When getting up and going on Judoscale I found that any time a replica was scaled down it was considered crashed, even though it was properly scaled down, it just wasn't exiting with a zero status, but rather 143.

When contacting Judoscale this was their response, so I'm reaching out to see if using raise_exception_on_sigterm is the accepted solution or if I'm missing something:

---

Thanks for reporting it. I remember hitting that exact same issue wth the 143 exit code from Puma back when we were building the integration, with my test application. And I came across that same somewhat hidden puma config: raise_exception_on_sigterm.

I had some back-and-forth with Railway back then, and tested the same app on Heroku which didn't show the issue. At the time Railway reviewed things on their side, and even built a sample app in Go with a 143 exit code on SIGTERM, but that app didn't show the same issue as mine, so they couldn't find anything wrong on their side (and they said they were handling 143 as expected iirc), and I couldn't find anything wrong on the Rails + Puma setup either.

Our main app was also running on Railway for a few weeks for our platform tour, and we apparently didn't seem to encounter that same issue for some reason.

I guess the TL;DR is: if you're seeing that same behavior, the only option I can think of is raise_exception_on_sigterm, indeed.