I have experienced this myself. I'm pretty sure that it is due to DNS caching and propagation speeds (or more to the point - slowness!). I have found that most ISP DNS don't update very often so although the initial propagation happens quite quickly, it is not updated anywhere near as quickly.

So when you first launch your server the initial DNS is propagated quite quickly. But when you restart your server it gets a new public IP and the propagation of the new association is quite slow. The cached DNS location of your old IP is still active, but as the IP is no longer associated with your appliance you get the HubDNS/tklapp.com message instead (because the old IP is no longer in use). In my experience it can take some time for the new IP to be associated with the domain name (but it does happen!).

One workaround (which is not perfect but seems to work much better) is instead of using your default ISP DNS, use a (better i.e. more quickly updated) public DNS for resolution. I find Google's public DNS (8.8.8.8) does a much better job.

Another (probably more satisfactory) way to go (especially if you have a production server that you don't want to have these sort of issues with) is to buy an AWS elastic IP. They are essentially like a static IP which you associate with your AWS server. However if you are going to do that and have your own domain (i.e. aren't using a free tklapp.com subdomain) then you can probably just adjust your nameserver records (with whoever your domain is through) to link directly to your elastic IP (and cut HubDNS out of it altogether).