Been perusing the ActivityPub protocol and I have concerns BUT am also aware that I could easily be misunderstanding some things.

1) When a user creates a new post, their instance has to fan it out to every follower with an HTTP POST per follower? So fanout scales with the cardinality of the follow graph, not the federation graph? Do I have that right?

2) There doesn’t appear to be a backpressure mechanism between servers?

Few failure scenarios that worry me.

1) A post goes viral and starts getting tons of engagement. Each engagement requires an individual HTTP POST back to the original instance? What happens if that server can’t keep up and requests time out or get rejected? How do you prevent retries from backing up across the fediverse?

2) If two high-follower users are replying to each other, the fanout is going to get immense fast.

If two people with 100k followers each (disjoint sets) are replying to each other how are their instances expected to handle that volume?

This protocol seems inherently uncacheable? How are instance operators able to keep costs under control given that?

A prediction: big instances are going to build their own out-of-band federation protocol to stream updates to each other such that servers fan out to their own users, rather than servers fanning out to every user.

@youfoundryan i predict smaller instances will cap follower limits for their users. a 1M follower acct will essentially be its own instance