support gpu to the grid client

packages/grid_client/scripts/vm_with_gpu.ts

@@ -0,0 +1,79 @@
+import { DiskModel, FilterOptions, MachineModel, MachinesModel, NetworkModel } from "../src";
+import { config, getClient } from "./client_loader";
+import { log } from "./utils";
+
+async function main() {
+  const grid3 = await getClient();
+
+  // create network Object
+  const n = new NetworkModel();
+  n.name = "vmgpuNetwork";
+  n.ip_range = "10.249.0.0/16";
+
+  // create disk Object
+  const disk = new DiskModel();
+  disk.name = "vmgpuDisk";
+  disk.size = 100;
+  disk.mountpoint = "/testdisk";
+
+  const vmQueryOptions: FilterOptions = {
+    cru: 8,
+    mru: 16, // GB
+    sru: 1000,
+    availableFor: grid3.twinId,
+    hasGPU: true,
+    rentedBy: grid3.twinId,
+  };
+
+  const nodes = await grid3.capacity.filterNodes(vmQueryOptions);
+  if (nodes.length === 0) {
+    throw Error(`Couldn't find a node satisfying these filter options: ${JSON.stringify(vmQueryOptions)}`);
+  }
+  const nodeId = +nodes[0].nodeId;
+
+  // create vm node Object
+  const vm = new MachineModel();
+  vm.name = "vmgpu";
+  vm.node_id = nodeId;
+  vm.disks = [disk];
+  vm.public_ip = false;
+  vm.planetary = true;
+  vm.cpu = 8;
+  vm.memory = 1024 * 16;
+  vm.rootfs_size = 0;
+  vm.flist = "https://hub.grid.tf/tf-official-vms/ubuntu-22.04.flist";
+  vm.entrypoint = "/";
+  vm.env = {
+    SSH_KEY: config.ssh_key,
+  };
+  let gpuList = await grid3.zos.getNodeGPUInfo({ nodeId: nodeId });
+  gpuList = gpuList.filter(g => g.contract === 0);
+  if (gpuList.length <= 0) {
+    throw Error(`Couldn't find GPU card available on node ${nodeId}`);
+  }
+  vm.gpu = [gpuList[0].id]; // gpu card's id, you can check the available gpu from the dashboard
+
+  // create VMs Object
+  const vms = new MachinesModel();
+  vms.name = "vmgpu";
+  vms.network = n;
+  vms.machines = [vm];
+  vms.metadata = "";
+  vms.description = "test deploying VM with GPU via ts grid3 client";
+
+  // deploy vms
+  const res = await grid3.machines.deploy(vms);
+  log(res);
+
+  // get the deployment
+  const l = await grid3.machines.getObj(vms.name);
+  log(l);
+
+  // // delete
+  // const d = await grid3.machines.delete({ name: vms.name });
+  // log(d);
+
+  await grid3.disconnect();
+}
+
+main();

packages/grid_client/scripts/zos_rmb_requests.ts

@@ -42,6 +42,12 @@ async function main() {
   } catch (e) {
     log(`Couldn't reach node ${nodeId} to get storage pools due to ${e}`);
   }
+  try {
+    log(await grid3.zos.getNodeGPUInfo({ nodeId }));
+  } catch (e) {
+    log(`Couldn't reach node ${nodeId} to get gpu info due to ${e}`);
+  }
+
   await grid3.disconnect();
 }
 

packages/grid_client/src/high_level/base.ts

@@ -1,5 +1,6 @@
 import { Addr } from "netaddr";
 
+import { RMB } from "../clients/rmb/client";
 import { GridClientConfig } from "../config";
 import { events } from "../helpers/events";
 import { Operations, TwinDeployment } from "../high_level/models";
@@ -11,8 +12,11 @@ import { Workload, WorkloadTypes } from "../zos/workload";
 
 class HighLevelBase {
   nodes: Nodes;
+  rmb: RMB;
+
   constructor(public config: GridClientConfig) {
     this.nodes = new Nodes(this.config.graphqlURL, this.config.proxyURL, this.config.rmbClient);
+    this.rmb = new RMB(config.rmbClient);
   }
 
   _filterWorkloads(

packages/grid_client/src/high_level/kubernetes.ts

@@ -32,6 +32,7 @@ class KubernetesHL extends HighLevelBase {
     corex = false,
     solutionProviderId: number,
     zlogsOutput?: string,
+    gpu: string[] = [],
   ) {
     events.emit("logs", `Creating a master with name: ${name} on node: ${nodeId}, network: ${network.name}`);
     const machine = new VMHL(this.config);
@@ -73,6 +74,7 @@ class KubernetesHL extends HighLevelBase {
       corex,
       solutionProviderId,
       zlogsOutput,
+      gpu,
     );
   }
 
