MySQL MTS复制: hitting slave_pending_jobs_size_max

测试步骤：

从库停止复制：stop slave;

主库创建大表400万条记录。

开启从库复制：start slave;

监测从库error log持续输出：

2018-12-06T10:40:52.616289+08:00 4 [Note] Multi-threaded slave: Coordinator has waited 2431 times hitting slave_pending_jobs_size_max; current event size = 8207.
2018-12-06T10:40:52.647618+08:00 4 [Note] Multi-threaded slave: Coordinator has waited 2441 times hitting slave_pending_jobs_size_max; current event size = 8207.
2018-12-06T10:40:52.679589+08:00 4 [Note] Multi-threaded slave: Coordinator has waited 2451 times hitting slave_pending_jobs_size_max; current event size = 8207.
2018-12-06T10:40:52.711510+08:00 4 [Note] Multi-threaded slave: Coordinator has waited 2461 times hitting slave_pending_jobs_size_max; current event size = 8207.
2018-12-06T10:40:52.750250+08:00 4 [Note] Multi-threaded slave: Coordinator has waited 2471 times hitting slave_pending_jobs_size_max; current event size = 8207.
2018-12-06T10:40:52.785731+08:00 4 [Note] Multi-threaded slave: Coordinator has waited 2481 times hitting slave_pending_jobs_size_max; current event size = 8207.

搜索发现报错有以下两种情况
第一种
Last_Error: Cannot schedule event Rows_query, relay-log name ./db-s18-relay-bin.000448, position 419156572 to Worker thread because its size 18483519 exceeds 16777216 of slave_pending_jobs_size_max.
第二种
[Note] Multi-threaded slave: Coordinator has waited 701 times hitting slave_pending_jobs_size_max; current event size = 8167.

BUG地址：https://bugs.mysql.com/bug.php?id=68462
以上两种报错，初步判断问题可能在 slave_pending_jobs_size_max 的大小上，此值，官方默认是 16M，此值可以动态调整

slave-pending-jobs-size-max参数说明
在多线程复制时，在队列中Pending的事件所占用的最大内存，默认为16M，如果内存富余，或者延迟较大时，可以适当调大;注意这个值要比主库的max_allowed_packet大

slave-pending-jobs-size-max有如下几种情况：
1.- 如果event大小已经超过了等待任务大小的上限（配置slave-pending-jobs-size-max )，就报event太大的错，然后返回；
2.- 如果event大小+已经在等待的任务大小超过了slave-pending-jobs-size-max，就等待，至到等待队列变小；
3.- 如果当前的worker的队列满的话，也等待。
---------------------

 

检查slave_pending_jobs_size_max参数值为默认：

+-----------------------------+----------+
| Variable_name               | Value    |
+-----------------------------+----------+
| slave_pending_jobs_size_max | 16777216 |
+-----------------------------+----------+

 

调大该参数：

root@localhost:3306.sock [(none)]>set global slave_pending_jobs_size_max=16777216*8;
Query OK, 0 rows affected (0.02 sec)

root@localhost:3306.sock [(none)]>show variables like '%job%';
+-----------------------------+-----------+
| Variable_name               | Value     |
+-----------------------------+-----------+
| slave_pending_jobs_size_max | 134217728 |
+-----------------------------+-----------+
1 row in set (0.00 sec)

 

重新测试，告警消失。

 

WL#11348: Defaults: Increase Slave's Multi-Threaded Event Applier Buffer

看来MySQL 8.0的下一个版本会将默认值提高了。