本文主要简单介绍下8.0.17新引入的功能multi-valued index, 顾名思义，索引上对于同一个Primary key, 可以建立多个二级索引项，实际上已经对array类型的基础功能做了支持 （感觉官方未来一定会推出类似pg的array 列类型）， 并基于array来构建二级索引，这意味着该二级索引的记录数可以是多于聚集索引记录数的，因而该索引不可以用于通常意义的查询，只能通过特定的接口函数来使用，下面的例子里会说明。

本文不对代码做深入了解，仅仅记录下相关的入口函数，便于以后工作遇到时能快速查阅。在最后附上了对应worklog的连接，感兴趣的朋友可以直接阅读worklog去了解他是如何实现的。

范例

摘录自官方文档

root@test 04:08:50>show create table customers\G *************************** 1. row *************************** Table: customersCreate Table: CREATE TABLE `customers` ( `id` bigint(20) NOT NULL AUTO_INCREMENT, `modified` datetime DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP, `custinfo` json DEFAULT NULL, PRIMARY KEY (`id`), KEY `zips` ((cast(json_extract(`custinfo`,_latin1'$.zip') as unsigned array)))) ENGINE=InnoDB AUTO_INCREMENT=6 DEFAULT CHARSET=latin11 row in set (0.00 sec)root@test 04:08:53>select * from customers;+----+---------------------+-------------------------------------------------------------------+| id | modified | custinfo |+----+---------------------+-------------------------------------------------------------------+| 1 | 2019-08-14 16:08:50 | {"user": "Jack", "user_id": 37, "zipcode": [94582, 94536]} || 2 | 2019-08-14 16:08:50 | {"user": "Jill", "user_id": 22, "zipcode": [94568, 94507, 94582]} || 3 | 2019-08-14 16:08:50 | {"user": "Bob", "user_id": 31, "zipcode": [94477, 94536]} || 4 | 2019-08-14 16:08:50 | {"user": "Mary", "user_id": 72, "zipcode": [94536]} || 5 | 2019-08-14 16:08:50 | {"user": "Ted", "user_id": 56, "zipcode": [94507, 94582]} |+----+---------------------+-------------------------------------------------------------------+5 rows in set (0.00 sec)

通过如下三个函数member of, json_contains, json_overlaps可以使用到该索引

root@test 04:09:00>SELECT * FROM customers WHERE 94507 MEMBER OF(custinfo->'$.zipcode');+----+---------------------+-------------------------------------------------------------------+| id | modified | custinfo |+----+---------------------+-------------------------------------------------------------------+| 2 | 2019-08-14 16:08:50 | {"user": "Jill", "user_id": 22, "zipcode": [94568, 94507, 94582]} || 5 | 2019-08-14 16:08:50 | {"user": "Ted", "user_id": 56, "zipcode": [94507, 94582]} |+----+---------------------+-------------------------------------------------------------------+2 rows in set (0.00 sec)root@test 04:09:41>SELECT * FROM customers WHERE JSON_CONTAINS(custinfo->'$.zipcode', CAST('[94507,94582]' AS JSON));+----+---------------------+-------------------------------------------------------------------+| id | modified | custinfo |+----+---------------------+-------------------------------------------------------------------+| 2 | 2019-08-14 16:08:50 | {"user": "Jill", "user_id": 22, "zipcode": [94568, 94507, 94582]} || 5 | 2019-08-14 16:08:50 | {"user": "Ted", "user_id": 56, "zipcode": [94507, 94582]} |+----+---------------------+-------------------------------------------------------------------+2 rows in set (0.00 sec)root@test 04:09:54>SELECT * FROM customers WHERE JSON_OVERLAPS(custinfo->'$.zipcode', CAST('[94507,94582]' AS JSON));+----+---------------------+-------------------------------------------------------------------+| id | modified | custinfo |+----+---------------------+-------------------------------------------------------------------+| 1 | 2019-08-14 16:08:50 | {"user": "Jack", "user_id": 37, "zipcode": [94582, 94536]} || 2 | 2019-08-14 16:08:50 | {"user": "Jill", "user_id": 22, "zipcode": [94568, 94507, 94582]} || 5 | 2019-08-14 16:08:50 | {"user": "Ted", "user_id": 56, "zipcode": [94507, 94582]} |+----+---------------------+-------------------------------------------------------------------+3 rows in set (0.00 sec)

multi-value index是functional index的一种实现，列的定义是一个虚拟列，值是从json column上取出来的数组

数组上存在相同值的话，会只存储一个到索引上。支持的类型：DECIMAL, INTEGER, DATETIME,VARCHAR/CHAR。另外index上只能有一个multi-value column。

下面简单介绍下相关的接口函数

数组最大容量：

入口函数： ha_innobase::mv_key_capacity

插入记录:

入口函数 row_ins_sec_index_multi_value_entry

通过类Multi_value_entry_builder_insert来构建tuple, 然后调用正常的接口函数row_ins_sec_index_entry插入到二级索引中.

已经解析好，排序并去重的数据存储在结构struct multi_value_data , 指针在dfield_t::data中. multi_value_data结构也是multi-value具体值的内存表现

删除记录：

入口函数: row_upd_del_multi_sec_index_entry

基于类Multi_value_entry_builder_normal构建tuple, 并依次从索引中删除

更新记录

入口函数：row_upd_multi_sec_index_entry

由于可能不是所有的二级索引记录都需要更新，需要计算出diff，找出要更新的记录calc_row_difference --> innobase_get_multi_value_and_diff, 设置一个需要更新的bitmap

事务回滚

相关函数：

row_undo_ins_remove_multi_secrow_undo_mod_upd_del_multi_secrow_undo_mod_del_mark_multi_sec

回滚的时候通过trx_undo_rec_get_multi_value从undo log中获取multi-value column的值，通过接口Multi_value_logger::read来构建并存储到field data中

记录undo log

函数: trx_undo_store_multi_value

通过Multi_value_logger::log将multi-value的信息存储到Undo log中. 'Multi_value_logger'是一个辅助类，用于记录multi-value column的值以及如何读出来

purge 二级索引记录

入口函数:

row_purge_del_markrow_purge_upd_exist_or_extern_func |--> row_purge_remove_multi_sec_if_poss

本文为云栖社区原创内容，未经允许不得转载。