[kyle-rb]

No I think ATProto is "built this way". The firehose is just the output of the Relay. If Bluesky wanted to shut off the firehose, they would have to make changes to ATProto or stop conforming to ATProto.

I understand that Bluesky's conformance to ATProto is just a promise, but it's a better promise than you get from most websites. Also in the meantime, if you migrate to a self-hosted PDS, you can ensure that even if Bluesky restricts access to their Relay's firehose, 3rd party Relay servers can still pick up your posts and publish their own unrestricted firehose.

[immibis]

What do you think would happen if the Bluesky company suddenly blocked everyone but https://bsky.app/ servers from using their relays?

And what if, before they did that, they updated the PDS code so it blocked all relays except for their one?

I'm not asking what you would do. I'm asking what would happen because of what everyone does. I think the name "Bluesky" would refer to the fully centralized bsky.app, and 99.9% of users would never notice a difference. Users who had other PDSes would either quit (nobody noticing their departure) or sign up to bsky.app like everyone else. The events of Twitter show it's probably the latter - people bent over backwards to comply with Musk to keep their accounts.

[OneDeuxTriSeiGo]

> What do you think would happen if the Bluesky company suddenly blocked everyone but https://bsky.app/ servers from using their relays?

That's not how it works. Appviews pick the relay they use, not the client/user. The relay is used for gossip into the appview (and other things).

More importantly, appviews never see the client/user directly. Appviews only talk to the PDS. Really most things other than the client ever only talk to the PDS or listen to the relay. The only thing that ever directly talks to the client is the PDS.

The way atproto services generally work is the client configures a series of XRPC requests with HTTP headers to determine what appview, labelers, etc to use and it issues that request to the PDS. The PDS then proxies that request to the appview or wherever and they respond back to the PDS which routes the response back to you.

So in a real sense your PDS is not just a data host, but also operates akin to an IRC bouncer.

-----

> And what if, before they did that, they updated the PDS code so it blocked all relays except for their one?

PDS relay routing, etc is mostly all handled manually via config files,etc so this isn't really a concern. And PDS code is probably the "easiest" part of the ecosystem to hack on which is why there are like 6 different implementations with the majority (like 4) that maintain near feature parity with the "bluesky PDS" software.

And importantly, the bluesky PDS is literally a sqlite DB, an OAUTH implementation, some go IPLD data structure manipulation code, and a go XRPC router. It's fairly trivial to hack on as needed.

------

> I'm not asking what you would do. I'm asking what would happen because of what everyone does. I think the name "Bluesky" would refer to the fully centralized bsky.app [...]

Migration currently isn't perfect but within ~6 months it should be ironed out by the community at which point migrating off a PDS to another is just a matter of:

1. click button on new PDS to transfer/"create new account".

2. set your new email, password, and list your old/current handle.

3. get auth code via email (one from the new PDS and one from your DID provider)

4. input codes into migrator interface (for whichever migrator you are using)

5. log into your apps again.

There are multiple large PDS operators working really really hard to spin up operations (proper backups, failover, HA, etc) so they can run reliably and avoid the "my mastodon instance imploded guess everything is gone" issue. Open federation is only about ~ a year old (plus change) so the community is only just now really reaching the "mature third parties" stage.

[immibis]

So, you're making stuff up that obviously has no basis in reality here.

Migration would be impossible because the bsky.app PDS wouldn't allow anyone to access the data except for the bsky.app relay.

other appviews wouldn't display bsky.app data because both the PDS and relay would block them.

[hoidofyolen]

First of all, Bluesky the company would have to implement auth on the communication between relay and AppView, cut off everyone from using their relay, and specifically stop ingesting content from non-Bluesky PDSs. There would be time when we realize that's happening to get another AppView up and running. There are arguably enough resources in either the ATproto dev community or other funding sources that another AppView could pop up within a week or two, and be maintained by the community for months without issue. And Bluesky wouldn't likely stop existing altogether, so people would log into Bluesky, see the news (or hear it elsewhere), and see people talking about how to get access to your account.

Secondly, while you're right that most people don't have copies of their repos, some copies of the entire network exist in the community, as well as copies of the PLC Directory. Within a couple weeks we would have community run versions of the AppView, PDSs, Relays, and PLC, and some seriously pissed off community members who would now want to do everything they can to take up the mantle temporarily. Soon Bluesky the app, as well as Instagram and Twitter, would be flooded with tutorials of how to recover your accounts and migrate to another PDS.

If your point is that we'd lose at least half of the users of Bluesky, then yea probably. But ATproto would be just fine, and if people want to get their data back they likely can, as long as they can stomach a little work. And that process is getting easier all the time.

[immibis]

Who would use the new network? It would just be Mastodon Vs Twitter again. All the people you want to read the tweets of would (still) be on Blue Sky, not Black Sky or Red Sky or President Zelen Sky or whatever.

