Post

the primary microcosm api gateway is still running on a digitalocean droplet. hopefully will be able to move it next week so i can close my account there. wish i didn’t always have such a backlog.

(it’s mostly luck thats that’s the only thing there. i know other folks have had more painful moves)

and in case you’re curious: the gateway is the public access to all the services that run from my house. the secondary runs on a lil vultr vps. the pair are dns load balanced. this is nice because the move should be zero downtime.

the gateways run nginx and reach my home network via tailscale-- cleartext http over wireguard. it has all just worked super reliably. the gateways also - terminate tls - cache responses for a few seconds - rate-limit by client ip - are small enough to fail before a DoS could ruin my home internet

these days they serve 250–1,800 req/sec, mostly for constellation (slingshot isn’t hosted at home) dns load balancing isn’t perfect, the volume is often unbalanced between the gateways. this unbalance is good actually, since their purpose is redundancy, not capacity. either can still do full load.

(eventually i’ll replace nginx with a custom proxy layer, when i finally build out microcosm dev accounts, so we can do access tokens for higher rate limits and priority. nginx has just been so solid it’s never become urgent)

the one weird operational thing recently: someone bypassing rate limits through a major residential proxy service (hundreds of thousands of distinct ips) with a fake browser user agent. not very sophisticated, but very annoying. but also not enough volume to impact service 🤷‍♀️