We have a rds t3.small instance which we perform write actions on. We have 2 read replicas for this instance which we route to using weighted routing policy via Route53. The reads are fine for now but we are getting massive operations for the write (master) database instance. CPU Utilization is nearing baseline of 20% and connections keep increasing (we are using connection pools but the traffic is too much)

Any possibilities on how we can manage load for writes? Is it possible to launch another instance for the same rds mysql database?

CodePudding user response：

You can't distribute writes across multiple instances, they all have to go to the master instance. It sounds like you will soon need to increase the size of the writer instance. If downtime is a concern, you could add a new, larger instance as a read replica, and then promote it as the new writer instance.

CodePudding user response：

All RDS instance types suck for write performance. They all use remotely-attached EBS storage, and remote storage incurs a heavy performance penalty for I/O.

At my last job, I benchmarked RDS versus Aurora for MySQL, and also benchmarked our physical servers (non-cloud) and also tested MySQL installed manually on EC2 i3 instances. RDS had consistent poor performance by a wide margin.

The t3 instances will be even worse, because they use burstable performance. Their baseline performance assumes a very light load, and they can add a burst of extra performance but only for short periods.

If you have write performance issues, especially for an application that requires consistent high performance, then you should upgrade to a more powerful instance type such as M5 or R5.

I would move away from RDS. I think it's useful only for very light load, or for temporary use during testing or development. I would recommend using Aurora instead of RDS, but my first preference would be to operate MySQL myself on EC2 i3 instances.