I'd recommend them switching from Graphite to Prometheus in order to resolve problems with high requirements for disk IO and storage space. Prometheus uses less disk IO and compresses time series data points better than Graphite - 12 bytes vs 4 bytes on typical data.
Prometheus' local storage doesn't scale, but this may be resolved with long-term remote storage systems from this list <a href="https://prometheus.io/docs/operating/integrations/#remote-endpoints-and-storage" rel="nofollow">https://prometheus.io/docs/operating/integrations/#remote-en...</a> . I'd recommend trying VictoriaMetrics, since it is the most cost-effective solution for storing huge amounts of time series data.