I’ve had a few really strange Xen problems that I was only able to resolve by doing a rolling-reboot of the individual Xen hosts. This was a real pain in the ass, but now I’ve discovered a way to restart the Xen toolstack without affecting the running VMs:
xe-toolstack-restart
Note that you sometimes need to run this command on the master in your pool rather than the actual XenServer(s) having problems.