如何解决由触发器导致mysql内存溢出问题

1、问题现象

一台从库服务器的内存使用率持续上升，最终导致 MySQL 服务被 kill 了。

内存监控视图如下：

从图中可以看出，在 00:00 左右触发了 kill，然后又被mysqld_safe进程拉起，然后内存又会持续上升。

2、排查过程

基本信息

• 数据库版本：MySQL 5.7.32
• 操作系统版本：Ubuntu 20.04
• 主机配置：8C64GB
• innodb_buffer_pool_size：8G

由于用户环境未打开内存相关的监控，所以在 my.cnf 配置文件中配置如下：

performance-schema-instrument = 'memory/% = COUNTED'

打开内存监控等待运行一段时间后，相关视图查询如下：

从上述截图可以看到，MySQL 的 buffer pool 大小分配正常，但是 memory/sql/sp_head::main_mem_root 占用了 8GB 内存。

查看源代码的介绍：

sp_head：sp_head represents one instance of a stored program.It might be of any type (stored procedure, function, trigger, event).

根据源码的描述可知，sp_head表示一个存储程序的实例，该实例可能是存储过程、函数、触发器或者定时任务。

查询当前环境存储过程与触发器数量：

当前环境存在大量的触发器与存储过程。

查询 MySQL 相关bug，这里面提到一句话：

Tried to tweak table_open_cache_instances to affect this?

查询此参数描述：

A value of 8 or 16 is recommended on systems that routinely use 16 or more cores. However, if you have many large triggers on your tables that cause a high memory load, the default setting for table_open_cache_instances might lead to excessive memory usage. In that situation, it can be helpful to set table_open_cache_instances to 1 in order to restrict memory usage.

根据官方的解释可以了解到，如果有许多大的触发器，参数 table_open_cache_instances 的默认值可能会造成内存使用过多。

比如 table_open_cache_instances 设置为 16，那么表缓存会划分为 16 个 table instance。当并发访问大时，最多的情况下一个表的缓存信息会出现在每一个 table instance 里面。

再有每次将表信息放入表缓存时，所有关联的触发器都被放入 memory/sql/sp_head::main_mem_root 中，table_open_cache_instances 设置的越大其所占内存也就越大，以及存储过程也会消耗更多的内存，所以导致内存一直上升最终导致 OOM。

下面简单验证一下触发器对内存的影响。

当 table_open_cache_instances 为 8 时：

#清空缓存

mysql> flush tables;
Query OK, 0 rows affected (0.00 sec)

[root@test ~]# cat test.sh
for i in `seq 1 1 8`
do
mysql -uroot -p test -e "select * from test;"
done

[root@test ~]# sh test.sh

mysql> show variables like '%table_open_cache_instances%';
+----------------------------+-------+
| Variable_name | Value |
+----------------------------+-------+
| table_open_cache_instances | 8 |
+----------------------------+-------+
1 row in set (0.00 sec)

mysql> SELECT current_alloc FROM sys.memory_global_by_current_bytes WHERE event_name='memory/sql/sp_head::main_mem_root';
+---------------+
| current_alloc |
+---------------+
| 119.61 KiB |
+---------------+
1 row in set (0.00 sec)

在该表上创建一个触发器。

mysql> d|
mysql> CREATE TRIGGER trigger_test BEFORE INSERT ON test FOR EACH ROW BEGIN SIGNAL SQLSTATE '45000' SET message_text='Very long string. MySQL stores table descriptors in a special memory buffer, calle
'> at holds how many table descriptors MySQL should store in the cache and table_open_cache_instances t
'> hat stores the number of the table cache instances. So with default values of table_open_cache=4000
'> and table_open_cache_instances=16, you will have 16 independent memory buffers that will store 250 t
'> able descriptors each. These table cache instances could be accessed concurrently, allowing DML to u
'> se cached table descriptors without locking each other. If you use only tables, the table cache doe
'> s not require a lot of memory, because descriptors are lightweight, and even if you significantly increased the value of table_open_cache, it would not be so high. For example, 4000 tables will take u
'> p to 4000 x 4K = 16MB in the cache, 100.000 tables will take up to 390MB that is also not quite a hu
'> ge number for this number of open tables. However, if your tables have triggers, it changes the gam
'> e.'; END|
Query OK, 0 rows affected (0.00 sec)

#清空缓存

mysql> flush tables;
Query OK, 0 rows affected (0.00 sec)

然后访问表，查看缓存。

[root@test ~]# cat test.sh
for i in `seq 1 1 8`
do
mysql -uroot -p test -e "select * from test;"
done

[root@test ~]# sh test.sh

mysql> SELECT current_alloc FROM sys.memory_global_by_current_bytes WHERE event_name='memory/sql/sp_head::main_mem_root';
+---------------+
| current_alloc |
+---------------+
| 438.98 KiB |
+---------------+
1 row in set (0.00 sec)

可以发现 memory/sql/sp_head::main_mem_root 明显增长较大。如果有很多大的触发器，那么所占内存就不可忽视（现场环境触发器里面很多是调用了存储过程）。

当 table_open_cache_instances 为 1 时：

mysql> flush tables;
Query OK, 0 rows affected (0.00 sec)

mysql> show variables like '%table_open_cache_instances%';
+----------------------------+-------+
| Variable_name | Value |
+----------------------------+-------+
| table_open_cache_instances | 1 |
+----------------------------+-------+
1 row in set (0.00 sec)

SELECT current_alloc FROM sys.memory_global_by_current_bytes WHERE event_name='memory/sql/sp_head::main_mem_root';
+---------------+
| current_alloc |
+---------------+
| 119.61 KiB |
+---------------+
1 row in set (0.00 sec)

mysql> #访问表

mysql> system sh test.sh

mysql> SELECT current_alloc FROM sys.memory_global_by_current_bytes WHERE event_name='memory/sql/sp_head::main_mem_root';
+---------------+
| current_alloc |
+---------------+
| 159.53 KiB |
+---------------+
1 row in set (0.00 sec)

可以发现 memory/sql/sp_head::main_mem_root 所占内存增长较小。

由于大量触发器会导致表缓存和 memory/sql/sp_head::main_mem_root 占用更多的内存，根据实际环境，尝试把该从库的 table_open_cache_instances 修改为 1 后观察情况。

可以看到内存值趋于稳定，未再次出现内存使用率异常的问题。

3、总结