The symptom occurs only in the following condition:
Running multiple 2016 servers with RDS enabled (servers without RDS are not affected as far as we see)
There are more total vCPU cores assigned to all virtual servers then there are pCPU cores in the host. (spreading the load when needed.) which is normal.
When the pCPU load goes to 100% than all RDS 2016 guest servers will go to 100% and will not go down.
Windows guest shows 100% CPU load, even when there is only 1 user logged on, not using any resources, Task manager shows total of 100%, but user load is around 2% (does not add up)
This condition causes all RDS machines to freeze up, even when the pCPU load drops, the vCPU is still 100%.
Servers without RDS enabled on the same machine are not affected and continue to function.
We solved this for now to match the pCPU cores to the total of vCPU cores assigned to the VM's.
This issue is preventing loadspreading of the vCPU's across multiple RDS server.
Server 2016 guests.
3 with RDS enabled, 1 without RDS enabled.
1 hardware unit.
dual CPU Xeon E5-2620 v4.
Different customer with the same CPU has the same issue.