Setup limits for GPU

Author: vishesh92

.github/workflows/ci.yml

@@ -137,6 +137,7 @@ jobs:
                   smoke/test_vm_deployment_planner
                   smoke/test_vm_strict_host_tags
                   smoke/test_vm_schedule
+                  smoke/test_deploy_vgpu_enabled_vm
                   smoke/test_vm_life_cycle
                   smoke/test_vm_lifecycle_unmanage_import
                   smoke/test_vm_snapshot_kvm

agent/src/main/java/com/cloud/agent/properties/AgentProperties.java

@@ -213,6 +213,15 @@ public class AgentProperties{
      */
     public static final Property<String> AGENT_HOOKS_LIBVIRT_VM_XML_TRANSFORMER_SCRIPT = new Property<>("agent.hooks.libvirt_vm_xml_transformer.script", "libvirt-vm-xml-transformer.groovy");
 
+    /**
+     * This property is used with the agent.hooks.basedir property to define the Libvirt VM XML transformer shell script.<br>
+     * The shell script is used to execute the Libvirt VM XML transformer script.<br>
+     * For more information see the agent.properties file.<br>
+     * Data type: String.<br>
+     * Default value: <code>libvirt-vm-xml-transformer.sh</code>
+     */
+    public static final Property<String> AGENT_HOOKS_LIBVIRT_VM_XML_TRANSFORMER_SHELL_SCRIPT = new Property<>("agent.hooks.libvirt_vm_xml_transformer.shell_script", "libvirt-vm-xml-transformer.sh");
+
     /**
      * This property is used with the agent.hooks.basedir and agent.hooks.libvirt_vm_xml_transformer.script properties to define the Libvirt VM XML transformer method.<br>
      * Libvirt XML transformer hook does XML-to-XML transformation.<br>
@@ -233,6 +242,15 @@ public class AgentProperties{
      */
     public static final Property<String> AGENT_HOOKS_LIBVIRT_VM_ON_START_SCRIPT = new Property<>("agent.hooks.libvirt_vm_on_start.script", "libvirt-vm-state-change.groovy");
 
+    /**
+     * This property is used with the agent.hooks.basedir property to define the Libvirt VM on start shell script.<br>
+     * The shell script is used to execute the Libvirt VM on start script.<br>
+     * For more information see the agent.properties file.<br>
+     * Data type: String.<br>
+     * Default value: <code>libvirt-vm-state-change.sh</code>
+     */
+    public static final Property<String> AGENT_HOOKS_LIBVIRT_VM_ON_START_SHELL_SCRIPT = new Property<>("agent.hooks.libvirt_vm_on_start.shell_script", "libvirt-vm-state-change.sh");
+
     /**
      * This property is used with the agent.hooks.basedir and agent.hooks.libvirt_vm_on_start.script properties to define the Libvirt VM on start method.<br>
      * The hook is called right after Libvirt successfully launched the VM.<br>
@@ -252,6 +270,15 @@ public class AgentProperties{
      */
     public static final Property<String> AGENT_HOOKS_LIBVIRT_VM_ON_STOP_SCRIPT = new Property<>("agent.hooks.libvirt_vm_on_stop.script", "libvirt-vm-state-change.groovy");
 
+    /**
+     * This property is used with the agent.hooks.basedir property to define the Libvirt VM on stop shell script.<br>
+     * The shell script is used to execute the Libvirt VM on stop script.<br>
+     * For more information see the agent.properties file.<br>
+     * Data type: String.<br>
+     * Default value: <code>libvirt-vm-state-change.sh</code>
+     */
+    public static final Property<String> AGENT_HOOKS_LIBVIRT_VM_ON_STOP_SHELL_SCRIPT = new Property<>("agent.hooks.libvirt_vm_on_stop.shell_script", "libvirt-vm-state-change.sh");
+
     /**
      * This property is used with the agent.hooks.basedir and agent.hooks.libvirt_vm_on_stop.script properties to define the Libvirt VM on stop method.<br>
      * The hook is called right after libvirt successfully stopped the VM.<br>

api/src/main/java/com/cloud/agent/api/VgpuTypesInfo.java

@@ -40,6 +40,7 @@ public class VgpuTypesInfo {
     private Long maxVgpuPerGpu;
     private Long remainingCapacity;
     private Long maxCapacity;
+    private boolean display = false;
 
     public String getModelName() {
         return modelName;
@@ -53,18 +54,34 @@ public Long getVideoRam() {
         return videoRam;
     }
 
