prelocking_list=false, ot_ctx=0x7fffe83f2380)
    at /mysql/sql/sql_base.cc:5118
#7  0x0000555558e833bd in open_tables (thd=0x7fff2c006890, start=0x7fffe83f23f0, counter=0x7fff2c01be48, 
    flags=0, prelocking_strategy=0x7fffe83f2408)
    at /mysql/sql/sql_base.cc:5928
#8  0x0000555558e85626 in open_tables_for_query (thd=0x7fff2c006890, tables=0x7fff2cbee9b8, flags=0)
    at /mysql/sql/sql_base.cc:6904
#9  0x0000555559075720 in Sql_cmd_dml::prepare (this=0x7fff2cbef400, thd=0x7fff2c006890)
    at /mysql/sql/sql_select.cc:372
#10 0x00005555590760bf in Sql_cmd_dml::execute (this=0x7fff2cbef400, thd=0x7fff2c006890)
    at /mysql/sql/sql_select.cc:527
#11 0x0000555558fedc8e in mysql_execute_command (thd=0x7fff2c006890, first_level=true)
    at /mysql/sql/sql_parse.cc:4794
#12 0x0000555558fefe25 in dispatch_sql_command (thd=0x7fff2c006890, parser_state=0x7fffe83f3990, 
    update_userstat=false) at /mysql/sql/sql_parse.cc:5399
#13 0x0000555558fe52d3 in dispatch_command (thd=0x7fff2c006890, com_data=https://www.cnblogs.com/greatsql/archive/2022/10/12/0x7fffe83f4b70, 
    command=COM_QUERY) at /mysql/sql/sql_parse.cc:2000
#14 0x0000555558fe3643 in do_command (thd=0x7fff2c006890)
    at /mysql/sql/sql_parse.cc:1448
#15 0x000055555920e200 in handle_connection (arg=0x555560a65790)
    at /mysql/sql/conn_handler/connection_handler_per_thread.cc:307
#16 0x000055555ae36375 in pfs_spawn_thread (arg=0x5555608a2e20)
    at /mysql/storage/perfschema/pfs.cc:2899
#17 0x00007ffff77a6609 in start_thread (arg=) at pthread_create.c:477
#18 0x00007ffff76cb163 in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95

#fill_column_from_dd函式里面最重要的是make_field函式，把欄位從dd::Column轉為table的field然后賦值給TABLE_SHARE，
  reg_field = make_field(*col_obj, charset, share, rec_pos, null_pos, null_bit_pos);

三、知識應用

session每次獲取表的資訊都是在第一次打開表的時候就做好了，下次如果表沒有變化就從Table_cache直接獲取表資訊，現在假設我們要改col4欄位的集合值又不想通過alter陳述句來執行，那就可以直接對dd表進行操作，注意，該操作對生產環境有很大風險，這里只用來進行知識實踐，不能用來在生產環境實際操作，

把col4的x,y值改為a,b：
首先試著插入col4=x的記錄，現在還沒改dd表插入成功，

mysql> insert into t1 values(1,1,'aa',1,'x','x1');
Query OK, 1 row affected (0.03 sec)

接著開始改col4的集合值：

mysql> SET SESSION debug='+d,skip_dd_table_access_check';
Query OK, 0 rows affected (0.02 sec)
mysql> update mysql.columns set default_value_utf8='a' ,column_type_utf8='enum(\'a\',\'b\'))' where table_id=383 and name='col4';
Query OK, 1 row affected (0.03 sec)
Rows matched: 1  Changed: 1  Warnings: 0

