4 ms·

> Bluesky uses content-addressed content, so that content can survive if a node goes down Does anyone have recommended reading about how Bluesky implements thi

by tmiku 2y ago

> Bluesky uses content-addressed content, so that content can survive if a node goes down Does anyone have recommended reading about how Bluesky implements this? I've skimmed the docs at atproto.com but I'm not sure how this fits in there. It's also not clear to me how content-addressed content protects against node failures in general.

steveklabnik 2y ago
Through stuff like https://ipld.io/specs/transport/car/carv1/ https://ipld.io/specs/transport/car/carv1/ > It's also not clear to me how content-addressed content protects against node failures in general. Imagine my PDS host disappears. Server's gone. I can spin up a new PDS of my own, or sign up for one that exists, and then replay all of my posts from the relay, and put them into the new PDS. I'm now back up and running without any involvement from the old PDS host.
- tmiku 2y ago
  Does the relay durably store data from the PDSes it crawls? I thought it just aggregates and distributes data from the PDSes it is actively connected to. Edit: I now see via https://docs.bsky.app/docs/advanced-guides/backfill https://docs.bsky.app/docs/advanced-guides/backfill that this is the case. I'm surprised that the docs don't mention it more clearly that relays store the full contents of the attached PDSes, but maybe I'm just not up to speed on learning how this works.
  - steveklabnik 2y ago
    Currently the relay stores the entire history. That’s why people are saying that it’s a centralization point that’s unlikely to be able to be hosted by others as the network grows. That said the team is aware of this and has ideas for how more limited subsets could work. That’s more important as time goes on, so this is yet another one of those prioritization related issues where improvements will happen, but it’s also good to know about the current state of things too.

by tmiku 2y ago

steveklabnik 2y ago

Through stuff like https://ipld.io/specs/transport/car/carv1/ https://ipld.io/specs/transport/car/carv1/ > It's also not clear to me how content-addressed content protects against node failures in general. Imagine my PDS host disappears. Server's gone. I can spin up a new PDS of my own, or sign up for one that exists, and then replay all of my posts from the relay, and put them into the new PDS. I'm now back up and running without any involvement from the old PDS host.

tmiku 2y ago
Does the relay durably store data from the PDSes it crawls? I thought it just aggregates and distributes data from the PDSes it is actively connected to. Edit: I now see via https://docs.bsky.app/docs/advanced-guides/backfill https://docs.bsky.app/docs/advanced-guides/backfill that this is the case. I'm surprised that the docs don't mention it more clearly that relays store the full contents of the attached PDSes, but maybe I'm just not up to speed on learning how this works.
- steveklabnik 2y ago
  Currently the relay stores the entire history. That’s why people are saying that it’s a centralization point that’s unlikely to be able to be hosted by others as the network grows. That said the team is aware of this and has ideas for how more limited subsets could work. That’s more important as time goes on, so this is yet another one of those prioritization related issues where improvements will happen, but it’s also good to know about the current state of things too.