bug fixes

backend/src/types.ts

@@ -247,6 +247,7 @@ export type KubeDecorator = KubeStatus & {
   customObjectsApi: k8s.CustomObjectsApi;
   rbac: k8s.RbacAuthorizationV1Api;
   currentToken: string;
+
 };
 
 export type KubeFastifyInstance = FastifyInstance & {
@@ -806,12 +807,17 @@ export type NotebookData = {
   notebookSizeName: string;
   imageName: string;
   imageTagName: string;
-  gpus: number;
+  accelerator: AcceleratorState;
   envVars: EnvVarReducedTypeKeyValues;
   state: NotebookState;
   username?: string;
 };
 
+export type AcceleratorState = {
+  accelerator?: AcceleratorKind;
+  count: number;
+};
+
 export const LIMIT_NOTEBOOK_IMAGE_GPU = 'nvidia.com/gpu';
 
 type DisplayNameAnnotations = Partial<{
@@ -863,18 +869,20 @@ export type SupportedModelFormats = {
   autoSelect?: boolean;
 };
 
-export type GPUCount = string | number;
+
+export enum ContainerResourceAttributes {
+  CPU = 'cpu',
+  MEMORY = 'memory',
+}
 
 export type ContainerResources = {
   requests?: {
     cpu?: string;
     memory?: string;
-    'nvidia.com/gpu'?: GPUCount;
   };
   limits?: {
     cpu?: string;
     memory?: string;
-    'nvidia.com/gpu'?: GPUCount;
   };
 };
 
@@ -917,4 +925,11 @@ export type AcceleratorKind = K8sResourceCommon & {
     description?: string;
     tolerations?: NotebookToleration[];
   };
-};
+};
+
+export enum KnownLabels {
+  DASHBOARD_RESOURCE = 'opendatahub.io/dashboard',
+  PROJECT_SHARING = 'opendatahub.io/project-sharing',
+  MODEL_SERVING_PROJECT = 'modelmesh-enabled',
+  DATA_CONNECTION_AWS = 'opendatahub.io/managed',
+}

backend/src/utils/constants.ts

@@ -1,6 +1,6 @@
 import * as path from 'path';
 import './dotenv';
-import { DashboardConfig, NotebookSize } from '../types';
+import { DashboardConfig, KnownLabels, NotebookSize } from '../types';
 
 export const PORT = Number(process.env.PORT) || Number(process.env.BACKEND_PORT) || 8080;
 export const IP = process.env.IP || '0.0.0.0';
@@ -133,3 +133,5 @@ export const DEFAULT_NOTEBOOK_SIZES: NotebookSize[] = [
 
 export const imageUrlRegex =
   /^([\w.\-_]+((?::\d+|)(?=\/[a-z0-9._-]+\/[a-z0-9._-]+))|)(?:\/|)([a-z0-9.\-_]+(?:\/[a-z0-9.\-_]+|))(?::([\w.\-_]{1,127})|)/;
+
+  export const LABEL_SELECTOR_DASHBOARD_RESOURCE = `${KnownLabels.DASHBOARD_RESOURCE}=true`;

backend/src/utils/notebookUtils.ts

@@ -1,10 +1,10 @@
 import { getDashboardConfig } from './resourceUtils';
 import {
+  ContainerResourceAttributes,
   EnvironmentVariable,
   ImageInfo,
   ImageTag,
   KubeFastifyInstance,
-  LIMIT_NOTEBOOK_IMAGE_GPU,
   Notebook,
   NotebookAffinity,
   NotebookData,
@@ -156,7 +156,7 @@ export const assembleNotebook = async (
   envName: string,
   tolerationSettings: NotebookTolerationSettings,
 ): Promise<Notebook> => {
-  const { notebookSizeName, imageName, imageTagName, gpus, envVars } = data;
+  const { notebookSizeName, imageName, imageTagName, accelerator, envVars } = data;
 
   const notebookSize = getNotebookSize(notebookSizeName);
 
@@ -186,39 +186,34 @@ export const assembleNotebook = async (
   const tolerations: NotebookToleration[] = [];
 
   let affinity: NotebookAffinity = {};
-  if (gpus > 0) {
+  if (accelerator.count > 0 && accelerator.accelerator) {
     if (!resources.limits) {
       resources.limits = {};
     }
     if (!resources.requests) {
       resources.requests = {};
     }
-    resources.limits[LIMIT_NOTEBOOK_IMAGE_GPU] = gpus;
-    resources.requests[LIMIT_NOTEBOOK_IMAGE_GPU] = gpus;
-    tolerations.push({
-      effect: 'NoSchedule',
-      key: LIMIT_NOTEBOOK_IMAGE_GPU,
-      operator: 'Exists',
-    });
+    resources.limits[accelerator.accelerator.spec.identifier] = accelerator.count;
+    resources.requests[accelerator.accelerator.spec.identifier] = accelerator.count;
   } else {
-    affinity = {
-      nodeAffinity: {
-        preferredDuringSchedulingIgnoredDuringExecution: [
-          {
-            preference: {
-              matchExpressions: [
-                {
-                  key: 'nvidia.com/gpu.present',
-                  operator: 'NotIn',
-                  values: ['true'],
-                },
-              ],
-            },
-            weight: 1,
-          },
-        ],
-      },
-    };
+    // step type down to string to avoid type errors
+    const containerResourceKeys: string[] = Object.values(ContainerResourceAttributes);
+
+    Object.keys(resources.limits || {}).forEach((key) => {
+      if (!containerResourceKeys.includes(key)) {
+        delete resources.limits?.[key];
+      }
+    });
+
+    Object.keys(resources.requests || {}).forEach((key) => {
+      if (!containerResourceKeys.includes(key)) {
+        delete resources.requests?.[key];
+      }
+    });
+  }
+
+  if (accelerator.accelerator?.spec.tolerations) {
+    tolerations.push(...accelerator.accelerator.spec.tolerations);
   }
 
