Did Sandstorm actually crash from things like "port already in use" before? Or did it go into an infinite loop of trying to start node and failing?

It would actually crash, which has the upside that systemd monitoring shows the process as FAILED. You can repro by setting PORT=22 on a Sandstorm on a systemd Sandstorm install.

For advanced sysadmins, having systemd be aware of failure may be useful, but for users like me who know nothing about systemd, it's most useful if Sandstorm tries its best to self-recover. Consider a user who launched a VM with a one-click launcher and doesn't know how to open a shell, much less use it.

That said, going into emergency update mode when the port is in-use does not seem right...

Perhaps we should hold off on this until someone has time to change it so that update checks happen more often in this mode with "type=emerg" or something, so that we can get alerts when this is happening.