Where are we at with relay design? Is anybody working on the deployment that can serve 1 million users?
Maybe it's "easy" enough to just split the write and read paths as long as we're willing to accept some eventual consistency.
Discussion
#[2] how many users you at now on the damus relay?
Do we even want a million users on one relay?
In theory it seems a "simple" way to achieve arbitrary scale could be to partition the database of notes by pubkey and scale out horizontally. When a request for a set of pubkeys comes in we split up the request and forward to whatever nodes handle which keys.