Those errors may not actually be user facing. When there is more than one VM available, we will retry against another one. You’ll see an error in the region the VM is running in, and then the edge proxy will connect to a different VM.
If you look at metrics on fly.io/apps/<name>/metrics and don’t see 502s, everything is working as it should be.
You can’t control the delay between rolling restarts directly. You can catch the SIGINT and wait up to 5 minutes before exiting, though, which might help.