Fix capacity & UI enhancements

Author: vishesh92

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

@@ -402,6 +402,7 @@ public class ApiConstants {
     public static final String PARENT = "parent";
     public static final String PARENT_ID = "parentid";
     public static final String PARENT_DOMAIN_ID = "parentdomainid";
+    public static final String PARENT_GPU_DEVICE_ID = "parentgpudeviceid";
     public static final String PARENT_SUBNET = "parentsubnet";
     public static final String PARENT_TEMPLATE_ID = "parenttemplateid";
     public static final String PASSWORD = "password";

api/src/main/java/org/apache/cloudstack/api/response/GpuDeviceResponse.java

@@ -70,6 +70,10 @@ public class GpuDeviceResponse extends BaseResponse {
     @Param(description = "the vGPU profile name assigned to this GPU device")
     private GpuDevice.State state;
 
+    @SerializedName(ApiConstants.PARENT_GPU_DEVICE_ID)
+    @Param(description = "the ID of the parent GPU device, if this is a vGPU")
+    private String parentGpuDeviceId;
+
 
     public GpuDeviceResponse() {
         // Empty constructor for serialization
@@ -163,4 +167,12 @@ public GpuDevice.State getState() {
     public void setState(GpuDevice.State state) {
         this.state = state;
     }
+
+    public String getParentGpuDeviceId() {
+        return parentGpuDeviceId;
+    }
+
+    public void setParentGpuDeviceId(String parentGpuDeviceId) {
+        this.parentGpuDeviceId = parentGpuDeviceId;
+    }
 }

engine/schema/src/main/java/com/cloud/gpu/dao/VGPUTypesDaoImpl.java

@@ -42,7 +42,7 @@ public class VGPUTypesDaoImpl extends GenericDaoBase<VGPUTypesVO, Long> implemen
     private static final String LIST_ZONE_POD_CLUSTER_WIDE_GPU_CAPACITIES =
             "SELECT host_gpu_groups.group_name, vgpu_type, max_vgpu_per_pgpu, SUM(remaining_capacity) AS remaining_capacity, SUM(max_capacity) AS total_capacity FROM" +
                     " `cloud`.`vgpu_types` INNER JOIN `cloud`.`host_gpu_groups` ON vgpu_types.gpu_group_id = host_gpu_groups.id INNER JOIN `cloud`.`host`" +
-                    " ON host_gpu_groups.host_id = host.id WHERE host.type =  'Routing' AND host.data_center_id = ?";
+                    " ON host_gpu_groups.host_id = host.id WHERE host.type =  'Routing' AND vgpu_types.max_capacity > 0 AND host.data_center_id = ?";
     private final SearchBuilder<VGPUTypesVO> _searchByGroupId;
     private final SearchBuilder<VGPUTypesVO> _searchByGroupIdVGPUType;
     @Inject
@@ -80,7 +80,7 @@ public List<VgpuTypesInfo> listGPUCapacities(Long dcId, Long podId, Long cluster
             finalQuery.append(" AND host.cluster_id = ?");
             resourceIdList.add(clusterId);
         }
-        finalQuery.append(" GROUP BY host_gpu_groups.group_name, vgpu_type");
+        finalQuery.append(" GROUP BY host_gpu_groups.group_name, vgpu_type, max_vgpu_per_pgpu");
 
         try {
             pstmt = txn.prepareAutoCloseStatement(finalQuery.toString());

engine/schema/src/main/java/com/cloud/vm/dao/VMInstanceDaoImpl.java

@@ -131,12 +131,22 @@ public class VMInstanceDaoImpl extends GenericDaoBase<VMInstanceVO, Long> implem
 
     private static final String ORDER_HOSTS_NUMBER_OF_VMS_FOR_ACCOUNT_PART2 = " GROUP BY host.id ORDER BY 2 ASC ";
 
