ATProto 中不存在实例。
There Are No Instances in ATProto

原始链接: https://overreacted.io/there-are-no-instances-in-atproto/

“Bluesky 的实例在哪里?”这个问题源于对 atproto 架构的根本误解。习惯了 Mastodon“联邦”模式的用户认为,社交网络必须由集成了托管和应用服务的互联“实例”组成。 在 Mastodon 模式下,用户被绑定在特定的服务器上。你的身份与该服务器挂钩;如果管理员停止联邦或服务器下线,你与网络的连接就会中断。这形成了孤立的“领地”,各领地之间必须通过复杂且容易出错的协议进行通信。 相比之下,atproto 将托管与聚合分离开来,其运作方式更像是 RSS 的进化版。在这种模式下,你的数据“驻留”在你的托管方,而各种应用程序则充当该数据的独立呈现。你不再是“处于”某个实例中,而是一个独立的个体,你的内容只是被各个应用程序聚合而已。 atproto 的去中心化不在于计算服务器副本的数量,而在于你可以自由选择托管服务商和交互界面。通过将这些层级解耦,atproto 避免了传统社交媒体和联邦平台的“原罪”,让用户能够真正掌控自己的数字身份和在线状态。

这篇 Hacker News 讨论帖围绕 Dan Abramov 的文章《ATProto 中没有实例》(There Are No Instances in ATProto)展开,该文认为 AT Protocol(Bluesky 的核心架构)的运作方式与 Mastodon 的服务器模型(ActivityPub)有所不同。 评论者们就 ATProto 是否真正消除了“实例”概念展开了辩论。一些用户认为,该协议并没有彻底移除实例,而是将架构细分为不同的专业化服务:个人数据服务器(PDS)、中继器(Relay)和应用视图(AppView)。普遍观点认为,虽然该系统的设计旨在解决 Mastodon 的“NxM”架构中固有的扩展性问题,但它依赖中继器作为高性能的数据传输管道。 批评者则指出,这些组件(如 AppView)仍然可能导致权力集中,并暗示 ATProto 与 Mastodon 之间的主要区别在于发送者、接收者和中介者的角色如何解耦。归根结底,这场讨论突显了 ATProto 的分段架构在性能效率与 ActivityPub 的去中心化社区管理模式之间的权衡。
相关文章

原文

Every single time a post about atproto hits Hacker News, somebody asks in the comments: “But where are all the Bluesky instances?”. The problem is, there are no instances in atproto! The question is a category error. Instances are a Mastodon-brained concept, and I wanted something I can link to that explains this clearly.

So this is that post.


I know RSS is still being used somewhere (podcasts?!) but its heyday is arguably behind. Which is a shame. For a few years, which some of us might fondly remember as the golden age of the web, it felt like blogging was a cool thing.

Now look at this picture because it’s going to be important:

alice'sblogcat'sblogbob'sbloggooglereaderfeedly

As a reminder, you publish stuff on your own blog, which you can either self-host or host on a popular blogging platform. But then everyone’s stuff gets aggregated into apps like Google Reader and Feedly, or collective blogs like Monologue (RIP).

Note that hosting and aggregation are two separate things. Your posts don’t “live” in an app like Google Reader. Apps are mere projections of the Blogosphere.

Seriously, make sure this thought sears into your brain; it’s going to be essential.


Here’s what you could call an evolution of this concept.

We put a box around the whole thing so that everyone is enclosed in the same space so we can show ads and stuff. Also, let’s leave only one app (we can let alternative apps live for a while, but not for long). That’s traditional social media.

alice'spostscat'spostsbob'spostsfacebookthe facebook newsfeed

Oh no, now we have centralization!

Oh no, runaway network effects!

Oh no, bla bla bla.

What do we do?

We need to decentralize this somehow.


I say “Mastodon” here because if I say “ActivityPub” instead, a crowd of people will show up and say that actually what I’m describing is how Mastodon chose to implement ActivityPub. Whereas ActivityPub by itself does not really specify how to actually use it in practice. I’m sure this is all very interesting—but I digress.

How do we decentralize a social network?

Let’s build a version of what we saw earlier, but make it self-hostable. Then every community can have their own “little Facebook” or “little Twitter”. We’ll call them instances. They’re kind of like countries—because you live “inside” one of them:

alice'spostsalex'spostsann'spostscat'spostscrow'spostscali'spostsbob'spostsbree'spostsboba'spostsmastodon instance #1mastodon instance #2mastodon instance #3the newsfeedthe newsfeedthe newsfeed

But wait, this opens a bunch of questions.