mysql> select name,ordinal_position,type,default_value_utf8,options,column_type_utf8 from mysql.columns where table_id=383;
+-------------+------------------+-----------------------+--------------------+-------------------+------------------+
| name        | ordinal_position | type                  | default_value_utf8 | options           | column_type_utf8 |
+-------------+------------------+-----------------------+--------------------+-------------------+------------------+
| col1        |                2 | MYSQL_TYPE_NEWDECIMAL | NULL               | interval_count=0; | decimal(65,0)    |
| col2        |                3 | MYSQL_TYPE_VARCHAR    | NULL               | interval_count=0; | varchar(100)     |
| col3        |                4 | MYSQL_TYPE_LONG       | NULL               | interval_count=0; | int              |
| col4        |                5 | MYSQL_TYPE_ENUM       | a                  | interval_count=2; | enum('a','b'))   |集合值已改
| col5        |                6 | MYSQL_TYPE_SET        | NULL               | interval_count=2; | set('x1','y1')   |
| DB_ROLL_PTR |                8 | MYSQL_TYPE_LONGLONG   | NULL               | NULL              |                  |
| DB_TRX_ID   |                7 | MYSQL_TYPE_INT24      | NULL               | NULL              |                  |
| id          |                1 | MYSQL_TYPE_LONG       | NULL               | interval_count=0; | int              |
+-------------+------------------+-----------------------+--------------------+-------------------+------------------+
8 rows in set (0.00 sec)

mysql> update mysql.column_type_elements set name='a' where column_id=4286 and element_index=1;
Query OK, 1 row affected (0.03 sec)
Rows matched: 1  Changed: 1  Warnings: 0

mysql> update mysql.column_type_elements set name='b' where column_id=4286 and element_index=2;
Query OK, 1 row affected (0.03 sec)
Rows matched: 1  Changed: 1  Warnings: 0

mysql>  select * from mysql.column_type_elements where column_id=4286;
+-----------+---------------+------+
| column_id | element_index | name |
+-----------+---------------+------+
|      4286 |             1 | a    |集合值已改
|      4286 |             2 | b    |集合值已改
+-----------+---------------+------+
2 rows in set (0.00 sec)

現在再插入一條col4=x的記錄發現還是成功的，這是因為t1沒有重新從dd表轉為TABLE資訊，需要重啟后再試，

mysql> insert into t1 values(2,1,'aa',1,'x','x1');
Query OK, 1 row affected (0.02 sec)

重啟資料庫，然后登錄，再次插入col4=x發現報錯了，因為這時候的TABLE資訊是重新從dd表轉化的，

mysql> insert into t1 values(2,1,'aa',1,'x','x1');
ERROR 1265 (01000): Data truncated for column 'col4' at row 1

插入col4=a的記錄成功，說明更改成功，

mysql> insert into t1 values(3,1,'aa',1,'a','x1');
Query OK, 1 row affected (0.02 sec)

查看建表資訊,發現已經成功更改，

mysql> show create table t1;
+-------+-------------------------+
| Table | Create Table     |
+-------+-------------------------+
| t1    | CREATE TABLE `t1` (
  `id` int NOT NULL AUTO_INCREMENT,
  `col1` decimal(65,0) DEFAULT NULL,
  `col2` varchar(100) DEFAULT NULL,
  `col3` int DEFAULT NULL,
  `col4` enum('a','b') DEFAULT 'a',更改成功
  `col5` set('x1','y1') DEFAULT NULL,
  PRIMARY KEY (`id`)
) ENGINE=InnoDB AUTO_INCREMENT=4 DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci
/*!50100 PARTITION BY HASH (`id`)
PARTITIONS 3 */ 
+-------+-------------------------+

四、總結

實際上更改表結構如果通過alter命令來改流程跟上面也是一樣的，也是通過更新dd表來實作表結構的變更，這里只是從更底層來介紹，以上的操作在實際生產中不能直接操作，風險很大，會影響現有的記錄和相關的功能，這里只是作為一個案例來更好的說明dd的作業流程，幫助大家遇到問題知道怎么從底層排查，

Enjoy GreatSQL ??

關于 GreatSQL

GreatSQL是由萬里資料庫維護的MySQL分支，專注于提升MGR可靠性及性能，支持InnoDB并行查詢特性，是適用于金融級應用的MySQL分支版本，

相關鏈接： GreatSQL社區 Gitee GitHub Bilibili

GreatSQL社區：

歡迎來GreatSQL社區發帖提問
https://greatsql.cn/

GreatSQL社區

技術交流群：

微信：掃碼添加GreatSQL社區助手微信好友，發送驗證資訊加群，

轉載請註明出處，本文鏈接：https://www.uj5u.com/shujuku/514134.html

標籤：其他

上一篇：9個SQL運維常遇到的問題

下一篇：Mysql雙機主從搭建