-    private static final String COUNT_VMS_BASED_ON_VGPU_TYPES1 =
+    private static final String COUNT_VMS_BASED_ON_VGPU_TYPES1_LEGACY =
             "SELECT pci, type, SUM(vmcount) FROM (SELECT MAX(IF(offering.name = 'pciDevice',value,'')) AS pci, MAX(IF(offering.name = 'vgpuType', value,'')) " +
             "AS type, COUNT(DISTINCT vm.id) AS vmcount FROM service_offering_details offering INNER JOIN vm_instance vm ON offering.service_offering_id = vm.service_offering_id " +
             "INNER JOIN `cloud`.`host` ON vm.host_id = host.id WHERE vm.state = 'Running' AND host.data_center_id = ? ";
+    private static final String COUNT_VMS_BASED_ON_VGPU_TYPES2_LEGACY =
+            "GROUP BY vm.service_offering_id) results GROUP BY pci, type";
+
+    private static final String COUNT_VMS_BASED_ON_VGPU_TYPES1 =
+            "SELECT gpu_card.device_name, vgpu_profile.name, COUNT(gpu_device.vm_id) "
+            + "FROM `cloud`.`gpu_device` "
+            + "INNER JOIN `cloud`.`host` ON gpu_device.host_id = host.id "
+            + "INNER JOIN `cloud`.`gpu_card` ON gpu_device.card_id = gpu_card.id "
+            + "INNER JOIN `cloud`.`vgpu_profile` ON vgpu_profile.id = gpu_device.vgpu_profile_id "
+            + "WHERE vm_id IS NOT NULL AND host.data_center_id = ? ";
     private static final String COUNT_VMS_BASED_ON_VGPU_TYPES2 =
-            "GROUP BY pci, type) results GROUP BY pci, type";
+            "GROUP BY gpu_card.name, vgpu_profile.name";
 
     private static final String UPDATE_SYSTEM_VM_TEMPLATE_ID_FOR_HYPERVISOR = "UPDATE `cloud`.`vm_instance` SET vm_template_id = ? WHERE type <> 'User' AND hypervisor_type = ? AND removed is NULL";
 
@@ -794,40 +804,52 @@ public List<Long> listHostIdsByVmCount(long dcId, Long podId, Long clusterId, lo
 
     @Override
     public HashMap<String, Long> countVgpuVMs(Long dcId, Long podId, Long clusterId) {
+        StringBuilder finalQueryLegacy = new StringBuilder();
         StringBuilder finalQuery = new StringBuilder();
         TransactionLegacy txn = TransactionLegacy.currentTxn();
+        PreparedStatement pstmtLegacy = null;
         PreparedStatement pstmt = null;
         List<Long> resourceIdList = new ArrayList<Long>();
         HashMap<String, Long> result = new HashMap<String, Long>();
 
         resourceIdList.add(dcId);
+        finalQueryLegacy.append(COUNT_VMS_BASED_ON_VGPU_TYPES1_LEGACY);
         finalQuery.append(COUNT_VMS_BASED_ON_VGPU_TYPES1);
 
         if (podId != null) {
+            finalQueryLegacy.append("AND host.pod_id = ? ");
             finalQuery.append("AND host.pod_id = ? ");
             resourceIdList.add(podId);
         }
 
         if (clusterId != null) {
+            finalQueryLegacy.append("AND host.cluster_id = ? ");
             finalQuery.append("AND host.cluster_id = ? ");
             resourceIdList.add(clusterId);
         }
+        finalQueryLegacy.append(COUNT_VMS_BASED_ON_VGPU_TYPES2_LEGACY);
         finalQuery.append(COUNT_VMS_BASED_ON_VGPU_TYPES2);
 
