Enable expanding root disk root_disk_size_gb

Applications such as Nvidia's libraries, required for many AI / LLM
applications, aren't easily installed on anywhere but the root disk;
however, the amount of free space in the root disk of a Jammy stemcell
is too small to accommodate these libraries.

To address this, we introduce a new property, `root_disk_size_gb`.
Setting this property will expand the root disk during VM creation (in
the `create_vm` CPI method, before the BOSH packages are installed).

Typical use (in a Cloud Config):

```yaml
vm_extensions:
- name: 20G_root
  cloud_properties:
    root_disk_size_gb: 20
```

Setting this property disables the VM's "linked clone" feature [0],
which is incompatible with extending the size of the root disk. The VM
will therefore require more disk space than expected, typically the size
of the root disk (in the above example, 20 GiB)

Adding this feature brings the vSphere CPI to parity with the AWS,
Azure, and GCP CPIs.

Root disk size is measured in GiB (1,073,741,824 bytes), not in GB
(1,000,000,000 bytes).

Drive-by:

Removed unnecessary initialization, `config_spec.device_change = []`:
`ConfigSpec` class initializes `device_change` to an empty array; we
don't need to do it.

[0] A linked clone is a copy of a virtual machine that shares virtual
disks with the parent virtual machine. Only changes from the parent disk
are recorded, saving on disk space.

Signed-off-by: Brian Cunnie <[email protected]>

src/vsphere_cpi/lib/cloud/vsphere/vm_config.rb

@@ -76,6 +76,10 @@ def pci_passthroughs
       vm_type.pci_passthroughs || []
     end
 
+    def root_disk_size_gb
+      vm_type.root_disk_size_gb.to_i
+    end
+
     def storage_policy_name
       @manifest_params[:storage_policy]
     end

src/vsphere_cpi/lib/cloud/vsphere/vm_creator.rb

@@ -183,6 +183,8 @@ def create(vm_config)
 
           # Clone VM
           logger.info("Cloning vm: #{replicated_stemcell_vm} to #{vm_config.name}")
+          # Don't link clone if expanding root disk, otherwise "Invalid operation for device '0'. Disks with parents cannot be expanded."
+          linked = vm_config.root_disk_size_gb > 0 ? false : true
           created_vm_mob = @client.wait_for_task do
             @cpi.clone_vm(
               replicated_stemcell_vm.mob,
@@ -191,7 +193,7 @@ def create(vm_config)
               cluster.resource_pool.mob,
               datastore: datastore.mob,
               host:,
-              linked: true,
+              linked: linked,
               snapshot: snapshot.current_snapshot,
               config: config_spec,
               datastore_cluster:
@@ -251,7 +253,6 @@ def create(vm_config)
             # Jammy stemcell at hardware version 13 only allows 1 vGPU; we want to be able to add more
             unless vm_config.vgpus.empty?
               config_spec = VimSdk::Vim::Vm::ConfigSpec.new
-              config_spec.device_change = []
               vm_config.vgpus.each do |vgpu|
                 vgpu = Resources::PCIPassthrough.create_vgpu(vgpu)
                 vgpu_config = Resources::VM.create_add_device_spec(vgpu)
@@ -261,7 +262,6 @@ def create(vm_config)
             end
             unless vm_config.pci_passthroughs.empty?
               config_spec = VimSdk::Vim::Vm::ConfigSpec.new
-              config_spec.device_change = []
               vm_config.pci_passthroughs.each do |pci_passthrough|
                 virtual_pci_passthrough = Resources::PCIPassthrough.create_pci_passthrough(
                   vendor_id: pci_passthrough['vendor_id'],
@@ -271,6 +271,15 @@ def create(vm_config)
               end
               @client.reconfig_vm(created_vm_mob, config_spec)
             end
+
+            if vm_config.root_disk_size_gb > 0
+              device_spec = Resources::VM.create_edit_device_spec(created_vm.system_disk)
+              device_spec.device.capacity_in_kb = vm_config.root_disk_size_gb * 2 ** 20 # GiB → kiB
+              config_spec = VimSdk::Vim::Vm::ConfigSpec.new
+              config_spec.device_change << device_spec
+              @client.reconfig_vm(created_vm_mob, config_spec)
+            end
+
             # DRS Rules
             create_drs_rules(vm_config, created_vm.mob, cluster)
 