How do you choose which instance to join? Maybe you’re a member of a few overlapping communities. Well, I guess you’re just gonna have to pick which community’s admins you trust the most with handling your identity and data.

Okay, now another problem—what if my friend’s on a different instance? How will they see my posts? Since each instance is basically its own little Facebook, they have no shared source of truth. So they have to send messages to each other:

alice'spostsalex'spostsann'spostscat'spostscrow'spostscali'spostsbob'spostsbree'spostsboba'spostsmastodon instance #1mastodon instance #2mastodon instance #3the newsfeedthe newsfeedthe newsfeed

This network topology might remind you of warring fiefdoms in Ancient China.

If Alice-from-instance-#1 follows Bree-from-instance-#2, the two instances make an agreement: Bree’s posts will be forwarded to instance #1 so that Alice can see them. That’s called “federation”. You post on your instance, and then it gets forwarded to other instances whose users wanted to hear from you.

This picture has a few interesting implications:

  • You “belong” to your instance. You’re not Alice, you are Alice-from-instance-#1. That’s why your Mastodon login is literally [email protected]. “Where you’re from” is an immutable part of your identity. (Somehow, this manages to be even more restrictive than countries and nationalities.)
  • If your instance’s admins pick a fight with another instance’s admins, they may choose to “stop federating”, and no longer forward any posts between them. That could be a surprising reason why you’re no longer seeing posts from your friends.
  • If your instance goes down, your identity ceases to exist. People who followed you followed you-from-that-instance, not some abstract platonic “actual you”.

Oh, and the arrows between instances scale as O(n²). This might not matter much now, but it could matter if this approach to social networking becomes popular.


Now forget all of that—full reset.

The mistake was when we drew this box:

alice'spostscat'spostsbob'spostsfacebookthe facebook newsfeed

Erase the box.

Go back to this:

alice'sblogcat'sblogbob'sbloggooglereaderfeedly

We have hosting where things actually “live”, and apps aggregate from them. This worked for blogs just fine, so why wouldn’t it work for literally everything else?

alice'sstuffcat'sstuffbob'sstuffapp #1app #2

Like RSS, but for all kinds of stuff.

That’s atproto.


Now you know! There are no instances in atproto.

Instances are these Mastodon-brained things:

alice'spostsalex'spostsann'spostscat'spostscrow'spostscali'spostsbob'spostsbree'spostsboba'spostsmastodon instance #1mastodon instance #2mastodon instance #3the newsfeedthe newsfeedthe newsfeed

They’re those isolated bundled hosting+app fiefdoms that send stuff to each other.

Compare this picture to atproto.

In atproto, we cut hosting apart from the aggregation at the network level:

alice'sstuffalex'sstuffcrow'sstuffcali'sstuffboba'sstuffatprotoapp #1atprotoapp #2atprotoapp #3bree'sstuffann'sstuffbob'sstuffcat'sstuffatproto hosting #1atproto hosting #2atproto hosting #3

There are no instances at all! There’s hosting you can swap, and there are apps that aggregate from everyone’s hosting. It’s very much like RSS and Google Reader.

The decentralization of atproto is richer in structure than “many copies of one app”:

You care about decentralization? You have full agency here. Decentralize away.


Now you see why every decentralized social media discussion is derailed by this.

Mastodon users measure decentralization by the number of instances because that’s the only thing you can do in Mastodon. If there’s only one type of “box”, and each box is “an app coupled with hosting”, the only thing you can do is to host more of these boxes and get them to talk to each other. They’re isolated by default.

In atproto, every app is a projection of the whole Atmosphere, just like Feedly and Google Reader are projections of the entire Blogosphere. You mostly “decentralize” by swapping your hosting, and/or by making and trying new apps. Running many full copies of the Bluesky database server is possible, but it’s not any more useful than running many copies of Google Reader. People do set them up (cue Blacksky), but they arise to meet someone’s specific needs (like a different moderation philosophy). There are other approaches too: this Bluesky client has no dedicated database at all, and it just hits a free community-run cache of everyone’s hosting. Shared network infrastructure like Relays has been cheap to run for a year now.

This is why “counting Bluesky instances” is so misleading. What matters is:

  1. Are people migrating to alternative hosting?
  2. Are people trying and making new apps?

Separating hosting and apps fixes broken incentives in closed and in federated social. Coupling hosting and apps was the original sin, and the fix is simple.

Keep our stuff outside the apps; let the apps aggregate over it.

alice'sstuffcat'sstuffbob'sstuffapp #1app #2

Like RSS and Google Reader.

联系我们 contact @ memedata.com