+    public void setVideoRam(Long videoRam) {
+        this.videoRam = videoRam;
+    }
+
     public Long getMaxHeads() {
         return maxHeads;
     }
 
+    public void setMaxHeads(Long maxHeads) {
+        this.maxHeads = maxHeads;
+    }
+
     public Long getMaxResolutionX() {
         return maxResolutionX;
     }
 
+    public void setMaxResolutionX(Long maxResolutionX) {
+        this.maxResolutionX = maxResolutionX;
+    }
+
     public Long getMaxResolutionY() {
         return maxResolutionY;
     }
 
+    public void setMaxResolutionY(Long maxResolutionY) {
+        this.maxResolutionY = maxResolutionY;
+    }
+
     public Long getMaxVpuPerGpu() {
         return maxVgpuPerGpu;
     }
@@ -177,6 +194,14 @@ public void setVmName(String vmName) {
         this.vmName = vmName;
     }
 
+    public boolean isDisplay() {
+        return display;
+    }
+
+    public void setDisplay(boolean display) {
+        this.display = display;
+    }
+
     public VgpuTypesInfo(GpuDevice.DeviceType deviceType, String groupName, String modelName, String busAddress,
             String vendorId, String vendorName, String deviceId, String deviceName, String numaNode, String pciRoot
     ) {

api/src/main/java/com/cloud/capacity/Capacity.java

@@ -33,7 +33,7 @@ public interface Capacity extends InternalIdentity, Identity {
     public static final short CAPACITY_TYPE_DIRECT_ATTACHED_PUBLIC_IP = 8;
     public static final short CAPACITY_TYPE_LOCAL_STORAGE = 9;
     public static final short CAPACITY_TYPE_VIRTUAL_NETWORK_IPV6_SUBNET = 10;
-    public static final short CAPACITY_TYPE_GPU = 19;
+    public static final short CAPACITY_TYPE_GPU = 11;
 
     public static final short CAPACITY_TYPE_CPU_CORE = 90;
 

api/src/main/java/com/cloud/user/ResourceLimitService.java

@@ -50,6 +50,12 @@ public interface ResourceLimitService {
                 "The default maximum number of projects that can be created for an account",false);
     static final ConfigKey<Long> DefaultMaxDomainProjects = new ConfigKey<>("Domain Defaults",Long.class,"max.domain.projects","50",
                         "The default maximum number of projects that can be created for a domain",false);
+    static final ConfigKey<Long> DefaultMaxAccountGpus = new ConfigKey<>("Account Defaults",Long.class,"max.account.gpus","20",
+            "The default maximum number of GPU devices that can be used for an account", false);
+    static final ConfigKey<Long> DefaultMaxDomainGpus = new ConfigKey<>("Domain Defaults",Long.class,"max.domain.gpus","20",
+            "The default maximum number of GPU devices that can be used for a domain", false);
+    static final ConfigKey<Long> DefaultMaxProjectGpus = new ConfigKey<>("Project Defaults",Long.class,"max.project.gpus","20",
+            "The default maximum number of GPU devices that can be used for a project", false);
 
     static final List<ResourceType> HostTagsSupportingTypes = List.of(ResourceType.user_vm, ResourceType.cpu, ResourceType.memory, ResourceType.gpu);
     static final List<ResourceType> StorageTagsSupportingTypes = List.of(ResourceType.volume, ResourceType.primary_storage);
@@ -284,4 +290,8 @@ void checkVmResourceLimitsForTemplateChange(Account owner, Boolean display, Serv
     void incrementVmMemoryResourceCount(long accountId, Boolean display, ServiceOffering serviceOffering, VirtualMachineTemplate template, Long memory);
     void decrementVmMemoryResourceCount(long accountId, Boolean display, ServiceOffering serviceOffering, VirtualMachineTemplate template, Long memory);
 
