-
@ Mike Dilger
2023-11-09 05:20:37
A lot of terms have been bandied about regarding relay models: Gossip relay model, outbox relay model, and inbox relay model. But this term "relay model" bothers me. It sounds stuffy and formal and doesn't actually describe what we are talking about very well. Also, people have suggested maybe there are other relay models. So I thought maybe we should rethink this all from first principles. That is what this blog post attempts to do.
Nostr is notes and other stuff transmitted by relays. A client puts an event onto a relay, and subsequently another client reads that event. OK, strictly speaking it could be the same client. Strictly speaking it could even be that no other client reads the event, that the event was intended for the relay (think about nostr connect). But in general, the reason we put events on relays is for other clients to read them.
Given that fact, I see two ways this can occur:
1) The reader reads the event from the same relay that the writer wrote the event to (this I will call relay rendezvous),
2) The event was copied between relays by something.
This second solution is perfectly viable, but it less scalable and less immediate as it requires copies which means that resources will be consumed more rapidly than if we can come up with workable relay rendezvous solutions. That doesn't mean there aren't other considerations which could weigh heavily in favor of copying events. But I am not aware of them, so I will be discussing relay rendezvous.
We can then divide the relay rendezvous situation into several cases: one-to-one, one-to-many, and one-to-all, where the many are a known set, and the all are an unbounded unknown set. I cannot conceive of many-to-anything for nostr so we will speak no further of it.
For a rendezvous to take place, not only do the parties need to agree on a relay (or many relays), but there needs to be some way that readers can become aware that the writer has written something.
So the one-to-one situation works out well by the writer putting the message onto a relay that they know the reader checks for messages on. This we call the INBOX model. It is akin to sending them an email into their inbox where the reader checks for messages addressed to them.
The one-to-(known)-many model is very similar, except the writer has to write to many people's inboxes. Still we are dealing with the INBOX model.
The final case, one-to-(unknown)-all, there is no way the writer can place the message into every person's inbox because they are unknown. So in this case, the writer can write to their own OUTBOX, and anybody interested in these kinds of messages can subscribe to the writer's OUTBOX.
Notice that I have covered every case already, and that I have not even specified what particular types of scenarios call for one-to-one or one-to-many or one-to-all, but that every scenario must fit into one of those models.
So that is basically it. People need INBOX and OUTBOX relays and nothing else for relay rendezvous to cover all the possible scenarios.
That is not to say that other kinds of concerns might not modulate this. There is a suggestion for a DM relay (which is really an INBOX but with a special associated understanding), which is perfectly fine by me. But I don't think there are any other relay models. There is also the case of a live event where two parties are interacting over the same relay, but in terms of rendezvous this isn't a new case, it is just that the shared relay is serving as both parties' INBOX (in the case of a closed chat) and/or both parties' OUTBOX (in the case of an open one) at the same time.
So anyhow that's my thinking on the topic. It has become a fairly concise and complete set of concepts, and this makes me happy. Most things aren't this easy.