PMM and large scale RDS deployments

Percona Monitoring and Management (PMM) PMM 2.x
1

We have hundreds of RDS MySQL, and Aurora DBs spread across many AWS accounts, regions, and Availability Zones. I have a few questions regarding this setup based on what I’ve read in the PMM docs:

1. The docs say that PMM Server must be deployed to the same AZ as the RDS instance. Our topology is designed for HA, and we do not always know which AZ the DB primary is hosted. AWS have high bandwidth connectivity between AZs, are you aware of any real cases where this has been problematic?

2. Service discovery: Is there any way currently to automatically add new instances to PMM without manual intervention? Perhaps via Cloudwatch events or something similar? Is there an API endpoint that we could target perhaps?

3. We have a fairly significant Prometheus setup with Grafana in front of it already. Is it possible to use that data source instead of the Prometheus that comes bundled in the server?

4. Can we use a different Clickhouse instance with QAN? and is it possible to break out the QAN components from the PMM server? If this is not possible, is there anything on the PMM roadmap to break these components apart to facilitate a hybrid approach?

Essentially, I’m trying to build a large scale monitoring platform for our global RDS estate but would prefer to use our own monitoring stack, with some of the good extras that come from PMM such as QAN. Any advice would be appreciated.

2
  1. The main question is the latency between PMM Server and the host it monitors.  If you’re not putting your PMM Server in  Asia while monitoring instance in Americas you should be OK.
    2) There is currently no built in service discovery but you surely can implement one yourself processing CloudWatch events and triggering API
    3) No you can’t because while you may have your own server and it may get similar or even same data what PMM captures it will be in different format (different labeling conventions etc) as such  PMM Dashboards would not work
    4) This is not supported out of the box but the power of  Open Source is of course it is possible :slight_smile:

    Thanks for your interest!
3

Thanks for the reply Peter. That all makes sense. I think we’ll take a look at how we can stand on the shoulders of the excellent tool you have already built. Appreciate you taking the time to respond. :slight_smile: