Modify Accelerator support for kserve

backend/src/types.ts

@@ -909,7 +909,7 @@ export type ServingRuntime = K8sResourceCommon & {
       args: string[];
       image: string;
       name: string;
-      resources: ContainerResources;
+      resources?: ContainerResources;
       volumeMounts?: VolumeMount[];
     }[];
     supportedModelFormats: SupportedModelFormats[];

frontend/src/__mocks__/mockAcceleratork8sResource.ts

@@ -0,0 +1,47 @@
+import { AcceleratorKind } from '~/k8sTypes';
+import { genUID } from './mockUtils';
+
+type MockResourceConfigType = {
+  name?: string;
+  namespace?: string;
+  displayName?: string;
+  identifier?: string;
+  enabled?: boolean;
+  tolerations?: {
+    key: string;
+    operator: string;
+    effect: string;
+  }[];
+};
+
+export const mockAcceleratork8sResource = ({
+  name = 'migrated-gpu',
+  namespace = 'test-project',
+  displayName = 'Nvidia GPU',
+  identifier = 'nvidia.com/gpu',
+  enabled = true,
+  tolerations = [
+    {
+      key: 'nvidia.com/gpu',
+      operator: 'Exists',
+      effect: 'NoSchedule',
+    },
+  ],
+}: MockResourceConfigType): AcceleratorKind => ({
+  apiVersion: 'dashboard.opendatahub.io/v1',
+  kind: 'AcceleratorProfile',
+  metadata: {
+    creationTimestamp: '2023-03-17T16:12:41Z',
+    generation: 1,
+    name,
+    namespace,
+    resourceVersion: '1309350',
+    uid: genUID('service'),
+  },
+  spec: {
+    identifier,
+    displayName,
+    enabled,
+    tolerations,
+  },
+});

frontend/src/__mocks__/mockInferenceServiceModalData.ts

@@ -0,0 +1,28 @@
+import {
+  CreatingInferenceServiceObject,
+  InferenceServiceStorageType,
+} from '~/pages/modelServing/screens/types';
+
+type MockResourceConfigType = Partial<CreatingInferenceServiceObject>;
+
+export const mockInferenceServiceModalData = ({
+  name = 'my-inference-service',
+  project = 'caikit-example',
+  servingRuntimeName = 'caikit',
+  storage = {
+    type: InferenceServiceStorageType.NEW_STORAGE,
+    path: '/caikit-llama',
+    dataConnection: 'aws-data-connection',
+    awsData: [],
+  },
+  format = {
+    name: 'caikit',
+    version: '1.0.0',
+  },
+}: MockResourceConfigType): CreatingInferenceServiceObject => ({
+  name,
+  project,
+  servingRuntimeName,
+  storage,
+  format,
+});

frontend/src/__mocks__/mockServingRuntimeModalData.ts

@@ -0,0 +1,33 @@
+import { CreatingServingRuntimeObject } from '~/pages/modelServing/screens/types';
+
+type MockResourceConfigType = Partial<CreatingServingRuntimeObject>;
+
+export const mockServingRuntimeModalData = ({
+  name = 'my-inference-service',
+  servingRuntimeTemplateName = 'caikit',
+  numReplicas = 1,
+  modelSize = {
+    name: 'small',
+    resources: {
+      requests: {
+        cpu: '1',
+        memory: '1Gi',
+      },
+      limits: {
+        cpu: '1',
+        memory: '1Gi',
+      },
+    },
+  },
+  externalRoute = false,
+  tokenAuth = false,
+  tokens = [],
+}: MockResourceConfigType): CreatingServingRuntimeObject => ({
+  name,
+  servingRuntimeTemplateName,
+  numReplicas,
+  modelSize,
+  externalRoute,
+  tokenAuth,
+  tokens,
+});

frontend/src/api/__tests__/inferenceServices.spec.ts

@@ -1,25 +1,14 @@
+import { mockAcceleratork8sResource } from '~/__mocks__/mockAcceleratork8sResource';
+import { mockInferenceServiceModalData } from '~/__mocks__/mockInferenceServiceModalData';
 import { assembleInferenceService } from '~/api/k8s/inferenceServices';
-import { InferenceServiceStorageType } from '~/pages/modelServing/screens/types';
+import { translateDisplayNameForK8s } from '~/pages/projects/utils';
+import { AcceleratorState } from '~/utilities/useAcceleratorState';
 
 global.structuredClone = (val: unknown) => JSON.parse(JSON.stringify(val));
 
