Add templated Ray cluster

iterative · Aug 16, 2023 · 5630521 · 5630521
1 parent a56a772
commit 5630521
Show file tree

Hide file tree

Showing 6 changed files with 111 additions and 2 deletions.
diff --git a/.github/workflows/helm-lint-and-install.yaml b/.github/workflows/helm-lint-and-install.yaml
@@ -37,6 +37,7 @@ jobs:
       - name: Add external repositories
         run: |
           helm repo add bitnami https://charts.bitnami.com/bitnami
+          helm repo add kuberay https://ray-project.github.io/kuberay-helm
 
       - name: Run chart-testing (list-changed)
         id: list-changed

diff --git a/charts/studio/Chart.lock b/charts/studio/Chart.lock
@@ -8,5 +8,8 @@ dependencies:
 - name: postgresql
   repository: https://charts.bitnami.com/bitnami
   version: 11.9.13
-digest: sha256:17b45dd98206bc3856d787b7399053996e6039ec8b52c9e00a501e7fcd896765
-generated: "2023-07-28T20:11:07.578307878Z"
+- name: kuberay-operator
+  repository: https://ray-project.github.io/kuberay-helm
+  version: 0.6.0
+digest: sha256:3b2dc8c1ca3e1841605034aca0e0dd7119c1fbcf17eecc47c997bf4510bce4cf
+generated: "2023-08-16T00:49:28.269211675Z"
diff --git a/charts/studio/Chart.yaml b/charts/studio/Chart.yaml
@@ -20,3 +20,7 @@ dependencies:
     condition: postgresql.enabled
     version: "11.9.13"
     repository: "https://charts.bitnami.com/bitnami"
+  - name: kuberay-operator
+    condition: ray.enabled
+    version: "0.6.0"
+    repository: "https://ray-project.github.io/kuberay-helm"
diff --git a/charts/studio/templates/configmap-studio.yaml b/charts/studio/templates/configmap-studio.yaml
@@ -121,6 +121,9 @@ data:
   DQL_UDF_ENABLED: {{ $dvcx.udfEnabled | default "False" | quote }}
   DVCX_ENABLED: {{ $dvcx.enabled | default "False" | quote }}
   DVCX_UDF_ENABLED: {{ $dvcx.udfEnabled | default "False" | quote }}
+  {{- if .Values.ray.enabled }}
+  DVCX_RAY_URL: "http://{{ .Values.ray.fullnameOverride }}-head-svc.{{ .Release.Namespace }}.svc.cluster.local:8265"
+  {{- end }}
 
   DQL_ROOT_DIR: {{ $dvcx.rootDir | default "/tmp" | quote }}
   DQL_CH_HOST: {{ $dvcxClickhouse.host | default "" | quote }}

diff --git a/charts/studio/templates/raycluster-studio.yaml b/charts/studio/templates/raycluster-studio.yaml
@@ -0,0 +1,70 @@
+{{- if .Values.ray.enabled }}
+apiVersion: ray.io/v1alpha1
+kind: RayCluster
+metadata:
+  labels:
+    controller-tools.k8s.io: "1.0"
+  name: {{ .Release.Name }}-ray
+spec:
+  rayVersion: {{ .Values.ray.version }}
+  enableInTreeAutoscaling: true
+  autoscalerOptions:
+    upscalingMode: Default
+    idleTimeoutSeconds: 60
+    imagePullPolicy: IfNotPresent
+    resources:
+      resources:
+        cpu: 500m
+        memory: 500M
+      resources:
+        cpu: 500m
+        memory: 500M
+  headGroupSpec:
+    rayStartParams:
+      dashboard-host: 0.0.0.0
+    template:
+      spec:
+        containers:
+        - name: head
+          image: {{ .Values.ray.image }}
+          resources:
+            requests:
+              cpu: {{ .Values.ray.head.resources.cpu }}
+              memory: {{ .Values.ray.head.resources.memory }}
+            limits:
+              cpu: {{ .Values.ray.head.resources.cpu }}
+              memory: {{ .Values.ray.head.resources.memory }}
+          ports:
+          - containerPort: 6379
+            name: gcs
+          - containerPort: 8265
+            name: dashboard
+          - containerPort: 10001
+            name: client
+          lifecycle:
+            preStop:
+              exec:
+                command: ["/bin/sh","-c","ray stop"]
+  workerGroupSpecs:
+  - replicas: {{ .Values.ray.worker.replicas.minimum }}
+    minReplicas: {{ .Values.ray.worker.replicas.minimum }}
+    maxReplicas: {{ .Values.ray.worker.replicas.maximum }}
+    groupName: workers
+    rayStartParams: {}
+    template:
+      spec:
+        containers:
+        - name: worker
+          image: {{ .Values.ray.image }}
+          resources:
+            requests:
+              cpu: {{ .Values.ray.worker.resources.cpu }}
+              memory: {{ .Values.ray.worker.resources.memory }}
+            limits:
+              cpu: {{ .Values.ray.worker.resources.cpu }}
+              memory: {{ .Values.ray.worker.resources.memory }}
+          lifecycle:
+            preStop:
+              exec:
+                command: ["/bin/sh", "-c", "ray stop"]
+{{- end }}
diff --git a/charts/studio/values.yaml b/charts/studio/values.yaml
@@ -335,6 +335,34 @@ postgresql:
         # -- Postgres database
         database: iterativeai
 
+ray:
+  # -- Ray enabled
+  enabled: false
+  # -- Ray name override
+  fullnameOverride: {{ .Release.Name }}-ray
+
+  # -- Container image used for Ray nodes
+  image: rayproject/ray:2.4.0-py310-cpu
+
+  # -- Ray version
+  version: 2.4.0
+
+  # Recommendation: adjust resources (i.e. requests & limits) so a single
+  # head or worker is scheduled on each Kubernetes node.
+
+  head:
+    resources:
+      cpu: 2
+      memory: 8G
+
+  worker:
+    resources:
+      cpu: 2
+      memory: 8G
+    replicas:
+      minimum: 1
+      maximum: 4
+
 # -- Studio UI settings group
 studioUi:
   # -- Additional environment variables for ui pods