Telemetry: Instrument Memory NSM
If/when we support Prometheus as a data source (#217666 (closed)) we can begin to gather our key metrics from self-managed instances.
When ready, we should instrument our north star metric as defined (https://gitlab.com/gitlab-data/analytics/-/issues/4435).
Proposal
Query: Requests / Hour (Rolling 7 day average) / ( (GB of RAM * 0.004237) + (Cores * 0.031611) )
Some unknowns:
- We will need to identify whether we need recording rules for throughput, or whether we can have Prometheus calculate that on the fly without too much impact.
- Exactly which metric name to use. We could start with NGINX since that is what we use for single-node, and then broaden support for HA Proxy in the event someone has implemented that instead.
Availability & Testing
Testing notes from @grantyoung on the Epic should cover this issue - &3209
Edited by Grant Young