(can you tell I'm mostly on fedi these days, on an instance that lets you have 2500 chars per post... I've forgotten how to speak in the clipped sentences you need for twitter clones :) )
Keeping resource allocs mostly fixed is a huge deal for reducing the amount of cluster chaos, and enables a bunch of other nice things (e.g. preload images on the allocs's machines during rollouts, so the rolling restart can be as fast as the healthchecks allow them to be, no waiting on the network)
That's the big thing I remember missing from Borg: an "alloc" is the unit of scheduling that reserves resources on a worker. "jobs" start inside alloc, possibly 1:N, but jobs can change without changing/moving the alloc. Means less scheduler load, and faster rolling restarts, just SIGTERM + exec.
I would still encourage/want to hear details about pod mutability. In my mind's eye, ~all fields of a pod should be mutable in place. That's not the same as allowing the _containers_ to mutate in place, some edits may result in a full in-place pod restart.
Hah, blast from the past. I just re-read it, and I think I still stand by most of what I said (especially the bit about most of it being half-baked and needing more design :) ). I sympathize with the network defeatism, having to operate in consensus reality is a real drag.
Fancy! We could use some of those, what actuators are you using? Do they open the whole top frame?
Maybe, though I don't see what the incentive is either, and AIUI the base load is much higher than a fedi instance (i.e. doesn't scale by local activity), so there still seems to be an issue of what gets done to justify the high cost? I dunno, we'll see I suppose.
The fediverse needs a whole bunch of work, but I understand how it sustains itself. Bluesky, so far, I don't see any alternative to speedrunning twitter's monetization strats once the runway grows short
I'm having similar thoughts, but also... I still don't know how this place is sustainable once the free money runs out. The distribution still concentrates all the costs and power on the one aggregator everyone uses, and I haven't seen a plan for revenue... Which makes me assume I'm that plan
The graph's changed a teeny bit since that post, as we find more bugs and/or cursed configurations. A notable one I recall was container runtimes that bind-mount /etc/resolv.conf, and thus sometimes contains stuff that makes no sense, and also can't be atomically overwritten with mv(1)
For all the issues the fediverse has (and it does have them), it has an understandable sustainability model that does not lead back to being twitter. It's a fragile model, undergoing growing pains... but I can see how fedi can sustainably continue to exist.
We'll see if bsky follows suit.
Still cheering loosely from the sidelines, because I hope answers do emerge... But until they do, this site is running on borrowed time and the high of not yet being big enough to require the policies and behaviors of twitter to survive.
I'm a firm believer in incentive structures, and so far I see a big gap re: how will bluesky sustain itself long-term? Unless that's answered, we have to assume the default answer: VC funding, which inevitably leads to twitter-alike behavior.
A lot of the excitement seems to be "bluesky is so much better!" but that can be entirely attributed to being smaller. The new moderation guidelines are just one example of the growing pains that lead straight back to being Twitter.
I'm not really posting here yet, because I don't yet see how bluesky will have an outcome any different to twitter's. AFAICT there's no known plan for profitability or even sustainability, so no way that I can see to avoid the exact incentives that made twitter what it is.
bah I was hoping that a .library TLD existed, so with federation and dril's consent we could have archives federated in at dril.library
So you're saying bsky needs a thing to import a twitter archive, so all of dril can be transplanted for future historians
while we were (mostly) okay with being able to see who liked an individual tweet, quite a few privacy and security folks at twitter considered the “full like history” to be perhaps its biggest foundational design mistake.
it has ruined a _lot_ of lives and careers.
Who said anything about friends, I'm talking about a social network graph, which is much more important
Some UX feedback: the "X followed you" notifications don't have a "follow back" button, which makes it a bit tedious to do so. This is especially important in this early "bootstrapping" stage, where a bunch of my social network is joining and so I need to follow-back a couple dozen people at a time.
not on hand, it was a tweet from Jack in early 2022 I think? I found it via a press article talking about the potential impact to bsky of the twitter buyout, from around the time of the initial purchase offer.
+1, remarkably smooth flow, other things that use domain verification should take notes!
A bit more digging: bsky is a Delaware PBLLC, which is one of the places where PBLLC public benefit objectives have teeth, which is good.
Also found a statement from Dorsey that twtr doesn't own bsky, despite being (afaict?) its sole funding source. So, what _did_ twtr get in return for funding?
In particular, I'm curious because depending on the PBLLC's home state, a big enough chunk of shareholders (2/3 in Delaware) can alter the public benefit objectives. So, knowing who effectively owns bsky feels important to understand the risk of future ethical drift.
Learning more about Bluesky, and TIL PBLLCs, which balance public benefit against shareholder financial interests. Neat!
Curious though: Bluesky's PBLLC blogpost says initial funding for the PBLLC was provided by Twitter in 2021. So... what's the current ownership split of bsky?
Hi! Look I'll be honest, my main place right now is over on the activitypub-based fediverse, so in the short term don't expect much from this account. I'm mostly curious about the tech and getting a feel for the vibe of the place :)
