The routine tracker that couldn't track my routine

This week the backend for my daily routine tracker went down, which means I couldn’t log my routine for the day. The app that tracks my routine was the thing disrupting my routine. I’d call that poetic, but I was mostly just clicking deploy.

The backend lives on Railway. It’s a prototype, nothing fancy, but it’s what powers the whole app. Then Railway had a platform-wide incident. The status page had a long thread of updates, and none of them made it work for me. So I clicked deploy. I got their error message. I clicked deploy again. It’s a quiet kind of rage. Mostly I just refresh and sigh.

What bugs me isn’t the error itself. If my own deploy fails, that’s on me, and I’ll go dig through logs like an adult. This was different. It was a system outage, and they were publicly reporting it. There was nothing to debug on my end. All I could do was wait and read updates that kept not fixing my problem.

I’ve only been on Railway for about five months, and that’s not a lot of trust to build up. It also doesn’t take much to dent. I’ve used Google Cloud professionally and never had this kind of trouble, so naturally I started thinking about moving the backend there. My bar is low: don’t go down for hours.

Railway is finally back up, and for now I’m staying put. But they’re on thin ice. An outage like that while I’m just testing is fine. It’s a prototype, and the only person who suffers is me and my unlogged routine. Once I’m ready to put this app in front of more people, though, I definitely don’t trust putting prod on Railway. That’s where Google Cloud comes back into the conversation.

The part that went well

The good news is that the app has been fun to build. It’s actually how I got into using an LLM for Blender. I made a 3D model of the human body so the app can show off different organs, and once that existed I realized how easy it would be to build workout features on top of it.

The lungs were the hard part. All those branches and lobes. My process was to keep tweaking and giving the LLM feedback until it looked right. Eventually I figured out I was never going to perfect them and should just move on to the next thing. A lung that’s good enough is still a lung.

So that’s the week: figured out Blender with an LLM, shipped passable lungs, and got a lesson in who actually controls my uptime. Railway gets to keep the prototype. It doesn’t get prod.