         try {
+            pstmtLegacy = txn.prepareAutoCloseStatement(finalQueryLegacy.toString());
             pstmt = txn.prepareAutoCloseStatement(finalQuery.toString());
             for (int i = 0; i < resourceIdList.size(); i++) {
+                pstmtLegacy.setLong(1 + i, resourceIdList.get(i));
                 pstmt.setLong(1 + i, resourceIdList.get(i));
             }
-            ResultSet rs = pstmt.executeQuery();
+            ResultSet rs = pstmtLegacy.executeQuery();
+            while (rs.next()) {
+                result.put(rs.getString(1).concat(rs.getString(2)), rs.getLong(3));
+            }
+            rs = pstmt.executeQuery();
             while (rs.next()) {
                 result.put(rs.getString(1).concat(rs.getString(2)), rs.getLong(3));
             }
             return result;
         } catch (SQLException e) {
-            throw new CloudRuntimeException("DB Exception on: " + finalQuery, e);
+            throw new CloudRuntimeException("DB Exception on: " + finalQueryLegacy, e);
         } catch (Throwable e) {
-            throw new CloudRuntimeException("Caught: " + finalQuery, e);
+            throw new CloudRuntimeException("Caught: " + finalQueryLegacy, e);
         }
     }
 

server/src/main/java/org/apache/cloudstack/gpu/GpuServiceImpl.java

@@ -555,10 +555,13 @@ public void doInTransactionWithoutResult(TransactionStatus status) {
 
                 // Add vGPU profile IDs if provided
                 if (vgpuProfileIds != null && !vgpuProfileIds.isEmpty()) {
+                    List<GpuOfferingDetailVO> detailList = new ArrayList<>();
+
                     for (VgpuProfile vgpuProfile : vgpuProfileList) {
-                        gpuOfferingDetailsDao.addDetail(newGpuOffering.getId(), GpuOfferingDetailVO.VgpuProfileId,
-                                String.valueOf(vgpuProfile.getId()), true);
+                        detailList.add(new GpuOfferingDetailVO(gpuOffering.getId(),
+                                GpuOfferingDetailVO.VgpuProfileId, String.valueOf(vgpuProfile.getId()), true));
                     }
+                    gpuOfferingDetailsDao.saveDetails(detailList);
                     newGpuOffering.setVgpuProfiles(vgpuProfileList);
                 }
             }
@@ -632,10 +635,13 @@ public void doInTransactionWithoutResult(TransactionStatus status) {
                     gpuOfferingDetailsDao.removeDetails(gpuOffering.getId());
 
                     // Then add the new ones if not empty
+                    List<GpuOfferingDetailVO> detailList = new ArrayList<>();
+
                     for (VgpuProfile vgpuProfile : vgpuProfileList) {
-                        gpuOfferingDetailsDao.addDetail(gpuOffering.getId(), GpuOfferingDetailVO.VgpuProfileId,
-                                String.valueOf(vgpuProfile.getId()), true);
+                        detailList.add(new GpuOfferingDetailVO(gpuOffering.getId(),
+                                GpuOfferingDetailVO.VgpuProfileId, String.valueOf(vgpuProfile.getId()), true));
                     }
+                    gpuOfferingDetailsDao.saveDetails(detailList);
 
                     // Refresh vGPU profiles list
                     gpuOffering.setVgpuProfiles(vgpuProfileList);
@@ -760,16 +766,14 @@ public HashMap<String, HashMap<String, VgpuTypesInfo>> getGpuGroupDetailsFromGpu
                 gpuDeviceInfo =
                         new VgpuTypesInfo(card.getDeviceName(), vgpuProfile.getName(), card.getVramSize(), null, null,
                                 null, maxVgpuPerPgpu, GpuDevice.State.Free.equals(device.getState()) ? 1L : 0L,
-                                GpuDevice.State.HasVGPUs.equals(device.getState()) ? 0L : 1L);
+                                1L);
                 gpuGroupDetails.get(card.getDeviceName()).put(vgpuProfile.getName(), gpuDeviceInfo);
             } else {
                 // Update the existing VgpuTypesInfo with the new device's information
                 if (GpuDevice.State.Free.equals(device.getState())) {
                     gpuDeviceInfo.setRemainingCapacity(gpuDeviceInfo.getRemainingCapacity() + 1);
                 }
-                if (!GpuDevice.State.HasVGPUs.equals(device.getState())) {
-                    gpuDeviceInfo.setMaxVmCapacity(gpuDeviceInfo.getMaxCapacity() + 1);
-                }
+                gpuDeviceInfo.setMaxVmCapacity(gpuDeviceInfo.getMaxCapacity() + 1);
             }
         }
         return gpuGroupDetails;
@@ -969,6 +973,15 @@ private GpuDeviceResponse createGpuDeviceResponse(GpuDeviceVO gpuDevice) {
             }
         }
 
