To decrease resource usage of Prometheus, we'll continue to drop more metrics. Having said that, even when metrics aren't stored, Prometheus still needs to parse and process the metrics. We need to evaluate what's the fixed cost to get an idea of the lower bounds.

We could run a cloned instance of prometheus-k8s (same targets and rules) but dropping all metrics at collection time.

DoD

• Compare CPU & RAM usage of the cloned Prometheus with the regular Prometheus.