Sync kubeflow training operator manifests v1.5.0 rc.0

README.md

@@ -41,7 +41,7 @@ This repo periodically syncs all official Kubeflow components from their respect
 
 | Component | Local Manifests Path | Upstream Revision |
 | - | - | - |
-| Training Operator | apps/training-operator/upstream | [v1.4.0](https://github.com/kubeflow/tf-operator/tree/v1.4.0/manifests) |
+| Training Operator | apps/training-operator/upstream | [v1.5.0-rc.0](https://github.com/kubeflow/training-operator/tree/v1.5.0-rc.0/manifests) |
 | Notebook Controller | apps/jupyter/notebook-controller/upstream | [v1.5.0](https://github.com/kubeflow/kubeflow/tree/v1.5.0/components/notebook-controller/config) |
 | Tensorboard Controller | apps/tensorboard/tensorboard-controller/upstream | [v1.5.0](https://github.com/kubeflow/kubeflow/tree/v1.5.0/components/tensorboard-controller/config) |
 | Central Dashboard | apps/centraldashboard/upstream | [v1.5.0](https://github.com/kubeflow/kubeflow/tree/v1.5.0/components/centraldashboard/manifests) |

apps/training-operator/upstream/base/deployment.yaml

@@ -40,12 +40,14 @@ spec:
               port: 8081
             initialDelaySeconds: 15
             periodSeconds: 20
+            timeoutSeconds: 3
           readinessProbe:
             httpGet:
               path: /readyz
               port: 8081
-            initialDelaySeconds: 5
-            periodSeconds: 10
+            initialDelaySeconds: 10
+            periodSeconds: 15
+            timeoutSeconds: 3
           resources:
             limits:
               cpu: 100m

apps/training-operator/upstream/overlays/kubeflow/kubeflow-training-roles.yaml

@@ -22,6 +22,7 @@ rules:
   - apiGroups:
       - kubeflow.org
     resources:
+      - mpijobs
       - tfjobs
       - pytorchjobs
       - mxjobs
@@ -37,6 +38,7 @@ rules:
   - apiGroups:
       - kubeflow.org
     resources:
+      - mpijobs/status
       - tfjobs/status
       - pytorchjobs/status
       - mxjobs/status
@@ -55,6 +57,7 @@ rules:
   - apiGroups:
       - kubeflow.org
     resources:
+      - mpijobs
       - tfjobs
       - pytorchjobs
       - mxjobs
@@ -66,6 +69,7 @@ rules:
   - apiGroups:
       - kubeflow.org
     resources:
+      - mpijobs/status
       - tfjobs/status
       - pytorchjobs/status
       - mxjobs/status

apps/training-operator/upstream/overlays/kubeflow/kustomization.yaml

@@ -6,5 +6,4 @@ resources:
   - kubeflow-training-roles.yaml
 images:
   - name: kubeflow/training-operator
-    newName: public.ecr.aws/j1r0q0g6/training/training-operator
-    newTag: "174e8813666951ded505daf334a37f60fd50c18d"
+    newTag: "v1-e1434f6"

apps/training-operator/upstream/overlays/standalone/kustomization.yaml

@@ -6,5 +6,4 @@ resources:
   - namespace.yaml
 images:
   - name: kubeflow/training-operator
-    newName: public.ecr.aws/j1r0q0g6/training/training-operator
-    newTag: "174e8813666951ded505daf334a37f60fd50c18d"
+    newTag: "v1-e1434f6"

hack/sync-training-operator-manifests.sh

@@ -55,8 +55,15 @@ cp $SRC_DIR/manifests $DST_DIR -r
 
 echo "Successfully copied all manifests."
 
+echo "Updating README..."
+SRC_TXT="\[.*\](https://github.com/kubeflow/training-operator/tree/.*/manifests)"
+DST_TXT="\[$COMMIT\](https://github.com/kubeflow/training-operator/tree/$COMMIT/manifests)"
+
+sed -i "s|$SRC_TXT|$DST_TXT|g" ${MANIFESTS_DIR}/README.md
+
 # DEV: Comment out these commands when local testing
 echo "Committing the changes..."
 cd $MANIFESTS_DIR
 git add apps
+git add README.md
 git commit -m "Update kubeflow/training-operator manifests from ${COMMIT}"