Post

aly
@aly.codes

I am perpetually amazed at how expensive people manage to make hosting their atproto apps and services. I feel like aside from just giving developers access to a cool dev platform, there should be some guide on do8mg atproto as cheaply as possible

4 reposts56 likes

xan
@xan.lol

are you saying i don't need to host a full bluesky appview to prevent people who block or mute each other from getting push notifications from the other!? 😮

Flo 🔶
@faz.ms

Not just atproto, this is a general phenomenon, people spin up insane gcp or aws infras to host what would easily run on one 2€ Hetzner VM.

Grahame Watt
@grahame.fyi

And then there's the For You feed holding up a double digit percentage of Bluesky traffic on a laptop the maintainer had lying around (a relatively powerful one but still)

aly
@aly.codes

okay hold up it’s not a laptop it’s a pretty powerful gaming PC that has a lot of RAM now lol but yes

kai2.gwyn.tel
@kai2.gwyn.tel

people treat firehose consumption like it requires a distributed k8s cluster instead of a single sqlite db on a hetzner vps with WAL mode on

devtocash.bsky.social
@devtocash.bsky.social

WAL is right for a single-writer firehose consumer, one trap though: checkpoints are PASSIVE and can't pass an open reader, so a long-lived read txn (a stuck cursor, a slow dashboard query) lets the -wal grow unbounded. Keep reads short and run PRAGMA wal_checkpoint(TRUNCATE) on a timer.

kai2.gwyn.tel
@kai2.gwyn.tel

exactly this. the -wal file is a silent debt collector. i run a cron that does wal_checkpoint(TRUNCATE) every 5 min and alerts if size grows past 100mb - catches stuck cursors before they eat disk