+        if (gpuDevice.getParentGpuDeviceId() != null) {
+            GpuDeviceVO parentGpuDevice = gpuDeviceDao.findById(gpuDevice.getParentGpuDeviceId());
+            if (parentGpuDevice != null) {
+                response.setParentGpuDeviceId(parentGpuDevice.getUuid());
+            } else {
+                s_logger.debug("Parent GPU device with ID {} not found for GPU device {}", gpuDevice.getParentGpuDeviceId(), gpuDevice.getUuid());
+            }
+        }
+
         return response;
     }
 }

ui/public/locales/en.json

@@ -265,6 +265,8 @@
 "label.add.f5.device": "Add F5 device",
 "label.add.firewall": "Add firewall rule",
 "label.add.firewallrule": "Add Firewall Rule",
+"label.add.gpu.card": "Add GPU card",
+"label.add.gpu.offering": "Add GPU offering",
 "label.add.guest.network": "Add guest Network",
 "label.add.guest.os": "Add guest OS",
 "label.add.guest.os.category": "Add guest OS category",
@@ -336,6 +338,7 @@
 "label.add.user": "Add User",
 "label.add.upstream.ipv4.routes": "Add upstream IPv4 routes",
 "label.add.upstream.ipv6.routes": "Add upstream IPv6 routes",
+"label.add.vgpu.profile": "Add vGPU profile",
 "label.add.vm": "Add Instance",
 "label.add.vms": "Add Instances",
 "label.add.vmware.datacenter": "Add VMware datacenter",
@@ -835,6 +838,7 @@
 "label.disable.webhook": "Disable Webhook",
 "label.disabled": "Disabled",
 "label.disconnected": "Last disconnected",
+"label.discover.gpu.devices": "Discover GPU devices",
 "label.disk": "Disk",
 "label.disk.offerings": "Disk offerings",
 "label.disk.path": "Disk Path",
@@ -1065,9 +1069,16 @@
 "label.glustervolume": "Volume",
 "label.go.back": "Go back",
 "label.gpu": "GPU",
+"label.gpucardid": "GPU Card",
 "label.gpucardname": "GPU Card Name",
 "label.gpu.card": "GPU Card",
+"label.gpu.count": "GPU Count",
+"label.gpucount": "GPU Count",
+"label.gpu.device": "GPU Device",
 "label.gpu.devices": "GPU Devices",
+"label.gpuofferingname": "GPU offering",
+"label.gpu.offering": "GPU Offering",
+"label.gpu.offerings": "GPU Offerings",
 "label.chart.info": "Information about the charts",
 "label.group": "Group",
 "label.group.optional": "Group (Optional)",
@@ -2534,9 +2545,10 @@
 "label.verify": "Verify",
 "label.version": "Version",
 "label.versions": "Versions",
-"label.vgpu": "VGPU",
-"label.vgpuprofilename": "VGPU Profile Name",
-"label.vgpu.profile": "VGPU Profile",
+"label.vgpu": "vGPU",
+"label.vgpuprofileids": "vGPU Profile",
+"label.vgpuprofilename": "vGPU Profile Name",
+"label.vgpu.profile": "vGPU Profile",
 "label.vgputype": "vGPU type",
 "label.view": "View",
 "label.view.all": "View all",
@@ -2781,6 +2793,7 @@
 "message.action.disable.pod": "Please confirm that you want to disable this pod.",
 "message.action.disable.static.nat": "Please confirm that you want to disable static NAT.",
 "message.action.disable.zone": "Please confirm that you want to disable this zone.",
+"message.action.discover.gpu.devices": "Please confirm that you want to discover GPU devices. Only devices on the host which match GPU Card & vGPU Profile will be discovered and will be available for use.",
 "message.action.download.iso": "Please confirm that you want to download this ISO.",
 "message.action.download.snapshot": "Please confirm that you want to download this Snapshot.",
 "message.action.download.template": "Please confirm that you want to download this Template.",

