Unable to apply latency-sensitivity setting to virtual machine <VM name>. No valid placement on the host
."Latency Sensitivity
" is powered on the same hostYYYY-MM-DDTHH:MM:SS cpu18:xxxxxxxx)WARNING: CpuSched: 311: vcpu xxxxxxxx is placed in adoption mode since all pcpus in its affinitySet 0xffffffffff have exclusive affinity to other vcpus
YYYY-MM-DDTHH:MM:SS cpu19:xxxxxxxx)WARNING: CpuSched: 311: vcpu xxxxxxxx is placed in adoption mode since all pcpus in its affinitySet 0xffffffffff have exclusive affinity to other vcpus
YYYY-MM-DDTHH:MM:SS cpu19:xxxxxxxx)WARNING: CpuSched: 311: vcpu xxxxxxxx is placed in adoption mode since all pcpus in its affinitySet 0xffffffffff have exclusive affinity to other vcpus
YYYY-MM-DDTHH:MM:SS cpu19:xxxxxxxx)WARNING: CpuSched: 311: vcpu xxxxxxxx is placed in adoption mode since all pcpus in its affinitySet 0x80000 have exclusive affinity to other vcpus
YYYY-MM-DDTHH:MM:SS cpu12:xxxxxxxx)WARNING: CpuSched: 1324: Unable to apply latency-sensitivity setting to virtual machine <vm name>. No valid placement on the host.
YYYY-MM-DDTHH:MM:SS cpu2:xxxxxxxx)CpuSched: vm xxxxxxxx: 2764: unset exclusive affinity to 2
YYYY-MM-DDTHH:MM:SS cpu18:xxxxxxxx)CpuSched: vm xxxxxxxx: 2764: unset exclusive affinity to 18
YYYY-MM-DDTHH:MM:SS cpu11:xxxxxxxx)CpuSched: vm xxxxxxxx: 2764: unset exclusive affinity to 11
YYYY-MM-DDTHH:MM:SS cpu1:xxxxxxxx)CpuSched: vm xxxxxxxx: 2764: unset exclusive affinity to 1
High Latency Sensitivity requires to set CPU and 100% memory reservation for the VM and each virtual CPU is granted exclusive access to a physical core. In case if the configured cores for VM is not available in a single NUMA node, the latency-sensitivity setting would default to Normal to ensure successful VM poweron operation.
This issue is caused due to unavailability of non-exclusive cores to give to the VM as exclusive cores because NUMA client(s) of the VM required more exclusive cores than were available on a particular NUMA node.
In order to resolve the issue, split the VM into multiple NUMA clients and place NUMA clients on different NUMA nodes so that there is enough non-exclusive cores to give each NUMA client from each NUMA node.
Perform the below steps to ensure VM is able to power-on with exclusive core affinity across sockets
numa.vcpu.maxPerClient = "6"
<- enforces NUMA clients to be capped to a certain size
numa.consolidate = "FALSE"
<- enforces that NUMA clients are spread across NUMA nodes
Note: This reference is based on the below configuration of the host and needs to be adjusted as per the hardware present on host. In the above example, the NUMA clients are spread across NUMA nodes where 6 clients(VM cores) would be affinitized per NUMA node(socket).
So the value for numa.vcpu.maxPerClient
= (number of vCPU on VM)/(number of sockets on host)
Host CPU:
Number of Sockets: 2
Number of cores (total): 40 (20 cores per socket) <- HyperThreading enabled
VM CPU:
Number of vCPU: 12
In order to validate the CPU affinity, execute the below command via ssh session on the related ESXi host
[root@esxi:~]
tail -f /var/log/vmkernel.log | grep "exclusive affinity"
YYYY-MM-DDTHH:MM:SS vmkernel: cpu6:891115)CpuSched: vm xxxxxx: 2813: unset exclusive affinity to 6
YYYY-MM-DDTHH:MM:SS vmkernel: cpu8:891522)CpuSched: vm xxxxxx: 2813: set exclusive affinity to 8
YYYY-MM-DDTHH:MM:SS vmkernel: cpu15:892677)CpuSched: vm xxxxxx: 2813: set exclusive affinity to 15
YYYY-MM-DDTHH:MM:SS vmkernel: cpu10:893175)CpuSched: vm xxxxxx: 2813: set exclusive affinity to 10
YYYY-MM-DDTHH:MM:SS vmkernel: cpu11:893173)CpuSched: vm xxxxxx: 2813: set exclusive affinity to 11
[root@esxi:~]
tail -f /var/log/vmkernel.log | grep "exclusive affinity"
YYYY-MM-DDTHH:MM:SS cpu12:26997166)WARNING: CpuSched: 1324: Unable to apply latency-sensitivity setting to virtual machine <VM name>. No valid placement on the host.
YYYY-MM-DDTHH:MM:SS cpu2:26997157)CpuSched: vm xxxxxxxx: 2764: unset exclusive affinity to 2
YYYY-MM-DDTHH:MM:SS cpu18:26997167)CpuSched: vm xxxxxxxx: 2764: unset exclusive affinity to 18
YYYY-MM-DDTHH:MM:SS cpu11:26997162)CpuSched: vm xxxxxxxx: 2764: unset exclusive affinity to 11
YYYY-MM-DDTHH:MM:SS cpu1:26997163)CpuSched: vm xxxxxxxx: 2764: unset exclusive affinity to 1
YYYY-MM-DDTHH:MM:SS cpu34:26997161)CpuSched: vm xxxxxxxx: 2764: unset exclusive affinity to 34
YYYY-MM-DDTHH:MM:SS cpu39:26997153)CpuSched: vm xxxxxxxx: 2764: unset exclusive affinity to 39
These additional commands can be executed via ssh session on the ESXi host for detailed information
[root@esxi:~] sched-stats -t cpu -z "name:group:cpu:mode:affinity" | awk 'NR==1 || /vmx-vcpu/'
[root@esxi:~] sched-stats -t numa-pnode
[root@esxi:~] sched-stats -t numa-clients