I think the main points that nostr:nprofile1qqsdulkdrc5hdf4dktl6taxmsxnasykghdnf32sqmnc7w6km2hhav3gppemhxue69uhkummn9ekx7mp0qywhwumn8ghj7mn0wd68ytfsxyhxymmvwshx7cnnv4e8vetj9uq3uamnwvaz7tmwdaehgu3dwp6kytnhv4kxcmmjv3jhytnwv46z7qmml2f somewhat missed is that there is no SQL for Nostr relays and there is, currently, no easy replication scheme to support load balancing, especially for geo-distributed semi-mirrored servers.  

Relays ARE databases in the sense they store stuff, but if you need advanced querying capability OR if your events are too big (binary or text) or has too many tags (too many indexes), the relay operator might not be able to support it and might either reject it or become super slow.

The lack of a query language is a limit we are forcing on ourselves. The rest is pretty much solvable with the usual enterprise stack without changing the protocol.