   if (tolerationSettings?.enabled) {
@@ -266,6 +261,7 @@ export const assembleNotebook = async (
         'notebooks.opendatahub.io/last-image-selection': imageSelection,
         'opendatahub.io/username': username,
         'kubeflow-resource-stopped': null,
+        'opendatahub.io/accelerator-name': accelerator.accelerator?.metadata.name || '',
       },
       name: name,
       namespace: namespace,

backend/src/utils/resourceUtils.ts

@@ -33,6 +33,7 @@ import {
 } from './componentUtils';
 import { createCustomError } from './requestUtils';
 import { getAcceleratorNumbers } from '../routes/api/accelerators/acceleratorUtils';
+import { getNotebooks } from './notebookUtils';
 
 const dashboardConfigMapName = 'odh-dashboard-config';
 const consoleLinksGroup = 'console.openshift.io';
@@ -678,7 +679,7 @@ export const cleanupGPU = async (fastify: KubeFastifyInstance): Promise<void> =>
         };
 
         try {
-          await await fastify.kube.customObjectsApi.createNamespacedCustomObject(
+          await fastify.kube.customObjectsApi.createNamespacedCustomObject(
             'dashboard.opendatahub.io',
             'v1alpha',
             fastify.kube.namespace,
@@ -688,7 +689,34 @@ export const cleanupGPU = async (fastify: KubeFastifyInstance): Promise<void> =>
         } catch (e) {
           // If bad detection — exit early and dont create config
           throw 'Unable to add migrated-gpu accelerator profile: ' + e.toString()
-        }      
+        }
+
+        // update already running notebooks to use the new profile
+        const notebooks = await getNotebooks(fastify, fastify.kube.namespace)
+        notebooks.items.forEach(async (notebook) => {
+          const gpuCount = notebook.spec.template.spec.containers[0].resources?.limits?.['nvidia.com/gpu']
+          if (gpuCount) {
+            notebook.metadata.annotations = {
+              ...notebook.metadata.annotations,
+              'opendatahub.io/recommended-accelerators' : 'migrated-gpu'
+            }
+            await fastify.kube.customObjectsApi.patchNamespacedCustomObject(
+              'kubeflow.org',
+              'v1',
+              fastify.kube.namespace,
+              'notebooks',
+              notebook.metadata.name,
+              notebook,
+              undefined,
+              undefined,
+              undefined,
+              {
+                headers: { 'Content-type': PatchUtils.PATCH_FORMAT_JSON_MERGE_PATCH },
+              },
+            ) 
+          }
+        }
+        )
       };          
     }
 

frontend/src/api/k8s/utils.ts

@@ -31,7 +31,7 @@ export const assemblePodSpecOptions = (
     resources.requests[accelerator.accelerator.spec.identifier] = accelerator.count;
   } else {
     // step type down to string to avoid type errors
-    const containerResourceKeys: string[] = Object.keys(ContainerResourceAttributes);
+    const containerResourceKeys: string[] = Object.values(ContainerResourceAttributes);
 
     Object.keys(resources.limits || {}).forEach((key) => {
       if (!containerResourceKeys.includes(key)) {

frontend/src/pages/notebookController/screens/server/AcceleratorSelectField.tsx

@@ -48,7 +48,7 @@ const AcceleratorSelectField: React.FC<AcceleratorSelectFieldProps> = ({
       )?.[1];
 
       if (detectedAcceleratorCount === undefined) {
-        return `No accelerator detected with the identifier ${accelerator?.spec.identifier} detected.`;
+        return `No accelerator detected with the identifier ${accelerator?.spec.identifier}.`;
       } else if (newSize > detectedAcceleratorCount) {
         return `Only ${detectedAcceleratorCount} accelerator${
           detectedAcceleratorCount > 1 ? 's' : ''
@@ -61,9 +61,7 @@ const AcceleratorSelectField: React.FC<AcceleratorSelectFieldProps> = ({
   );
 
   React.useEffect(() => {
-    if (acceleratorCount > 0) {
-      setAcceleratorCountWarning(validateAcceleratorCount(acceleratorCount));
-    }
+    setAcceleratorCountWarning(validateAcceleratorCount(acceleratorCount));
   }, [acceleratorCount, validateAcceleratorCount]);
 
   const [acceleratorCountWarning, setAcceleratorCountWarning] = React.useState(

frontend/src/pages/projects/screens/spawner/spawnerUtils.ts

@@ -1,6 +1,6 @@
 import * as React from 'react';
 import compareVersions from 'compare-versions';
-import { BYONImage, K8sResourceCommon, NotebookSize, Volume, VolumeMount } from '~/types';
+import { BYONImage, NotebookSize, Volume, VolumeMount } from '~/types';
 import {
   BuildKind,
   ImageStreamKind,
@@ -414,7 +414,7 @@ export const isInvalidBYONImageStream = (imageStream: ImageStreamKind) => {
   );
 };
 
-export const convertBYONImageToK8sResource = (image: BYONImage): K8sResourceCommon => ({
+export const convertBYONImageToK8sResource = (image: BYONImage) => ({
   kind: 'ImageStream',
   apiVersion: 'image.openshift.io/v1',
   metadata: {