Dear Guys,
I have a problem at one of our customers and don't know what to do with that. Two weeks ago we had made an Exchange migration, they bought a brand new Dell server for that (16 CPU cores, 48GB memory), which was installed a month before. Windows Server 2016
Standard with only Hyper-V roles, another 2016 Standard as guest and one FreeBSD based VM. Before you install Exchange I've tried to install all the new updates for Windows. I've found the two VMs in a saved state and couldn't start them. "The change
of the (VMNAME) status was unsuccessfull, the server has no more memory." (I think it's not the exact error message in English, I've translated it, I have a Hungarian OS.)
So Exchange had 40GB of fixed memory, FreeBSD had 4GB, and another 4GB was reserved for the host OS. The Windows guest was empty as the FreeBSD too, so I had dropped their saved state without any hesitation to set other memory values. First I tried fixed
32GB for Exchange, then modified to 8-32GB dynamic settings. I've tried to restart Hyper-V VMM as well, but VMs didn't start with the same message. I couldn't start the Exchange, not the FreeBSD with only 4GB settings. I had tons of free memory, according
to resource manager. Finally, the host reboot had solved the problem, the question is that it is resolved permanently?
Today I have another error report from them. Same customer, another new Dell server from another series and more resources. Windows Server 2016 Standard Hyper-V host with 256GB of RAM and 2x12 CPUs (48 cores with HT), shared between 5 VMs (just Windows guests,
2008 Server, 2012 R2 and Windows 7), only ~30% of the server's resources are assigned to them. And the same error, saved state on all 5 VMs and couldn't start any of them. After a host restart they started correctly, and still running fine.
I've found lot of post about this kind of error on forums when NUMA is not enabled on the Hyper-V server, so I had checked them and NUMA is enabled on all of them. What can cause this problem on Server 2016's Hyper-V?