Reservior sampling does have legitimate use in monitoring: to track real-time percentiles (e.g. over the last 15 minutes), without logging everything and re-calculating on each new data entry. I do think it is important and useful to know about it, but probably would not use it for the database sampling (as it was in the article).
