[extra_rice]

Sorry, I didn't know how to make the question a bit clearer. Basically, it's: how do you ensure that queries on very large, highly dynamic datasets return in acceptable amount of time (especially if clients call/poll it at regular short intervals)?

[sethammons]

indexes, caching (pass through, LRU, etc), query read replicas, sharding, pre-fetching, sampling, maybe look into columnar storage ... Hard to answer not knowing more specifics.

Something to always remember: if it is valuable, charge for it. If it is really valuable, you can spend all kinds of hardware on it. Give each customer their dedicated instance and rinse and repeat the strategies above.