Add overprovisioner (#52)

charts/overprovisioner/.helmignore

@@ -0,0 +1,23 @@
+# Patterns to ignore when building packages.
+# This supports shell glob matching, relative path matching, and
+# negation (prefixed with !). Only one pattern per line.
+.DS_Store
+# Common VCS dirs
+.git/
+.gitignore
+.bzr/
+.bzrignore
+.hg/
+.hgignore
+.svn/
+# Common backup files
+*.swp
+*.bak
+*.tmp
+*.orig
+*~
+# Various IDEs
+.project
+.idea/
+*.tmproj
+.vscode/

charts/overprovisioner/Chart.yaml

@@ -0,0 +1,7 @@
+apiVersion: v2
+name: overprovisioner
+description: Overprovisioner for cluster-autoscaler.
+type: application
+
+version: 0.1.0
+appVersion: 0.0.1

charts/overprovisioner/templates/deployment.yaml

@@ -0,0 +1,55 @@
+{{- range $g := .Values.nodeGroups }}
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: overprovisioner-{{ $g.nickname }}
+  {{- with $g.deploymentLabels }}
+  labels:
+    {{- toYaml . | nindent 4 }}
+  {{- end }}
+spec:
+  replicas: {{ $g.count | int | toYaml }}
+  selector:
+    matchLabels:
+      app.kubernetes.io/name: {{ $.Chart.Name | toString | toYaml }}
+      app.kubernetes.io/instance: {{ $.Release.Name | toString | toYaml }}
+      overprovisioner.v1.vessl.ai/instance: {{ $g.nickname }}
+  strategy:
+    type: Recreate
+  template:
+    metadata:
+      {{- with $g.podAnnotations }}
+      annotations:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+      labels:
+        app.kubernetes.io/name: {{ $.Chart.Name | toString | toYaml }}
+        app.kubernetes.io/instance: {{ $.Release.Name | toString | toYaml }}
+        overprovisioner.v1.vessl.ai/instance: {{ $g.nickname }}
+        {{- with $g.podLabels }}
+        {{- toYaml . | nindent 8 }}
+        {{- end }}
+    spec:
+      priorityClassName: {{ $.Values.priorityClass.name }}
+      containers:
+        - name: {{ $.Chart.Name }}
+          image: "{{ $.Values.image.repository }}:{{ $.Values.image.tag }}"
+          imagePullPolicy: {{ $.Values.image.pullPolicy }}
+          {{- with $g.resources }}
+          resources:
+            {{- toYaml . | nindent 12 }}
+          {{- end }}
+      {{- with $g.nodeSelectorLabels }}
+      nodeSelector:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+      {{- with $g.affinity }}
+      affinity:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+      {{- with $g.tolerations }}
+      tolerations:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+---
+{{- end }}

charts/overprovisioner/templates/priority-class.yaml

@@ -0,0 +1,12 @@
+{{- if .Values.priorityClass.create }}
+apiVersion: scheduling.k8s.io/v1
+kind: PriorityClass
+metadata:
+  namespace: {{ .Release.Namespace }}
+  name: {{ .Values.priorityClass.name }}
+value: {{ .Values.priorityClass.value | int | toYaml }}
+description: |
+  Used by overprovisioner.
+  Pods with this priority class can be evicted (preempted) to yield resources
+  whenever workload pods need to be scheduled.
+{{- end }}

charts/overprovisioner/values.yaml

@@ -0,0 +1,45 @@
+nodeGroups:
+  - nickname: l4-spot
+    count: 1
+    nodeSelector: {}
+    tolerations:
+      - key: nvidia.com/gpu
+        operator: Exists
+    affinity:
+      nodeAffinity:
+        requiredDuringSchedulingIgnoredDuringExecution:
+          nodeSelectorTerms:
+            - matchExpressions:
+                - key: v1.k8s.vessl.ai/v1.l4-1.mem-42.spot
+                  operator: In
+                  values:
+                    - 'true'
+    resources:
+      requests: { cpu: 6100m }
+
+    deploymentLabels: {}
+    podAnnotations: {}
+    podLabels: {}
+
+priorityClass:
+  create: true
+  name: overprovisioner-placeholder
+
+  # NOTE: value should be...
+  #
+  #   - LESS than workload pod's priority.
+  #     By default, we do not set PriorityClass to workload pods, and thus it
+  #     uses cluster's default PriorityClass if exists. In most of our configuration,
+  #     there is no default PriorityClass, and priority value falls back to zero.
+  #     Thus, in most configurations, this value should be < 0.
+  #
+  #   - GREATER OR EQUAL than cluster autoscaler's priority cutoff.
+  #     If the placeholder pods are deemed too little to cluster-autoscaler,
+  #     it won't allow the pods to hold onto resources and evict the nodes.
+  #     Default cutoff is -10, so this value should be >= -10 in most configurations.
+  value: -5
+
+image:
+  repository: k8s.gcr.io/pause
+  tag: 3.9
+  pullPolicy: IfNotPresent