@@ -100,6 +102,7 @@ class KubernetesHL extends HighLevelBase {
     corex = false,
     solutionProviderId: number,
     zlogsOutput?: string,
+    gpu: string[] = [],
   ) {
     events.emit("logs", `Creating a worker with name: ${name} on node: ${nodeId}, network: ${network.name}`);
     const machine = new VMHL(this.config);
@@ -141,6 +144,7 @@ class KubernetesHL extends HighLevelBase {
       corex,
       solutionProviderId,
       zlogsOutput,
+      gpu,
     );
   }
 

packages/grid_client/src/high_level/machine.ts

@@ -45,6 +45,7 @@ class VMHL extends HighLevelBase {
     corex = false,
     solutionProviderId: number,
     zlogsOutput?: string,
+    gpu: string[] = [],
   ): Promise<[TwinDeployment[], string]> {
     const deployments: TwinDeployment[] = [];
     const workloads: Workload[] = [];
@@ -120,6 +121,7 @@ class VMHL extends HighLevelBase {
     }
 
     // ipv4
+    // TODO: make sure that the farm has a free public ip before continuing the deployment
     let ipName = "";
     let publicIps = 0;
     if (publicIp || publicIp6) {
@@ -131,6 +133,28 @@ class VMHL extends HighLevelBase {
       }
     }
 
+    if (gpu && gpu.length > 0) {
+      const nodeTwinId = await this.nodes.getNodeTwinId(nodeId);
+      const gpuList = await this.rmb.request([nodeTwinId], "zos.gpu.list", "");
+      if (gpuList.length <= 0) {
+        throw Error(`The selected node ${nodeId} doesn't have GPU card`);
+      }
+      for (const g of gpu) {
+        const found = gpuList.filter(item => item.id === g);
+        if (found.length === 0) {
+          throw Error(`Couldn't find the GPU with id: "${g}" in node: ${nodeId}`);
+        }
+        if (found[0].contract !== 0) {
+          throw Error(`This GPU: "${g}" is currently in use by another VM with contract id: ${found[0].contract}`);
+        }
+      }
+
+      const node = await this.nodes.getNode(nodeId);
+      if (node.rentedByTwinId !== this.config.twinId) {
+        throw Error(`This node ${nodeId} is not rented by the current user`);
+      }
+    }
+
     // validate user ip subnet in case of no networks already
     let userIPsubnet;
     let accessNodeSubnet;
@@ -247,6 +271,7 @@ class VMHL extends HighLevelBase {
         description,
         0,
         corex,
+        gpu,
       ),
     );
 

packages/grid_client/src/modules/base.ts

@@ -207,6 +207,7 @@ class BaseModule {
       description: workload.description,
       rootfs_size: data.size,
       corex: data.corex,
+      gpu: data.gpu,
     };
   }
 

packages/grid_client/src/modules/k8s.ts

@@ -115,6 +115,7 @@ class K8sModule extends BaseModule {
         master.corex,
         master.solutionProviderId!,
         master.zlogsOutput,
+        master.gpu,
       );
 
       deployments = deployments.concat(twinDeployments);
@@ -158,6 +159,7 @@ class K8sModule extends BaseModule {
         worker.corex,
         worker.solutionProviderId!,
         worker.zlogsOutput,
+        worker.gpu,
       );
 
       deployments = deployments.concat(twinDeployments);
@@ -295,6 +297,7 @@ class K8sModule extends BaseModule {
       options.corex,
       options.solutionProviderId!,
       options.zlogsOutput,
+      options.gpu,
     );
 
     return await this._add(options.deployment_name, options.node_id, oldDeployments, twinDeployments, network);

packages/grid_client/src/modules/machines.ts

@@ -71,6 +71,7 @@ class MachinesModule extends BaseModule {
         machine.corex,
         machine.solutionProviderId!,
         machine.zlogsOutput,
+        machine.gpu,
       );
       twinDeployments = twinDeployments.concat(TDeployments);
       if (wgConfig) {
@@ -188,6 +189,7 @@ class MachinesModule extends BaseModule {
       options.corex,
       options.solutionProviderId!,
       options.zlogsOutput,
+      options.gpu,
     );
     return await this._add(options.deployment_name, options.node_id, oldDeployments, twinDeployments, network);
   }

packages/grid_client/src/modules/models.ts

@@ -100,6 +100,7 @@ class MachineModel {
   @Expose() @IsOptional() @IsBoolean() corex?: boolean;
   @Expose() @IsInt() @IsOptional() solutionProviderId?: number;
   @Expose() @IsString() @IsOptional() zlogsOutput?: string;
+  @Expose() @IsString({ each: true }) @IsOptional() gpu?: string[];
 }
 
