This is, in my opinion, the most severe problem with the Fediverse.

@MrCheeze this is a feature… until server X goes down. But server X going down is exactly what we are afraid of

@mcc @MrCheeze I'm pretty sure there are people today running ActivityPub profiles on redundant servers or using distributed identities, but I have yet to understand how that works for post availability and retention, like how I would know where to go to read their stuff if the place I know them from is gone.

@julian @MrCheeze I don't think there's a systematic infrastructure for it. Bluesky on the other hand is built around a systematic architecture for that exact thing (although as I mentioned it has other decisions I don't like)

@mcc @MrCheeze Define systematic infrastructure? Mitra exists as a practical implementation compatible with Mastodon, and it implements https://codeberg.org/silverpill/feps/src/branch/main/c390/fep-c390.md and https://codeberg.org/silverpill/feps/src/branch/main/ae97/fep-ae97.md. So you can divorce your identity from your AP instance today if you like. I just don't fully understand what that gets us.

If you decide to take your Bluesky PDS down, does their relay keep serving cached copies of your posts forever? Is that a good thing, do we want that?

@mcc @whitequark would a service that subscribes to your posts and writes them out to an archive automatically (private, or public and indexable) be interesting for the permanence of posts concern? This seems relatively doable.

@julian @mcc @MrCheeze

>Mitra exists as a practical implementation compatible with Mastodon, and it implements...

The main specification is FEP-ef61, which describes how decentralized ("nomadic") identity works. FEP-c390 is an older idea and it doesn't provide full data portability. FEP-ae97 is about combining FEP-ef61 with ActivityPub Client-to-server API.

>I just don't fully understand what that gets us.

You can clone your account to multiple servers, and those multiple accounts would work as one! And more: peer to peer, local-first applications.
Mitra's implementation is actually incomplete, I recommend also checking out Streams, which supports FEP-ef61 and non-ActivityPub nomadic identity (Nomad protocol; this is where the "nomadic identity" comes from).

@silverpill Thanks, that helps me a bit. So I'd set up my client to send the posts I write to all of my multiple servers, and if one of them goes down I'm still reachable on the others with all my stuff? And the idea is that people who reply to me send their replies to all of my servers, or do my servers synchronize themselves on their own?

Maybe I just need to read up on this more...

@mcc @MrCheeze

@silverpill @mcc @julian @MrCheeze Sorry if this is not a right place to ask. I'm interested in FEP-ed61 but I'm stuck at Webfinger and gateways. What if a user only knows a webfinger handle [email protected] but a gateway example.com is dead? The Webfinger section of FEP-ef61 reads like you somehow know the actor's gateway before you look up the actor.

@julian I think servers should synchronize data between themselves, but the other approach where clients write to multiple servers (as they do in Nostr) is also fine

@mcc @MrCheeze

@mochi Unfortunately, handle is tied to a domain name. If primary gateway is dead, actor should update its gateways array, and thus obtain a new handle.

Actor ID ('ap' URL) is a permanent identifier that doesn't depend on a specific gateway.

@mcc @julian @MrCheeze

@onyxraven @mcc @whitequark it would certainly be interesting but it only solves part of the problem. URIs are an intrinsic part of posts and breaking URI resolution is for most intents and purposes the same as deleting the post. Yes archiving the content would be great, but conversation structure and discoverability would still be gone (unless you can browse the whole Fediverse via the archive like with archive.org).

@caesar @onyxraven @whitequark maybe urls shouldn't be an intrinsic part of posts then? Alternately if we assume all moves are one way we could imagine a directory of dead urls and what new url domains to map them to

@mcc @onyxraven @whitequark I totally agree, they shouldn't. It's a difficult challenge to solve in a decentralised system, so I can understand why it is the way it is, but it's one of the big drawbacks of the Fediverse at the moment and it's why I hate the "just pick a random server, you can always move later" advice that gets thrown at new users (and also why I haven't moved to hosting my own server).

@mcc @onyxraven @whitequark The directory of dead URLs is an interesting idea, but my preferred solution would be a proper migration mechanism where the new server can authoritatively declare that all posts posted from the old server should have their URLs updated to point to the new one.