yeah this feature drives me nuts, I never want to @ someone from their page

a picture of the tracy UI showing various spans, and a histogram of timing on a particular zone type

If you're writing C, C++, Rust, or Zig and not using tracy, you should take a look at it! Very useful

github.com/wolfpld/tracy

Yeeep they just give you like some aws credits which are gone in 3 months

on-prem is hard work though, but it scales very nicely if you treat it as a repeatable process with extensive automation

A bunch of those switches were "used" but actually new, just at like a 10x discount

Some were a bit more beat up but eh who cares for that price

I'd recommend just building things as light as possible and honestly just doing what is fun fun you to build lol

I'm still thinking about solving this problem! It's just gonna be some time

Honestly I am a BIG fan of buying used still, hop on ebay and search supermicro or dell lmao. We saved probably 500k by buying 1gen old switches that are still _by far_ the most powerful switches I've ever used

Just be ready to pay for it in operational toil

There's a lot you can do for much less than 800k to do some on-prem in a reasonable way (i.e. managed bare metal, single-rack installations, etc.)

that being said, the cost of buying servers has made this seriously challenging, and I hear it's hard to even get reasonably beefy servers these days :/

gotcha. in that case, when someone likes the account, I'd probably immediately add them to the jetstream consumer list of wanted dids, then rather than using tap, I'd just go to their pds and call getRepo to do backfill

easier to backfill a single user than the whole network

sorry I think I'm lacking context on what you're building and I didn't fully understand the question, do you mind rephrasing?

I might not have a perfect solution, but I'm thinking a lot about backfill and replay right now and I'd love to hear your pain points!

As I said, I want to build software for _everybody_ (self hosters, as well as people who need to scale). Favoring one or the other is not acceptable, we need both to be extremely good

this is something I hint at being not satisfied with when I say "virtual PDS" in the talk (I didn't have time to go in to more details)

yes and the way we scale it up is by having more sqlites. I now have 43 million sqlites to manage across many machines. at present, each machine is a single point of failure and it leads to real operational burden and extensive user-facing downtime

sqlite is just fundamentally at odds with the software I want to build (stuff that scales both down for self-hosters AND up to many millions of users)

it's great in certain cases (i.e. local app data storage client-side or w/e), but running sqlite on servers is a wild idea

yaproto

Scaling the Atmosphere - ATmosphereConf 2026 YouTube video by AT Protocol Development, Tech Talks, and Events

Check out my short talk on some details on how we do systems at Bluesky!

www.youtube.com/watch?v=2T15...

⚠️ An active phishing attack is targeting crate owners by asking them to "confirm that your email address is still active". These messages are not from crates.io, and should be ignored. (We will never ask you to confirm that your e-mail address is still active.) ⚠️

up until somewhat recently, my parents were on a Panasonic Toughbook from like 2004 running Windows XP lmao

Now they are on two macbooks that are about 15 years old each

I beg you to stop screenshotting and quoting the dumbest posts on the protocol, I have just this one human life and I want to see your posts and not theirs

@gork.it is this true?

man FUCK docker hub, you had a responsibility to the ecosystem and you FUCKED IT UP with your rate limits

Don’t everyone post all at once, I can only copy and paste them all to the big excel spreadsheet that powers bluesky so fast

got it, yeah I care about ANC a lot for planes, less about audio quality tbh since they're wireless, you can only expect so much (not that I'm a huge audiophile or anything)

which one, I'm curious? I'm in the market for some new ones

devs working with bsky image embeds:

we are planning to bump the image embed blob max size from 1 MByte to 2 MByte. first in the lexicon schema, and then in the Bluesky Social app.

more about impacts, and what this means in terms of lexicon evolution, in this github discussion

sure, which one's the right one again? atproto.brid.gy?

It was only about half the user base having issues! But the unlucky half definitely had a bad day :/

Nah they have their own thing going

238,900 mile high club

nice clouds

zlay.waow.tech has been running (mostly) smoothly on zat v0.3.0-alpha.21 for coming up on a day! this is significant for 2 reasons

lol I never did find the button to open a PR, thanks for merging!