MGR周期性性能下降原因

MySQL Goup Replication周期性性能下降原因

MySQL版本:5.7.22
在我的压测中发现一个有意思的现象: 16-256并发write_only各个负载下qps波谷有重合

当时觉得很不理解.想不明白原因, 从图上看, 这个性能衰减还是很严重的

下面这篇文章也有体现
https://cloud.tencent.com/developer/article/1054465

后来搜到这个bug [Performance drop every 60 seconds] https://bugs.mysql.com/bug.php?id=84774
这个bug是proxysql作者提的. 是我在看他的这篇博客发现的
Testing performance of MySQL Group Replication

官方回应是
感谢您评估Group Replication，您（以及所有社区反馈）非常重要！
与您的建议一样，Group Replication会定期执行维护，更确切地说，每60秒执行一次维护。
我们的表现结果显示，第2.4节。 随着时间的推移稳定
http://mysqlhighavailability.com/performance-evaluation-mysql-5-7-group-replication/
更确切地说，每个成员每隔60秒交换其持久化的事务集并且这些集合用于垃圾收集每个成员维护的认证信息。 在写密集型工作负载（如您的工作负载）上，此操作可能比预期的要长。
我们计划改善这一点。

Thank you for evaluating Group Replication, your (and all community feedback) is important!
Like you did suggest, Group Replication performs maintenance at a regular interval, more precisely each 60 seconds.
Our performance results show that, section 2.4. Stability over time at
http://mysqlhighavailability.com/performance-evaluation-mysql-5-7-group-replication/
More precisely, every 60 seconds every member exchange its persisted transactions set and the intersection of these sets is used to garbage collect the certification info that each member maintains. On write intensive workloads, like yours, this operation can be longer than expected.
We have plans to improve this.

在<<MySQL运维内参>>中也有描述

MGR目前个人感受从性能,可维护性,可诊断性都不如PXC好. 看以后吧~