dfranke wrote:
db3l, that does mitigate the problem, yes. However, it's inadequate for more serious problems which require booting from a rescue system or just reimaging the node altogether. It also requires that I remember how to get to Lish without logging into the Manager to look it up

Well, you can boot into specific profiles (and list the ones you have) from the lish command prompt, so you should be able to do anything in that regard that you can do from Linode Manager. If the box is just failing to start fully, you'll have console/single user access from lish. I'm assuming you would have already set up a profile for the rescue environment (e.g., finnix boot against same disk images), which I think is a good policy, at least as far as getting the box basically back up and operational. If you hadn't, then the alternative would be the Linode API if you didn't have manager access.
I guess if you do end up at a point of the entire host being down or something else outside of your control you'll have a problem - of course, even with Linode Manager in that case you would likely not be able to interact with your failing Linode.
For remembering how to access, I'd suggest setting things up beforehand to simplify that. For example, what I have set up is a local /etc/hosts entry on my personal systems that provides an alias for the appropriate <datacenter>##.linode.com host address. Then, in my ~/.ssh/config I set up an automatic username for that alias which thus keeps track of my linode#### username. Set the appropriate ssh key for lish, and you're down to just "ssh <alias>" to get access to lish for your box at any time.
Since this is really just an emergency approach if you do have a full loss of service on your box, simultaneous with your source address changing for Linode Manager, the exposure window should be quite small to start with, and I'd think this provides enough tools to help ensure you could take care of things in that rare case.
Of course, I still don't disagree that additional whitelist management methods could be helpful - just not sure you're really risking a catch-22 in the current scenario.
Perhaps the Linode API could be expanded to permit updating the whitelist, which would probably be less back-end work then trying to integrate phone or SMS service.
-- David