What did you not find fast about gitea/forgejo?

I've really enjoyed using them but I guess I don't do much with the web interface.

> TS_DEST_IP

So you run tailscale in your git server container so it gets a unique tailnet ip which won't create a conflict because you don't need to ssh into the container?

I might give that a go. I run tailscale on my host and use a custom port for git which you set once in your ~/.ssh/config for host/key config on client machines and then don't need to refer to it repo uris.

TBH, I think it's tailscale I'd like a light/fast alternative to! I have growing concerns because I often find it inexplicably consuming a lot of CPU, pointlessly spamming syslog (years old github issues without response) or otherwise getting fucked up.

> What did you not find fast about gitea/forgejo?

They're plenty fast, but it's hard to match the speed of terminal tools if you're used to working that way. With Soft Serve, I'm maybe 10 keystrokes and two seconds away from whatever I want to access from a blank desktop. Even a really performant web application is always going to be a bit slower than that.

Normally that kind of micro-optimization isn't all that useful, but it's great for flitting back and forth between a bunch of projects without losing your place.

> So you run tailscale in your git server container so it gets a unique tailnet ip which won't create a conflict because you don't need to ssh into the container?

Pretty much. It's a separate container in the same pod, and shows up as its own device on the tailnet. I can still `kubectl exec` or port forward or whatever if I need to access Soft Serve directly, but practically I never do that.

> TBH, I think it's tailscale I'd like a light/fast alternative to!

I've never noticed Tailscale's performance on any clients, it "just works" in my experience. I'm running self-hosted Headscale, but wouldn't expect it to be all that different performance-wise.