[datascienced]

Leaderless, eventually consistent DB, with DB running on same servers as app.

Could be good, you need to rethink all the distributed DB stuff.

Does a node deal with certain subset (e.g. certain customers) so you use customerid as a partition key.

CRDTs?

Once a node fills up to say x% capacity do you split that out (using consistent hashing to minimize effect of such spits).

You probably end up reinventing one of the distributed DBs once you make it production ready.

I think it is a lot of work unless you have simplifying requirements e.g. just give each customer a seperate VM and then use a mounted cloud storage to scale up space to some max like 10Tb then maybe cap it there. Write your db log to s3 glacier for a backup.

Second replica for failover.