ui/src/components/view/GPUTab.vue

@@ -17,30 +17,54 @@
 
 <template>
   <div>
-    <list-view
-      :tabLoading="tabLoading"
+    <a-table
+      :loading="tabLoading"
       :columns="columns"
-      :items="items"
-      :actions="actions"
-      :columnKeys="columnKeys"
-      :selectedColumns="selectedColumnKeys"
-      ref="listview"
-      @update-selected-columns="updateSelectedColumns"
-      @refresh="this.fetchData"
-      @enable-gpu-device="enableGpuDevice"
-      @disable-gpu-device="disableGpuDevice" />
-    </div>
+      :dataSource="items"
+      :pagination="false"
+      :rowKey="record => record.id"
+      :childrenColumnName="'children'"
+      :defaultExpandAllRows="true"
+      size="small"
+    >
+      <template #bodyCell="{ column, record }">
+        <template v-if="column.key === 'gpuDeviceActions'">
+          <a-space>
+            <a-button
+              v-if="record.state === 'Disabled'"
+              type="primary"
+              size="small"
+              @click="enableGpuDevice(record)"
+            >
+              {{ $t('label.enable') }}
+            </a-button>
+            <a-button
+              v-if="record.state === 'Free'"
+              type="primary"
+              danger
+              size="small"
+              @click="disableGpuDevice(record)"
+            >
+              {{ $t('label.disable') }}
+            </a-button>
+          </a-space>
+        </template>
+        <template v-else-if="column.key === 'busaddress'">
+          <span :style="{ paddingLeft: record.parentgpudeviceid ? '20px' : '0px' }">
+            {{ record.busaddress }}
+          </span>
+        </template>
+      </template>
+    </a-table>
+  </div>
 </template>
 
 <script>
 import { api } from '@/api'
 import { genericCompare } from '@/utils/sort.js'
-import ListView from '@/components/view/ListView'
+
 export default {
   name: 'GPUTab',
-  components: {
-    ListView
-  },
   props: {
     resource: {
       type: Object,
@@ -88,14 +112,47 @@ export default {
       }
       this.tabLoading = true
       api('listGpuDevices', params).then(json => {
-        this.items = []
-        this.items = json?.listgpudevicesresponse?.gpudevice || []
+        const devices = json?.listgpudevicesresponse?.gpudevice || []
+        this.items = this.buildGpuTree(devices)
       }).catch(error => {
         this.$notifyError(error)
       }).finally(() => {
         this.tabLoading = false
       })
     },
+    buildGpuTree (devices) {
+      // Separate parent devices and vGPUs
+      const parentDevices = []
+      const vgpuDevices = []
+      for (const device of devices) {
+        if (device.parentgpudeviceid) {
+          vgpuDevices.push(device)
+        } else {
+          parentDevices.push(device)
+        }
+      }
+
+      // Group vGPUs by their parent ID
+      const vgpusByParent = {}
+      vgpuDevices.forEach(vgpu => {
+        const parentId = vgpu.parentgpudeviceid
+        if (!vgpusByParent[parentId]) {
+          vgpusByParent[parentId] = []
+        }
+        vgpusByParent[parentId].push(vgpu)
+      })
+
+      // Build tree structure
+      const treeData = parentDevices.map(parent => {
+        const children = vgpusByParent[parent.id] || []
+        return {
+          ...parent,
+          children: children.length > 0 ? children : undefined
+        }
+      })
+
+      return treeData
+    },
     updateSelectedColumns (key) {
       if (this.selectedColumnKeys.includes(key)) {
         this.selectedColumnKeys = this.selectedColumnKeys.filter(x => x !== key)