+    void checkVmGpuResourceLimit(Account owner, Boolean display, ServiceOffering serviceOffering, VirtualMachineTemplate template, Long gpu) throws ResourceAllocationException;
+    void incrementVmGpuResourceCount(long accountId, Boolean display, ServiceOffering serviceOffering, VirtualMachineTemplate template, Long gpu);
+    void decrementVmGpuResourceCount(long accountId, Boolean display, ServiceOffering serviceOffering, VirtualMachineTemplate template, Long gpu);
+
 }

api/src/main/java/org/apache/cloudstack/api/ApiConstants.java

@@ -390,7 +390,9 @@ public class ApiConstants {
     public static final String GPU_CARD_ID = "gpucardid";
     public static final String GPU_CARD_NAME = "gpucardname";
     public static final String GPU_COUNT = "gpucount";
+    public static final String GPU_DISPLAY = "gpudisplay";
     public static final String GPU_DEVICE_TYPE = "gpudevicetype";
+    public static final String GPU_ENABLED = "gpuenabled";
     public static final String MAX_VGPU_PER_PHYSICAL_GPU = "maxvgpuperphysicalgpu";
     public static final String GUEST_OS_LIST = "guestoslist";
     public static final String GUEST_OS_COUNT = "guestoscount";

api/src/main/java/org/apache/cloudstack/api/BaseCmd.java

@@ -39,6 +39,7 @@
 import org.apache.cloudstack.alert.AlertService;
 import org.apache.cloudstack.annotation.AnnotationService;
 import org.apache.cloudstack.context.CallContext;
+import org.apache.cloudstack.gpu.GpuService;
 import org.apache.cloudstack.network.RoutedIpv4Manager;
 import org.apache.cloudstack.network.lb.ApplicationLoadBalancerService;
 import org.apache.cloudstack.network.lb.InternalLoadBalancerVMService;
@@ -130,6 +131,8 @@ public static enum CommandType {
     @Inject
     public UserVmService _userVmService;
     @Inject
+    public GpuService gpuService;
+    @Inject
     public ManagementService _mgr;
     @Inject
     public StorageService _storageService;

api/src/main/java/org/apache/cloudstack/api/command/admin/gpu/CreateGpuCardCmd.java

@@ -30,18 +30,13 @@
 import org.apache.cloudstack.api.ServerApiException;
 import org.apache.cloudstack.api.response.GpuCardResponse;
 import org.apache.cloudstack.gpu.GpuCard;
-import org.apache.cloudstack.gpu.GpuService;
 
-import javax.inject.Inject;
 
 @APICommand(name = "createGpuCard", description = "Creates a GPU card definition in the system",
-        responseObject = GpuCardResponse.class,
-        requestHasSensitiveInfo = false, responseHasSensitiveInfo = false, since = "4.21.0")
+        responseObject = GpuCardResponse.class, requestHasSensitiveInfo = false, responseHasSensitiveInfo = false,
+        since = "4.21.0")
 public class CreateGpuCardCmd extends BaseCmd {
 
-    @Inject
-    private GpuService gpuService;
-
     /// //////////////////////////////////////////////////
     /// ///////////// API parameters /////////////////////
     /// //////////////////////////////////////////////////
@@ -54,8 +49,8 @@ public class CreateGpuCardCmd extends BaseCmd {
             description = "the device name of the GPU card")
     private String deviceName;
 
