If Kubernetes is the answer ... you very likely asked the wrong questions.
Reading about JamSocket and what it does, it seems that it essentially lets you run Docker instances inside the Jamsocket infrastructure.
Why not just take Caddy in a clustered configuration, add some modules to control Docker startup/shutdown and reduce your services usage by 50%? As one example.
I’m not sure what you mean by that reducing service usage.
The earliest version of the product really was just nginx behind some containers, but we outgrew the functionality of existing proxies pretty quickly. See e.g. keys (https://plane.dev/developing/keys) which would not be possible with clustered Caddy alone.
My understanding was that K8s itself has overhead, which ultimately has to be paid for, even if using a managed service (it might be included in the cost of what you pay, of course).
I did add the caveat of "with modules" and the idea of sharing values around to different servers would be easy to do, since you have Postgres around as a database to hold those values/statuses.
HTTP proxying is not much of our codebase. I wouldn’t want to shoehorn what we’re doing into being a module of a proxy service just to avoid writing that part. That proxy doesn’t run on Kubernetes currently anyway, so it wouldn’t change anything we currently use Kubernetes for.
Reading about JamSocket and what it does, it seems that it essentially lets you run Docker instances inside the Jamsocket infrastructure.
Why not just take Caddy in a clustered configuration, add some modules to control Docker startup/shutdown and reduce your services usage by 50%? As one example.