Change oversubscription settings for existing compute nodes

Available since MOSK 23.1

MOSK enables you to control the oversubscription of compute node resources through the placement service API.

To manage the oversubscription through the placement API:

1. Obtain the host name of the hypervisor in question:

   openstack hypervisor list -f yaml
   

   Example of system response:

   - Host IP: 10.10.0.78
     Hypervisor Hostname: ps-ps-obnqilm4xxlu-0-gdy3x46euaeu-server-ftp6p7j6pyjl.cluster.local
     Hypervisor Type: QEMU
     ID: 1
     State: up
   - Host IP: 10.10.0.118
     Hypervisor Hostname: ps-ps-obnqilm4xxlu-1-n36gax6zqgef-server-xtby2leuercd.cluster.local
     Hypervisor Type: QEMU
     ID: 7
     State: up
   

2. Determine the resource provider that corresponds to the hypervisor:

   openstack resource provider list -f yaml --name <hypervisor_hostname>
   

   Example of system response:

   - generation: 4
     name: ps-ps-obnqilm4xxlu-1-n36gax6zqgef-server-xtby2leuercd.cluster.local
     parent_provider_uuid: null
     root_provider_uuid: b16e9094-3f0e-4b8e-a138-e0b1f0a980db
     uuid: b16e9094-3f0e-4b8e-a138-e0b1f0a980db
   

3. Verify the current values in the resource provider by its UUID:

   openstack resource provider inventory list <provider_uuid> -f yaml
   

   Example of system response:

   - allocation_ratio: 8.0
     max_unit: 8
     min_unit: 1
     reserved: 0
     resource_class: VCPU
     step_size: 1
     total: 8
     used: 0
   - allocation_ratio: 1.0
     max_unit: 7956
     min_unit: 1
     reserved: 512
     resource_class: MEMORY_MB
     step_size: 1
     total: 7956
     used: 0
   - allocation_ratio: 1.6
     max_unit: 145
     min_unit: 1
     reserved: 0
     resource_class: DISK_GB
     step_size: 1
     total: 145
     used: 0
   

4. Update the allocation ratio for the required resource class in the resource provider and inspect the system response to verify that the change has been applied:

   openstack resource provider inventory set <provider_uuid> --amend --resource VCPU:allocation_ratio=10
   

   Caution

   To ensure accurate resource updates, it is crucial to specify the --amend argument when making requests. Failure to do so will require the inclusion of values for all fields associated with the resource provider.

   Example of system response:

   - allocation_ratio: 10.0
     max_unit: 8
     min_unit: 1
     reserved: 0
     resource_class: VCPU
     step_size: 1
     total: 8
     used: 0
   - allocation_ratio: 1.0
     max_unit: 7956
     min_unit: 1
     reserved: 512
     resource_class: MEMORY_MB
     step_size: 1
     total: 7956
     used: 0
   - allocation_ratio: 1.6
     max_unit: 145
     min_unit: 1
     reserved: 0
     resource_class: DISK_GB
     step_size: 1
     total: 145
     used: 0
   

See also

Reference Architecture: Resource oversubscription