We should monitor the "upness" (according to the up metric from Prometheus) of various prometheus exporters we have deployed now. The metric is exported automatically by Prometheus and set to 0 whenever Prometheus is unable to scrape metrics from the given exporter.
In addition to the up metrics, exporters often export the status of the underlying daemon as <daemon>_up and we should monitor that too.
Dashboard linked to the alerts: https://grafana.wikimedia.org/d/NEJu05xZz/prometheus-targets