Merge pull request #60 from kubeslice-ent/release-v0.8.0

Release v0.8.0
kubeslice-ent · Oct 29, 2024 · 1ebecd2 · 1ebecd2
2 parents d14f73a + 1a84d3e
commit 1ebecd2
Show file tree

Hide file tree

Showing 18 changed files with 351 additions and 27 deletions.
diff --git a/charts/kubeslice-controller-egs/Chart.yaml b/charts/kubeslice-controller-egs/Chart.yaml
@@ -38,5 +38,5 @@ dependencies:
     repository: file://./charts/prometheus
     version: 25.*.*
     condition: kubeslice.prometheus.enabled
-version: 0.7.2
-appVersion: 0.7.2
+version: 0.8.0
+appVersion: 0.8.0
diff --git a/.../kubeslice-controller-egs/templates/crd.clustergpuallocations.inventory.kubeslice.io.yaml b/.../kubeslice-controller-egs/templates/crd.clustergpuallocations.inventory.kubeslice.io.yaml
@@ -1,3 +1,4 @@
+---
 apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
@@ -102,9 +103,54 @@ spec:
                       type: integer
                   type: object
                 type: array
+              unmanagedNodeInventory:
+                items:
+                  properties:
+                    cloudProvider:
+                      type: string
+                    free:
+                      type: integer
+                    gpuModelName:
+                      type: string
+                    gpuNodeShape:
+                      type: string
+                    gpuPowerThreshold:
+                      type: string
+                    gpuSharingType:
+                      type: string
+                    gpuSlicingProfile:
+                      items:
+                        properties:
+                          memory:
+                            type: string
+                          numGPUs:
+                            type: integer
+                          profileName:
+                            type: string
+                        type: object
+                      type: array
+                    gpuTempThreshold:
+                      type: string
+                    instanceType:
+                      type: string
+                    memory:
+                      type: integer
+                    nodeHealth:
+                      enum:
+                      - Healthy
+                      - Unhealthy
+                      type: string
+                    nodeName:
+                      type: string
+                    region:
+                      type: string
+                    totalGPUs:
+                      type: integer
+                  type: object
+                type: array
             type: object
         type: object
     served: true
     storage: true
     subresources:
-      status: {}
+      status: {}
diff --git a/...lice-controller-egs/templates/crd.workerclustergpuallocations.inventory.kubeslice.io.yaml b/...lice-controller-egs/templates/crd.workerclustergpuallocations.inventory.kubeslice.io.yaml
@@ -1,3 +1,4 @@
+---
 apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
@@ -103,9 +104,54 @@ spec:
                       type: integer
                   type: object
                 type: array
+              unmanagedNodeInventory:
+                items:
+                  properties:
+                    cloudProvider:
+                      type: string
+                    free:
+                      type: integer
+                    gpuModelName:
+                      type: string
+                    gpuNodeShape:
+                      type: string
+                    gpuPowerThreshold:
+                      type: string
+                    gpuSharingType:
+                      type: string
+                    gpuSlicingProfile:
+                      items:
+                        properties:
+                          memory:
+                            type: string
+                          numGPUs:
+                            type: integer
+                          profileName:
+                            type: string
+                        type: object
+                      type: array
+                    gpuTempThreshold:
+                      type: string
+                    instanceType:
+                      type: string
+                    memory:
+                      type: integer
+                    nodeHealth:
+                      enum:
+                      - Healthy
+                      - Unhealthy
+                      type: string
+                    nodeName:
+                      type: string
+                    region:
+                      type: string
+                    totalGPUs:
+                      type: integer
+                  type: object
+                type: array
             type: object
         type: object
     served: true
     storage: true
     subresources:
-      status: {}
+      status: {}
diff --git a/charts/kubeslice-controller-egs/templates/gpr-manager-deployment.yaml b/charts/kubeslice-controller-egs/templates/gpr-manager-deployment.yaml
@@ -3,7 +3,7 @@ kind: Deployment
 metadata:
   labels:
     control-plane: gpr-controller-manager
-  name: gpr-manager
+  name: egs-gpr-manager
   namespace: {{ .Release.Namespace }}
 spec:
   replicas: 1
@@ -35,6 +35,8 @@ spec:
           value: "{{ .Values.egs.gprManager.qmgrCheckPeriod }}"
         - name: MAX_DELAYED_COUNT
           value: "{{ .Values.egs.gprManager.maxDelayedCount }}"
+        - name: WAIT_TIME_REFRESH_PERIOD
+          value: "{{ .Values.egs.gprManager.refreshWaitTimeInterval }}"
         image: "{{ .Values.global.imageRegistry }}/{{ .Values.egs.gprManager.image }}:{{ .Values.egs.gprManager.tag }}"
         imagePullPolicy: "{{ .Values.egs.gprManager.pullPolicy }}"
         livenessProbe:
@@ -75,3 +77,9 @@ spec:
           secret:
             defaultMode: 420
             secretName: gpr-admission-webhook-certs
+      {{- if and .Values.imagePullSecrets .Values.imagePullSecrets.repository .Values.imagePullSecrets.username .Values.imagePullSecrets.password }}
+      imagePullSecrets:
+        - name: kubeslice-image-pull-secret
+      {{- end }}
+      tolerations:
+        - operator: "Exists"
diff --git a/charts/kubeslice-controller-egs/templates/gpr-manager-service.yaml b/charts/kubeslice-controller-egs/templates/gpr-manager-service.yaml
@@ -3,7 +3,7 @@ kind: Service
 metadata:
   labels:
     app: gpr-manager
-  name: gpr-manager
+  name: egs-gpr-manager
   namespace: {{ .Release.Namespace }}
 spec:
   ports:

