3はグローバル前提だとトータルで安上がりだと思うけど、その分コストの一極集中はあるので、そこを上手く運用できるかは今後の課題よね。技術よりも人の問題かもしれないけど。lexiconが十分に増えてjetstream的なのが一般的な未来を考えると、relayがどこまで頑張れるかが主題。3はグローバル前提だとトータルで安上がりだと思うけど、その分コストの一極集中はあるので、そこを上手く運用できるかは今後の課題よね。技術よりも人の問題かもしれないけど。lexiconが十分に増えてjetstream的なのが一般的な未来を考えると、relayがどこまで頑張れるかが主題。
jetstream is so good
そうよね。リアルタイムフィードをJetStreamに変更しなくても最近は結構安定していたから減っているんだろうなぁとは思ってた。
This might be of interest. I'm just writing the correct lexicon format to the PDS, which is then picked up by the Relay, which is then picked up by psky's instance of Jetstream and finally forward to the clients. I'm also connecting to psky's websocket to get new events github.com/Zetaphor/pui
A TUI client for picosky. Contribute to Zetaphor/pui development by creating an account on GitHub.
come to think of it, this may be an overly bold assumption because one of the thing jetstream does is strips all the verification stuff, which i could technically do while backfilling but ehhhh
i can backfill 11 million posts for emojistats.bsky.sh on my M1 Pro MBP in about 40 minutes which means i can backfill the rest of Bluesky on an equally fast server in about... 37 hours, give or take, if all goes well
Live counters and stats on emoji usage on Bluesky, broken down by language.
this assumes i can process things as fast as i do from jetstream which is a big assumption but will see
yeah, it was def not batching postgres inserts. this also finally gave me some ballpark perf metrics for the actual backfill (granted this backfill is with jetstream and i have to do the rest the usual way, need to benchmark that too) bsky.app/profile/alic...
i can backfill 11 million posts for emojistats.bsky.sh on my M1 Pro MBP in about 40 minutes which means i can backfill the rest of Bluesky on an equally fast server in about... 37 hours, give or take, if all goes well
Live counters and stats on emoji usage on Bluesky, broken down by language.
curious what others think! there are a couple different ways to do this, and feels like it might depend on that, but I think a reasonable golang client is still basically 1-2x vCPU, a couple GByte of RAM. jetstream can be very small. doing identity resolution/validation bumps things up?
The Air and the Eye are both rollerballs. They use a liquid ink (like a fountain pen) so they can be picky about papers in the same way. The Jetstreams are ballpoints (oil based ink) and will be totally different. Grab a Jetstream, but not the 1.0 size tips. .5 and .7mm are both great.