-    @Parameter(name = ApiConstants.NAME, type = CommandType.STRING, required = true, description
-            = "the display name of the GPU card")
+    @Parameter(name = ApiConstants.NAME, type = CommandType.STRING, required = true,
+            description = "the display name of the GPU card")
     private String name;
 
     @Parameter(name = ApiConstants.VENDOR_NAME, type = CommandType.STRING, required = true,
@@ -66,6 +61,11 @@ public class CreateGpuCardCmd extends BaseCmd {
             description = "the vendor ID of the GPU card")
     private String vendorId;
 
+    // Optional parameters for the passthrough vGPU profile display properties
+    @Parameter(name = ApiConstants.VIDEORAM, type = CommandType.LONG,
+            description = "the video RAM size in MB for the passthrough vGPU profile")
+    private Long videoRam;
+
     /// //////////////////////////////////////////////////
     /// //////////////// Accessors ///////////////////////
     /// //////////////////////////////////////////////////
@@ -90,23 +90,24 @@ public String getVendorId() {
         return vendorId;
     }
 
+    public Long getVideoRam() {
+        return videoRam;
+    }
+
     @Override
-    public void execute() throws ResourceUnavailableException, InsufficientCapacityException,
-            ServerApiException, ConcurrentOperationException,
-            ResourceAllocationException, NetworkRuleConflictException {
+    public void execute() throws ResourceUnavailableException, InsufficientCapacityException, ServerApiException,
+            ConcurrentOperationException, ResourceAllocationException, NetworkRuleConflictException {
         try {
             GpuCard gpuCard = gpuService.createGpuCard(this);
             if (gpuCard != null) {
                 GpuCardResponse response = new GpuCardResponse(gpuCard);
                 response.setResponseName(getCommandName());
                 setResponseObject(response);
             } else {
-                throw new ServerApiException(ApiErrorCode.INTERNAL_ERROR,
-                        "Failed to create GPU card");
+                throw new ServerApiException(ApiErrorCode.INTERNAL_ERROR, "Failed to create GPU card");
             }
         } catch (Exception e) {
-            throw new ServerApiException(ApiErrorCode.INTERNAL_ERROR,
-                    "Failed to create GPU card: " + e.getMessage());
+            throw new ServerApiException(ApiErrorCode.INTERNAL_ERROR, "Failed to create GPU card: " + e.getMessage());
         }
     }
 

api/src/main/java/org/apache/cloudstack/api/command/admin/gpu/CreateGpuDeviceCmd.java

@@ -17,6 +17,7 @@
 package org.apache.cloudstack.api.command.admin.gpu;
 
 import com.cloud.user.Account;
+import org.apache.cloudstack.acl.RoleType;
 import org.apache.cloudstack.api.APICommand;
 import org.apache.cloudstack.api.ApiConstants;
 import org.apache.cloudstack.api.ApiErrorCode;
@@ -28,73 +29,44 @@
 import org.apache.cloudstack.api.response.HostResponse;
 import org.apache.cloudstack.api.response.VgpuProfileResponse;
 import org.apache.cloudstack.gpu.GpuDevice;
-import org.apache.cloudstack.gpu.GpuService;
 import org.apache.commons.lang3.EnumUtils;
 import org.apache.commons.lang3.StringUtils;
 
-import javax.inject.Inject;
 
-@APICommand(name = "createGpuDevice",
-        description = "Creates a GPU device manually on a host",
-        responseObject = GpuDeviceResponse.class,
-        since = "4.21.0",
-        requestHasSensitiveInfo = false,
-        responseHasSensitiveInfo = false)
+@APICommand(name = "createGpuDevice", description = "Creates a GPU device manually on a host",
+            responseObject = GpuDeviceResponse.class, since = "4.21.0", requestHasSensitiveInfo = false,
+            responseHasSensitiveInfo = false, authorized = {RoleType.Admin})
 public class CreateGpuDeviceCmd extends BaseCmd {
 
