Good catch! github.com/rubys/showca...

Fixed. Thanks!

I don't care about other real world apps, each app is different. I care only about my own. I'm even fine making changes to my app to take advantage of frozen strings. I will try to find some time tomorrow to repeat and vary this test, and will post my results.

I'm not clear what you mean by "both environments". I created two VMs on the same platform (fly.io) where the only differences was an environment variable. I'm open to trying multiple requests, but skeptical of synthetic benchmarks and arguments based on knowledge of the implementation.

Don't fully remember, but at the time we were using Python 2, so strings were bytes not unicode. Looks like a buggy attempt to limit the strings to displayable ASCII.

Ultimately what Kamal/Capistrano/Ansible etc. do is ssh into servers and run commands. Kamal proxy handles the zero downtime deployments. You can run Kamal proxy on fly.io; triggering updates could be as simple as a webhook that runs git pull; prisma migrate deploy; kamal-proxy deploy

I'm assuming you have a volume? If so, yes Kamal's zero downtime deployment of apps with volumes is sweet and effortless. Can you explain what you mean by auto-update? Is this different than a GitHub action?

Give this a try: fly.io/sveltekit/ , if you have questions let me know. Fly.io is easier to get started with, but once you have a working Dockerfile you can switch back and forth. If you want to switch to Kamal, my docs will walk you through what you need to do.