 class MachinesModel {
@@ -138,6 +139,7 @@ class KubernetesNodeModel {
   @Expose() @IsOptional() @IsBoolean() corex?: boolean;
   @Expose() @IsInt() @IsOptional() solutionProviderId?: number;
   @Expose() @IsString() @IsOptional() zlogsOutput?: string;
+  @Expose() @IsString({ each: true }) @IsOptional() gpu?: string[];
 }
 
 class K8SModel {
@@ -556,9 +558,11 @@ class FilterOptions {
   @Expose() @IsOptional() @IsString() country?: string;
   @Expose() @IsOptional() @IsString() city?: string;
   @Expose() @IsOptional() @IsBoolean() dedicated?: boolean;
-  @Expose() @IsOptional() @IsInt() availableFor?: number;
+  @Expose() @IsOptional() @IsInt() @Min(1) availableFor?: number;
   @Expose() @IsOptional() @IsInt() page?: number;
   @Expose() @IsOptional() @IsInt() size?: number;
+  @Expose() @IsOptional() @IsBoolean() hasGPU?: boolean;
+  @Expose() @IsOptional() @IsInt() @Min(1) rentedBy?: number;
 }
 
 class CalculatorModel {

packages/grid_client/src/modules/zos.ts

@@ -98,6 +98,13 @@ class Zos {
     const nodeTwinId = await this.capacity.getNodeTwinId(options.nodeId);
     return await this.rmb.request([nodeTwinId], "zos.storage.pools", "");
   }
+
+  @expose
+  @validateInput
+  async getNodeGPUInfo(options: ZOSNodeModel) {
+    const nodeTwinId = await this.capacity.getNodeTwinId(options.nodeId);
+    return await this.rmb.request([nodeTwinId], "zos.gpu.list", "");
+  }
 }
 
 export { Zos as zos };

packages/grid_client/src/primitives/nodes.ts

@@ -45,6 +45,9 @@ interface NodeInfo {
   publicConfig: PublicConfig;
   status: string;
   certificationType: string;
+  hasGPU: boolean;
+  extraFee: number;
+  rentedByTwinId: number;
 }
 interface PublicConfig {
   domain: string;
@@ -332,6 +335,8 @@ class Nodes {
       status: "up",
       page: options.page,
       size: options.size,
+      has_gpu: options.hasGPU,
+      rented_by: options.rentedBy,
     };
     if (options.gateway) {
       params["ipv4"] = true;

packages/grid_client/src/primitives/vm.ts

@@ -39,6 +39,7 @@ class VMPrimitive {
     description = "",
     version = 0,
     corex = false,
+    gpu: string[] = [],
   ): Workload {
     const zmachine = new Zmachine();
     zmachine.flist = flist;
@@ -49,6 +50,7 @@ class VMPrimitive {
     zmachine.compute_capacity = this._createComputeCapacity(cpu, memory);
     zmachine.env = env;
     zmachine.corex = corex;
+    zmachine.gpu = gpu;
 
     const zmachine_workload = new Workload();
     zmachine_workload.version = version || 0;

packages/grid_client/src/zos/zmachine.ts

@@ -1,5 +1,16 @@
 import { Expose, Transform, Type } from "class-transformer";
-import { IsBoolean, IsDefined, IsInt, IsIP, IsNotEmpty, IsString, Max, Min, ValidateNested } from "class-validator";
+import {
+  IsBoolean,
+  IsDefined,
+  IsInt,
+  IsIP,
+  IsNotEmpty,
+  IsOptional,
+  IsString,
+  Max,
+  Min,
+  ValidateNested,
+} from "class-validator";
 
 import { ComputeCapacity } from "./computecapacity";
 import { WorkloadData, WorkloadDataResult } from "./workload_base";
@@ -47,6 +58,7 @@ class Zmachine extends WorkloadData {
   @Expose() @IsString() @IsNotEmpty() entrypoint: string;
   @Expose() env: Record<string, unknown>;
   @Expose() @Transform(({ value }) => (value ? true : false)) @IsBoolean() corex: boolean;
+  @Expose() @IsString({ each: true }) @IsOptional() gpu?: string[];
 
   challenge(): string {
     let out = "";
@@ -62,6 +74,11 @@ class Zmachine extends WorkloadData {
       out += key;
       out += "=";
       out += this.env[key];
+      if (this.gpu) {
+        for (const g of this.gpu) {
+          out += g;
+        }
+      }
     }
     return out;
   }