-    @Inject
-    private GpuService gpuService;
-
-    @Parameter(name = ApiConstants.HOST_ID,
-            type = CommandType.UUID,
-            entityType = HostResponse.class,
-            required = true,
-            description = "ID of the host where the GPU device is located")
+    @Parameter(name = ApiConstants.HOST_ID, type = CommandType.UUID, entityType = HostResponse.class, required = true,
+               description = "ID of the host where the GPU device is located")
     private Long hostId;
 
-    @Parameter(name = ApiConstants.BUS_ADDRESS,
-            type = CommandType.STRING,
-            required = true,
-            description = "PCI bus address of the GPU device (e.g., 0000:01:00.0) or UUID for MDEV devices.")
+    @Parameter(name = ApiConstants.BUS_ADDRESS, type = CommandType.STRING, required = true,
+               description = "PCI bus address of the GPU device (e.g., 0000:01:00.0) or UUID for MDEV devices.")
     private String busAddress;
 
-    @Parameter(name = ApiConstants.GPU_CARD_ID,
-            type = CommandType.UUID,
-            entityType = GpuCardResponse.class,
-            required = true,
-            description = "ID of the GPU card type")
+    @Parameter(name = ApiConstants.GPU_CARD_ID, type = CommandType.UUID, entityType = GpuCardResponse.class,
+               required = true, description = "ID of the GPU card type")
     private Long gpuCardId;
 
-    @Parameter(name = ApiConstants.VGPU_PROFILE_ID,
-            type = CommandType.UUID,
-            entityType = VgpuProfileResponse.class,
-            required = true,
-            description = "ID of the vGPU profile")
+    @Parameter(name = ApiConstants.VGPU_PROFILE_ID, type = CommandType.UUID, entityType = VgpuProfileResponse.class,
+               required = true, description = "ID of the vGPU profile")
     private Long vgpuProfileId;
 
-    @Parameter(name = ApiConstants.TYPE,
-            type = CommandType.STRING,
-            description = "Type of GPU device (PCI, MDEV, VGPUOnly). Defaults to PCI.")
+    @Parameter(name = ApiConstants.TYPE, type = CommandType.STRING,
+               description = "Type of GPU device (PCI, MDEV, VGPUOnly). Defaults to PCI.")
     private String type;
 
-    @Parameter(name = ApiConstants.PARENT_GPU_DEVICE_ID,
-            type = CommandType.UUID,
-            entityType = GpuDeviceResponse.class,
-            description = "ID of the parent GPU device (for virtual GPU devices)")
+    @Parameter(name = ApiConstants.PARENT_GPU_DEVICE_ID, type = CommandType.UUID, entityType = GpuDeviceResponse.class,
+               description = "ID of the parent GPU device (for virtual GPU devices)")
     private Long parentGpuDeviceId;
 
-    @Parameter(name = ApiConstants.NUMA_NODE,
-            type = CommandType.STRING,
-            entityType = GpuDeviceResponse.class,
-            description = "NUMA node of the GPU device (e.g., 0, 1, etc.). This is optional and can be used to specify the NUMA node for the GPU device which is used during allocation. Defaults to -1")
+    @Parameter(name = ApiConstants.NUMA_NODE, type = CommandType.STRING,
+            description = "NUMA node of the GPU device (e.g., 0, 1, etc.). This is optional and can be used to "
+                          + "specify the NUMA node for the GPU device which is used during allocation. Defaults to -1")
     private String numaNode;
 
-    @Parameter(name = ApiConstants.PCI_ROOT,
-            type = CommandType.STRING,
-            entityType = GpuDeviceResponse.class,
-            description = "PCI root of the GPU device.")
-    private String pciRoot;
-
     public Long getHostId() {
         return hostId;
     }
@@ -133,10 +105,6 @@ public String getNumaNode() {
         return numaNode;
     }
 
-    public String getPciRoot() {
-        return pciRoot;
-    }
-
     @Override
     public void execute() {
         try {