diff --git a/charts/kubeslice-controller-egs/templates/inventory-manager-deployment.yaml b/charts/kubeslice-controller-egs/templates/inventory-manager-deployment.yaml
@@ -3,7 +3,7 @@ kind: Deployment
 metadata:
   labels:
     control-plane: inventory-controller-manager
-  name: inventory-controller-manager
+  name: egs-inventory-controller-manager
   namespace: {{ .Release.Namespace }}
 spec:
   replicas: 1
@@ -60,3 +60,9 @@ spec:
         runAsNonRoot: true
       serviceAccountName: inventory-manager-access
       terminationGracePeriodSeconds: 10
+      {{- if and .Values.imagePullSecrets .Values.imagePullSecrets.repository .Values.imagePullSecrets.username .Values.imagePullSecrets.password }}
+      imagePullSecrets:
+        - name: kubeslice-image-pull-secret
+      {{- end }}
+      tolerations:
+        - operator: "Exists"
diff --git a/charts/kubeslice-controller-egs/templates/inventory-manager-service.yaml b/charts/kubeslice-controller-egs/templates/inventory-manager-service.yaml
@@ -3,7 +3,7 @@ kind: Service
 metadata:
   labels:
     app: inventory-manager
-  name: inventory-manager
+  name: egs-inventory-manager
   namespace: {{ .Release.Namespace }}
 spec:
   ports:

diff --git a/charts/kubeslice-controller-egs/templates/queue-manager.yaml b/charts/kubeslice-controller-egs/templates/queue-manager.yaml
@@ -1,7 +1,7 @@
 apiVersion: apps/v1
 kind: Deployment
 metadata:
-  name: queue-manager
+  name: egs-queue-manager
   namespace: {{ .Release.Namespace }}
   labels:
     app: "queue-manager"
@@ -43,6 +43,8 @@ spec:
         - name: kubeslice-image-pull-secret
       {{- end }}
       schedulerName: default-scheduler
+      tolerations:
+        - operator: "Exists"
   strategy:
     type: RollingUpdate
     rollingUpdate:

diff --git a/charts/kubeslice-controller-egs/values.yaml b/charts/kubeslice-controller-egs/values.yaml
@@ -2,11 +2,11 @@
 # This is a YAML-formatted file.
 # Declare variables to be passed into your templates.
 
-# if you're installing in openshift cluster make this variable true
 global:
   imageRegistry: docker.io/aveshasystems
   # Profile settings (e.g., for OpenShift)
   profile:
+# if you're installing in openshift cluster make this variable true
     openshift: false
   kubeTally:
     # Enable or disable KubeTally
@@ -55,7 +55,7 @@ kubeslice:
     # Image name for the KubeSlice controller
     image: kubeslice-controller-ent-egs
     # Image tag for the KubeSlice controller
-    tag: 0.7.0
+    tag: 0.8.0
     # Image pull policy for the KubeSlice controller
     pullPolicy: IfNotPresent
     # Configuration for kubeTally, which handles chargeback and metrics
@@ -107,21 +107,24 @@ egs:
     # Image name for the gpr-manager
     image: gpr-manager
     # Image tag for the gpr-manager
-    tag: 0.7.1
+    tag: 0.8.0
     imagePullPolicy: IfNotPresent
     # Period for checking if there is queue processing needed (in seconds)
     qmgrCheckPeriod: 60
     # Number of times a GPR can be delayed if there is not enough inventory
     maxDelayedCount: 10
+    # Period for refreshing the GPRs
+    refreshWaitTimeInterval: 300
 
   queueManager:
+    imageRegistry: aveshasystems # global registry will be choosen if removed 
     # Image name for the gpr-manager
     image: queue-manager
     # Image tag for the gpr-manager
-    tag: 0.7.0
+    tag: 0.8.0
     imagePullPolicy: IfNotPresent
     service:
-      name: queue-manager
+      name: egs-queue-manager
       port: 80
     # turn off reflection
     appEnv: "production"
@@ -130,11 +133,11 @@ egs:
     # Image name for inventory-manager
     image: inventory-manager
     # Image tag for inventory-manager
-    tag: 0.7.1
+    tag: 0.8.0
     imagePullPolicy: IfNotPresent
     logLevel: INFO
     service:
-      name: inventory-manager
+      name: egs-inventory-manager
       port: 80
 
 prometheus:

diff --git a/charts/kubeslice-ui-egs/Chart.yaml b/charts/kubeslice-ui-egs/Chart.yaml
@@ -28,5 +28,5 @@ keywords:
 - application
 kubeVersion: '>= 1.19.0-0'
 home: https://avesha.io/products/product-slice
-version: 0.7.2
-appVersion: 0.7.2
+version: 0.8.0
+appVersion: 0.8.0