So we collected and aggregated more than 1 billion samples of metrics per second, which resulted in writing more than 30-40 million unique metric datapoints per second to storage. This resulted in more than 10 billion unique time series being stored (each with a very large number of distinct datapoints each).

This was 3.6 trillion metric samples per hour or 2.5 trillion metric datapoints stored a day (after aggregating samples).