"Convince a HN user that a corporation with the ability to cut you off from your feed sources for more profit won't do exactly that and just because they use something decentralised-ish today doesn't mean they have to keep using that thing if they don't want to" challenge (difficulty: impossible)

[OneDeuxTriSeiGo]

> "Convince a HN user that a corporation with the ability to cut you off from your feed sources for more profit won't do exactly that

It's not just a technical restriction though. It's also a legal restriction. Bluesky PBLLC is a public benefit corporation and they are at least in moderate part beholden to their charter.

More importantly, their initial investment contract requires them to further the decentralisation of social media and exposes them to legal consequences should they deviate from that mission.

Those combined make it effectively impossible for them to lock in bluesky. Doing so would require technical changes that would be under no uncertain terms against the charter of the company and against the terms of the investment contracts that initially funded the company. It's a poison pill that would kill the company and destroy any "shareholder value" the moment they try to lock down the service or lock in the users.

You aren't going to see them try this type of brazen lock-in because it'd be explicitly harmful to every investor/VC and saddle the company in legal hell until it smoulders into ash.

[immibis]

> First of all, Bluesky the company would have to implement auth on the communication between relay and AppView, cut off everyone from using their relay, and specifically stop ingesting content from non-Bluesky PDSs

Yes, that's as simple as an IP address check, since they own all of it. Could alternatively be a password (bearer token).

> And Bluesky wouldn't likely stop existing altogether,

In fact, the change would be noticed by almost nobody, because 99.9% of users would still see 99.9% of their following (including 100% of the ones who populate the default feed) and would have to go far out of their way to see any change.

> There would be time when we realize that's happening to get another AppView up and running. There are arguably enough resources in either the ATproto dev community or other funding sources that another AppView could pop up within a week or two, and be maintained by the community for months without issue. And Bluesky wouldn't likely stop existing altogether, so people would log into Bluesky, see the news (or hear it elsewhere), and see people talking about how to get access to your account.

All irrelevant since it wouldn't have the content and people go there for the content. Views go where the content is.

> some copies of the entire network exist in the community

All irrelevant since it wouldn't have the new content posted after the block and people go there for the new content. New views go where the new content is.

> and some seriously pissed off community members who would now want to do everything they can to take up the mantle temporarily

Yes, exactly like the people who were pissed off at Twitter so they started Mastodon. Made literally no difference to Twitter. Most of those people even remained on Twitter and posted as much on Twitter as on Mastodon. Because views go where the content is. Thinking anything else is wilful delusion.

> Soon Bluesky the app, as well as Instagram and Twitter, would be flooded with tutorials of how to recover your accounts

Why recover? If you're part of the 99.9%, nothing changed for you.

> and migrate to another PDS.

I have no doubt there would be a flood of tutorials of how to make sure that only 0.1% of former Bluesky users will ever see your tweets, but nobody will follow that tutorial because they don't want to make sure that only 0.1% of former Bluesky users will ever see their tweets.

> If your point is that we'd lose at least half of the users of Bluesky

0.1% would be lost.

> But ATproto would be just fine

but completely useless, since its main use case is getting Bluesky feeds and that would be switched off

> and if people want to get their data back they likely can

This is nobody's goal. People log into Bluesky to see tweets from the people and topics they follow, not to "get their data".

[OneDeuxTriSeiGo]

No. That's just not remotely close to true or feasible.

> So, you're making stuff up that obviously has no basis in reality here.

I cannot understand why you are claiming this. I'm basing off the actual architecture and the way the parts interact. The design is just not feasible for locking down. Doing so completely breaks the model and it still leaks like a sieve if you try to.

----------

> Migration would be impossible because the bsky.app PDS wouldn't allow anyone to access the data except for the bsky.app relay.

Nope. Migration is still fully possible. Migration doesn't happen via the relay or any PDS->PDS mechanism. Migration is done via the client. The client/user runs operations on the source PDS, the destination PDS, and the DID registry. All the data is transported between by the client.

Specifically the way it works is you export/backup your information from your current PDS (in the form of a CAR file + blobs). Technically this step is optional. Even if the PDS goes offline or becomes hostile you can actually largely reconstruct this data from the network. Then you "create a new account" on the new PDS and upload your data that you backup up/recovered onto the new PDS. Then you update your DID to point to the new PDS. And finally you deactivate the account on the original PDS (basically saying I no longer store stuff here anymore).

This is part of the reason why migration tooling is a bit bumpy. Your JS script or app has to do the entire process by itself rather than letting the backends handle it. However it does make them extraordinarily resistant to data loss and/or takeover.

----------

> other appviews wouldn't display bsky.app data because both the PDS and relay would block them.

Relays work via gossip. If you can see the relay at any point, you can gossip 100% of their contents to another relay.

In the event bluesky PBLLC locked down their appview and PDS, they'd still have to make the relay open or everything breaks. Feed providers need access to the firehose. Labelers/Moderation Services need access to the firehose. And so on.

Everything is built with an assumption of a public firehose and if you lock down the firehose, all you need is one person to listen to the locked down firehose to 100% replicate it and gossip onto any other relay.