src/vsphere_cpi/lib/cloud/vsphere/vm_type.rb

@@ -27,6 +27,7 @@ def initialize(datacenter, cloud_properties, pbm)
       nsxt
       pci_passthroughs
       ram
+      root_disk_size_gb
       storage_policy
       tags
       upgrade_hw_version

src/vsphere_cpi/spec/integration/root_size_disk_gb_spec.rb

@@ -0,0 +1,37 @@
+require 'integration/spec_helper'
+
+describe 'root_disk_size_gb property' do
+
+  let(:vm_type) do
+    {
+      'ram' => 512,
+      'disk' => 2048,
+      'cpu' => 1,
+    }
+  end
+
+  context 'when "root_disk_size_gb" is not set' do
+    it 'creates a VM whose system disk is a linked-clone to the stemcell' do
+      simple_vm_lifecycle(@cpi, @vlan, vm_type) do |vm_id|
+        vm = @cpi.vm_provider.find(vm_id)
+        stemcell = @cpi.vm_provider.find(@stemcell_id)
+        system_disk = vm.system_disk
+        stemcell_disk = stemcell.system_disk
+        expect(system_disk.backing.parent.uuid).to eq(stemcell_disk.backing.parent.uuid)
+      end
+    end
+  end
+
+  context 'when "root_disk_size_gb" is set' do
+    let(:root_disk_size_gb) { 15 }
+    it 'creates a VM whose system disk is a linked-clone to the stemcell' do
+      vm_type['root_disk_size_gb'] = root_disk_size_gb
+      simple_vm_lifecycle(@cpi, @vlan, vm_type) do |vm_id|
+        vm = @cpi.vm_provider.find(vm_id)
+        system_disk = vm.system_disk
+        expect(system_disk.backing.parent).to be_nil # no parent disk, not a linked-clone
+        expect(system_disk.capacity_in_kb / 1024 / 1024).to eq(root_disk_size_gb) # convert kiB → GiB
+      end
+    end
+  end
+end

src/vsphere_cpi/spec/unit/cloud/vsphere/vm_creator_spec.rb

@@ -21,7 +21,15 @@ module VSphereCloud
       )
     }
     let(:agent_env) { instance_double('VSphereCloud::AgentEnv')  }
-    let(:config) {  [ 'default_disk_type'=> default_disk_type, 'resource_pool' => 'test',  'datacenters'=> [ datacenter ],'host' => 'localhost', 'user' => 'admin', 'password' => 'password' ] }
+    let(:config) {
+      [
+        default_disk_type: default_disk_type,
+        resource_pool: 'test',
+        datacenters: [datacenter],
+        host: 'localhost',
+        user: 'admin',
+        password: 'password'
+      ] }
     let(:cpi_config) do
       instance_double(
         'VSphereCloud::Config',
@@ -36,6 +44,13 @@ module VSphereCloud
     let(:custom_fields_manager) { instance_double('VimSdk::Vim::CustomFieldsManager') }
     let(:datacenter) { {  mob: datacenter_mob , name: 'dc-1', 'name' => 'dc-1', 'persistent_datastore_pattern' => 'ds-ps-*', 'datastore_pattern'=> 'ds-*', 'vm_folder'=> 'bosh_vms', 'template_folder'=> 'stemcells', 'disk_path' => 'disks', 'clusters' => [] } }
     let(:datacenter_mob) { instance_double('VimSdk::Vim::Datacenter', name: 'dc-1') }
+    let(:cluster) { instance_double(Resources::Cluster, mob: cluster_mob,
+                                    resource_pool: resource_pool, host_group: nil, accessible_datastores: { "ds-1": {}, "ds-2": {} }) }
+    let(:cluster_mob) { instance_double('VimSdk::Vim::ClusterComputeResource') }
+    let(:resource_pool) { instance_double(Resources::ResourcePool, mob: resource_pool_mob) }
+    let(:resource_pool_mob) { instance_double(VimSdk::Vim::ResourcePool) }
+    let(:datastore) { instance_double(Resources::Datastore, name: 'ds-1', mob: datastore_mob) }
+    let(:datastore_mob) { instance_double(VimSdk::Vim::Datastore) }
     let(:default_disk_type) { 'preallocated' }
     let(:ip_conflict_detector) { instance_double(IPConflictDetector, ensure_no_conflicts: nil) }
     let(:ensure_no_ip_conflicts) { true }
@@ -162,13 +177,13 @@ module VSphereCloud
 
