I've seen similar behavior before. Have you tried to put the
host in
maintenance and once all VMs are moved away to reboot it ?
We did so last night. It _did_ fix the issue!
Shutting down the VMs on the host, putting the host in and out of
maintenance mode did not help. In fact it further illustrated the
problem.
It took the host from 94% memory used down to 50% memory used, even
though there was nothing running on the host at all, and really 99% of
memory was available.
Rebooting the host resolved the issue.
I'm going to apply latest updates to a cluster and see if the issues
persist.
This therefore sounds like a bug, which is quite bad. Unless there is
some communication issue from the engine to the host, which the reboot
assists with?