Content
@
https://warpcast.com/~/channel/fc-devs
0 reply
0 recast
0 reaction
christopher
@christopher
In December, we replaced @farcaster/shuttle with Waypoint, our custom event processing system built with Rust, protobufs, and sqlx. It's very fast and much more memory efficient. Running on an optimized Kubernetes cluster with Terraform infrastructure, it also requires minimal costs and delivers instant upgrades. This change addressed memory leaks and reliability issues with Farcaster hubs and shuttle containers, while reducing our support dependency on Merkle as they pivot focus to Snapchain development. Our architecture also leverages GCP's managed services like PubSub and Cloud SQL/Bigtable for more coherent data systems. Waypoint will be open-sourced as an app and lib soon so anyone can run their own internal protocol data services. DM me directly to contribute to early development. Below is our Farcaster cluster slowly eating up memory and resources, causing disruptions in event subscriptions as it randomly restarts to heal.
5 replies
6 recasts
30 reactions
YB
@yb
i don't understand a lot of this, but can u give eli5 of why neynar doesn't solve infra side for the unofficial app? curious why you're investing so much resources on infrastructure, some longer term play that I'm missing?
1 reply
0 recast
7 reactions
christopher
@christopher
Neynar does a great job of providing APIs and services, and it's likely we'll find some way to work with the team eventually. Our core focus is discovery and personalizations, so having infra like this sit together on the same physical machine is important. 200ms delay * 18,000 events = 1 hour delay.
2 replies
0 recast
2 reactions
christopher
@christopher
Running this is free since it's right-sized, so if cost actually mattered, it's also one less invoice to pay.
0 reply
0 recast
2 reactions
YB
@yb
🫡
0 reply
0 recast
0 reaction