 describe('assembleInferenceService', () => {
   it('should have the right annotations when creating for Kserve', async () => {
-    const inferenceService = assembleInferenceService({
-      name: 'my-inference-service',
-      project: 'caikit-example',
-      servingRuntimeName: 'caikit',
-      storage: {
-        type: InferenceServiceStorageType.NEW_STORAGE,
-        path: '/caikit-llama',
-        dataConnection: 'aws-data-connection',
-        awsData: [],
-      },
-      format: {
-        name: 'caikit',
-        version: '1.0.0',
-      },
-    });
+    const inferenceService = assembleInferenceService(mockInferenceServiceModalData({}));
 
     expect(inferenceService.metadata.annotations).toBeDefined();
     expect(inferenceService.metadata.annotations?.['serving.kserve.io/deploymentMode']).toBe(
@@ -36,21 +25,7 @@ describe('assembleInferenceService', () => {
 
   it('should have the right annotations when creating for modelmesh', async () => {
     const inferenceService = assembleInferenceService(
-      {
-        name: 'my-inference-service',
-        project: 'caikit-example',
-        servingRuntimeName: 'caikit',
-        storage: {
-          type: InferenceServiceStorageType.NEW_STORAGE,
-          path: '/caikit-llama',
-          dataConnection: 'aws-data-connection',
-          awsData: [],
-        },
-        format: {
-          name: 'caikit',
-          version: '1.0.0',
-        },
-      },
+      mockInferenceServiceModalData({}),
       undefined,
       undefined,
       true,
@@ -68,4 +43,66 @@ describe('assembleInferenceService', () => {
       undefined,
     );
   });
+
+  it('should handle name and display name', async () => {
+    const displayName = 'Llama model';
+
+    const inferenceService = assembleInferenceService(
+      mockInferenceServiceModalData({ name: displayName }),
+    );
+
+    expect(inferenceService.metadata.annotations).toBeDefined();
+    expect(inferenceService.metadata.annotations?.['openshift.io/display-name']).toBe(displayName);
+    expect(inferenceService.metadata.name).toBe(translateDisplayNameForK8s(displayName));
+  });
+
+  it('should add accelerator if kserve and accelerator found', async () => {
+    const acceleratorState: AcceleratorState = {
+      accelerator: mockAcceleratork8sResource({}),
+      accelerators: [mockAcceleratork8sResource({})],
+      initialAccelerator: mockAcceleratork8sResource({}),
+      count: 1,
+      additionalOptions: {},
+      useExisting: false,
+    };
+
+    const inferenceService = assembleInferenceService(
+      mockInferenceServiceModalData({}),
+      undefined,
+      undefined,
+      false,
+      undefined,
+      acceleratorState,
+    );
+
+    expect(inferenceService.spec.predictor.tolerations).toBeDefined();
+    expect(inferenceService.spec.predictor.tolerations?.[0].key).toBe(
+      mockAcceleratork8sResource({}).spec.tolerations?.[0].key,
+    );
+    expect(inferenceService.spec.predictor.model.resources?.limits?.['nvidia.com/gpu']).toBe(1);
+    expect(inferenceService.spec.predictor.model.resources?.requests?.['nvidia.com/gpu']).toBe(1);
+  });
+
+  it('should not add accelerator if modelmesh and accelerator found', async () => {
+    const acceleratorState: AcceleratorState = {
+      accelerator: mockAcceleratork8sResource({}),
+      accelerators: [mockAcceleratork8sResource({})],
+      initialAccelerator: mockAcceleratork8sResource({}),
+      count: 1,
+      additionalOptions: {},
+      useExisting: false,
+    };
+
+    const inferenceService = assembleInferenceService(
+      mockInferenceServiceModalData({}),
+      undefined,
+      undefined,
+      true,
+      undefined,
+      acceleratorState,
+    );
+
+    expect(inferenceService.spec.predictor.tolerations).toBeUndefined();
+    expect(inferenceService.spec.predictor.model.resources).toBeUndefined();
+  });
 });

frontend/src/api/__tests__/servingRuntimes.spec.ts

@@ -1,23 +1,20 @@
+import { mockAcceleratork8sResource } from '~/__mocks__/mockAcceleratork8sResource';
 import { mockServingRuntimeK8sResource } from '~/__mocks__/mockServingRuntimeK8sResource';
+import { mockServingRuntimeModalData } from '~/__mocks__/mockServingRuntimeModalData';
 import { mockServingRuntimeTemplateK8sResource } from '~/__mocks__/mockServingRuntimeTemplateK8sResource';
 import { assembleServingRuntime } from '~/api/k8s/servingRuntimes';
 import { ServingRuntimeKind } from '~/k8sTypes';
+import { AcceleratorState } from '~/utilities/useAcceleratorState';
 
 global.structuredClone = (val: unknown) => JSON.parse(JSON.stringify(val));
 
