@bad-example.com - fig (aka:[phil])

27 Jul 2026 18:52 +0000

roomy still just blank-screen fails for me in firefox :(

25 Jul 2026 21:01 +0000

hubble’s (and hubble-sync’s) backfill is probably acceptable now but the depths of despair i was reaching this week when backfill throughput was absolutely *crashing* at what i thought was about the halfway point… [contains quote post or other embedded content]

24 Jul 2026 13:45 +0000

haven't though this through but: if every sync consumer periodically crawled `listRepos` on their upstream(s), they could discover account status changes that were missed over subscribeRepos, and we could actually have eventually-consistent account status in the network?

24 Jul 2026 02:25 +0000

it’s a spec SHOULD to getRepo from your upstream and follow redirects but i’m struggling to get much more than 300 redirects/sec out of the bsky relay

23 Jul 2026 14:05 +0000

major fomo not being at ietf 126 in person, but very excited to remotely attend the working group! gonna talk about how repository archives need to explicitly signal whether their blocks are pre-ordered or not near the end, which might one day finally make backfill slightly easier

22 Jul 2026 15:08 +0000

waow 50% less memory for a mimalloc -> jemalloc swap (yes that is under-3G-rss while backfilling)

21 Jul 2026 19:44 +0000

thank you robot boris [contains quote post or other embedded content]

20 Jul 2026 23:48 +0000

hubble-sync doing full-network (full-content!) backfill on a lil 4 core 8GiB machine, using less than 4GiB 😌

19 Jul 2026 17:13 +0000

full network backfill, disk-i/o saturated edition

17 Jul 2026 16:26 +0000

eyyy my block order spec concern *is* real - affects 0.4% of atproto repos (1% sample) - are they weird tho? no. dup occurrences: 512833 app.bsky.graph.block 36512 app.bsky.feed.post 32726 app.bsky.graph.listitem 2391 app.bsky.feed.like 2025 fm.teal.alpha.feed.play ...

17 Jul 2026 14:38 +0000

another day, another chance to re-remember again that the reference pds returns xrpc error "NotFound", not the documented "RepoNotFound" error, at `getRepo` and `getRepoStatus` :/

15 Jul 2026 17:55 +0000

"talking a little bit more of the terribly-named backends" wow roasted by @flo-bit.dev lmao <3

15 Jul 2026 15:10 +0000

random atproto sync notes - for whole-network backfill, you have to spread your `getRepo` requests across PDS hosts to get reasonable throughput, so it's a scheduling problem. it's also a spec-SHOULD to make these requests directly to your upstream (not PDSes directly) and follow its redirects. https://tangled.org/microcosm.blue/hubble/blob/main/hubble-sync/src/resync_scheduler.rs#L16

14 Jul 2026 17:33 +0000

i was thinking service proxying for allowing suspended/takendown accounts access to their archives on hubble, but... do pdses buffer the whole service-proxied response in memory or stream it back? (even with streaming it's potentially a lot of bandwidth through the pds) direct req + token i guess