Say one has HA Prometheus set up with two instances collecting the same metrics, with both of them wired to push data up to Grafana. They’re also configured (via external_labels) so that the replicas are distinguishable by the __replica__ label. (This as prescribed by Grafana.) Now, finally, my questions:
I assume these are deduped on the Grafana side. In terms of the bill, is there still a 2x cost or does one only pay for the metrics Grafana ends up using after deduping?
How can I see the effects of using HA and the __replica__ label on the Grafana Cloud dashboards? (e.g. On the “Metrics Instance Stats” panel, what will this do to the “Metrics Ingestion Rate” or the “Discarded Metrics Samples” graphs?)
As in that doc you linked, if the __replica__ label is set to a unique value within the HA Prometheus cluster, Grafana Cloud will deduplicate when those metrics are ingested. You should not be billed for these metrics, and if you find anything unusual please open a ticket with the Support team for further review.
You can use the grafanacloud_instance_samples_discarded metric in Explore with the grafanacloud-usage data source to review reasons for discarded metrics or something like mimirtool. You can use the queries from the Discarded Metric Samples panel on the Billing & Usage dashboard located in your GrafanaCloud folder to quickly review in Explore by selecting the drop down arrow near the Panel title and select Explore.
