The RHODS dashboard needs to display metrics from various workloads run by RHODS on behalf of the user:

• models
• ...

(the list is not exhaustive)

We need a single monitoring stack collecting from all those workloads in a single location, so that the Dashboard queries are kept relatively simple.

In that spirit, we need two things:

• Deploy as part of RHODS a metrics collector + storage, dedicated to these "user metrics" => that would be a Prometheus instance
•  Document how RHODS component should expose metrics for their workload type
• (Optionnal, bonus): define a standard set of metrics which each workload should expose to facilitate aggregation.