                                       cluster_placements: [
                                         instance_double(VmPlacement,
-                                                                                 cluster: instance_double(Resources::Cluster, host_group: nil, mob: nil,  accessible_datastores: {"ds-1": {}, "ds-2": {}}),
-                                                        fallback_disk_placements: [instance_double(Resources::Datastore, name: "ds-2")],
-                                                        disk_placement: instance_double(Resources::Datastore, name: "ds-1")
+                                                        cluster: cluster,
+                                                        disk_placement: datastore,
+                                                        fallback_disk_placements: [instance_double(Resources::Datastore, name: "ds-2")]
                                                        )
                                       ],
-                                      stemcell_cid: 'here-stemcell-cid'
-
+                                      stemcell_cid: 'here-stemcell-cid',
+                                      root_disk_size_gb: 0
                                      )
     }
 
@@ -199,7 +214,21 @@ module VSphereCloud
         expect(client).to receive(:upgrade_vm_virtual_hardware).with(cloned_vm_mob)
         expect(client).to receive(:reconfig_vm).with(cloned_vm_mob, anything)
         subject.create(vm_config)
-        # expect{subject.create(vm_config)}.not_to raise_exception
+      end
+    end
+
+    context 'with root_disk_size_gb set to 15 GiB' do
+      let(:system_disk) { instance_double('VimSdk::Vim::Vm::Device::VirtualDisk') }
+      let(:device_spec) { instance_double(VimSdk::Vim::Vm::Device::VirtualDeviceSpec, device: system_disk) }
+      let(:new_disk_size_gb) { 15 }
+      before do
+        allow(vm_config).to receive(:root_disk_size_gb).and_return(new_disk_size_gb)
+        allow(Resources::VM).to receive(:create_edit_device_spec).and_return(device_spec)
+      end
+      it "reconfigures the VM with a larger root disk and we don't check linked clones because the mocking is already too much" do
+        expect(system_disk).to receive(:capacity_in_kb=).with(new_disk_size_gb * 2 ** 20) # convert kiB → GiB
+        expect(client).to receive(:reconfig_vm).with(cloned_vm_mob, anything)
+        subject.create(vm_config)
       end
     end
 
@@ -243,13 +272,13 @@ module VSphereCloud
 
                                           cluster_placements: [
                                             instance_double(VmPlacement,
-                                                            cluster: instance_double(Resources::Cluster, host_group: nil, mob: nil,  accessible_datastores: {"ds-1": {}, "ds-2": {}}),
+                                                            cluster: cluster,
                                                             fallback_disk_placements: [],
                                                             disk_placement: instance_double(Resources::Datastore, name: "ds-1")
                                                            )
                                           ],
-                                          stemcell_cid: 'here-stemcell-cid'
-
+                                          stemcell_cid: 'here-stemcell-cid',
+                                          root_disk_size_gb: 0
                                          )
         }
         it 'still fails if there are no viable fallback_disk_placements' do