 describe('assembleServingRuntime', () => {
   it('should omit enable-xxxx annotations when creating', async () => {
     const servingRuntime = assembleServingRuntime(
-      {
-        name: 'my-serving-runtime',
-        servingRuntimeTemplateName: 'ovms',
-        numReplicas: 2,
-        modelSize: { name: 'Small', resources: {} },
-        tokens: [],
-        // test false values
+      mockServingRuntimeModalData({
         externalRoute: false,
         tokenAuth: false,
-      },
+      }),
       'test',
       mockServingRuntimeTemplateK8sResource({}).objects[0] as ServingRuntimeKind,
       false,
@@ -31,16 +28,10 @@ describe('assembleServingRuntime', () => {
 
   it('should remove enable-xxxx annotations when editing', async () => {
     const servingRuntime = assembleServingRuntime(
-      {
-        name: 'my-serving-runtime',
-        servingRuntimeTemplateName: 'ovms',
-        numReplicas: 2,
-        modelSize: { name: 'Small', resources: {} },
-        tokens: [],
-        // test false values
+      mockServingRuntimeModalData({
         externalRoute: false,
         tokenAuth: false,
-      },
+      }),
       'test',
       mockServingRuntimeK8sResource({ auth: true, route: true }),
       false,
@@ -54,16 +45,10 @@ describe('assembleServingRuntime', () => {
 
   it('should add enable-xxxx annotations when creating', async () => {
     const servingRuntime = assembleServingRuntime(
-      {
-        name: 'my-serving-runtime',
-        servingRuntimeTemplateName: 'ovms',
-        numReplicas: 2,
-        modelSize: { name: 'Small', resources: {} },
-        tokens: [],
-        // test true values
+      mockServingRuntimeModalData({
         externalRoute: true,
         tokenAuth: true,
-      },
+      }),
       'test',
       mockServingRuntimeTemplateK8sResource({}).objects[0] as ServingRuntimeKind,
       false,
@@ -77,16 +62,10 @@ describe('assembleServingRuntime', () => {
 
   it('should add enable-xxxx annotations when editing', async () => {
     const servingRuntime = assembleServingRuntime(
-      {
-        name: 'my-serving-runtime',
-        servingRuntimeTemplateName: 'ovms',
-        numReplicas: 2,
-        modelSize: { name: 'Small', resources: {} },
-        tokens: [],
-        // test true values
+      mockServingRuntimeModalData({
         externalRoute: true,
         tokenAuth: true,
-      },
+      }),
       'test',
       mockServingRuntimeK8sResource({ auth: false, route: false }),
       false,
@@ -97,4 +76,62 @@ describe('assembleServingRuntime', () => {
     expect(servingRuntime.metadata.annotations?.['enable-auth']).toBe('true');
     expect(servingRuntime.metadata.annotations?.['enable-route']).toBe('true');
   });
+
+  it('should add tolerations and gpu on modelmesh', async () => {
+    const acceleratorState: AcceleratorState = {
+      accelerator: mockAcceleratork8sResource({}),
+      accelerators: [mockAcceleratork8sResource({})],
+      initialAccelerator: mockAcceleratork8sResource({}),
+      count: 1,
+      additionalOptions: {},
+      useExisting: false,
+    };
+
+    const servingRuntime = assembleServingRuntime(
+      mockServingRuntimeModalData({
+        externalRoute: true,
+        tokenAuth: true,
+      }),
+      'test',
+      mockServingRuntimeK8sResource({ auth: false, route: false }),
+      true,
+      false,
+      acceleratorState,
+      true,
+    );
+
+    expect(servingRuntime.spec.tolerations).toBeDefined();
+    expect(servingRuntime.spec.containers[0].resources?.limits?.['nvidia.com/gpu']).toBe(1);
+    expect(servingRuntime.spec.containers[0].resources?.requests?.['nvidia.com/gpu']).toBe(1);
+  });
+
+  it('should not add tolerations and gpu on kserve', async () => {
+    const acceleratorState: AcceleratorState = {
+      accelerator: mockAcceleratork8sResource({}),
+      accelerators: [mockAcceleratork8sResource({})],
+      initialAccelerator: mockAcceleratork8sResource({}),
+      count: 1,
+      additionalOptions: {},
+      useExisting: false,
+    };
+
+    const servingRuntime = assembleServingRuntime(
+      mockServingRuntimeModalData({
+        externalRoute: true,
+        tokenAuth: true,
+      }),
+      'test',
+      mockServingRuntimeK8sResource({ auth: false, route: false }),
+      true,
+      false,
+      acceleratorState,
+      false,
+    );
+
+    expect(servingRuntime.spec.tolerations).toBeUndefined();
+    expect(servingRuntime.spec.containers[0].resources?.limits?.['nvidia.com/gpu']).toBeUndefined();
+    expect(
+      servingRuntime.spec.containers[0].resources?.requests?.['nvidia.com/gpu'],
+    ).toBeUndefined();
+  });
 });