feat: Add k8s glue agent deployment (#49)

charts/clearml/Chart.yaml

@@ -2,7 +2,7 @@ apiVersion: v2
 name: clearml
 description: MLOps platform
 type: application
-version: "3.3.1"
+version: "3.4.0"
 appVersion: "1.1.1"
 home: https://clear.ml
 icon: https://raw.githubusercontent.com/allegroai/clearml/master/docs/clearml-logo.svg

charts/clearml/README.md

@@ -1,6 +1,6 @@
 # ClearML Ecosystem for Kubernetes
 
-![Version: 3.3.1](https://img.shields.io/badge/Version-3.3.1-informational?style=flat-square) ![Type: application](https://img.shields.io/badge/Type-application-informational?style=flat-square) ![AppVersion: 1.1.1](https://img.shields.io/badge/AppVersion-1.1.1-informational?style=flat-square)
+![Version: 3.4.0](https://img.shields.io/badge/Version-3.4.0-informational?style=flat-square) ![Type: application](https://img.shields.io/badge/Type-application-informational?style=flat-square) ![AppVersion: 1.1.1](https://img.shields.io/badge/AppVersion-1.1.1-informational?style=flat-square)
 
 MLOps platform
 
@@ -163,6 +163,15 @@ For detailed instructions, see the [Optional Configuration](https://github.com/a
 | agentGroups.agent-group-gpu.replicaCount | int | `0` |  |
 | agentGroups.agent-group-gpu.tolerations | list | `[]` |  |
 | agentGroups.agent-group-gpu.updateStrategy | string | `"Recreate"` |  |
+| agentk8sglue.defaultDockerImage | string | `"nvidia/cuda:11.3.1-cudnn8-runtime-ubuntu18.04"` |  |
+| agentk8sglue.enabled | bool | `false` |  |
+| agentk8sglue.id | string | `"k8s-agent"` |  |
+| agentk8sglue.image.repository | string | `"allegroai/clearml-agent-k8s"` |  |
+| agentk8sglue.image.tag | string | `"aws-latest-1.21"` |  |
+| agentk8sglue.maxPods | int | `10` |  |
+| agentk8sglue.podTemplate.nodeSelector | object | `{}` |  |
+| agentk8sglue.podTemplate.tolerations | object | `{}` |  |
+| agentk8sglue.queue | string | `"aws-instances"` |  |
 | agentservices.affinity | object | `{}` |  |
 | agentservices.agentVersion | string | `""` |  |
 | agentservices.awsAccessKeyId | string | `nil` |  |

charts/clearml/templates/configmap-agentk8s-template.yaml

@@ -0,0 +1,35 @@
+{{- if .Values.agentk8sglue.enabled }}
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: k8sagent-pod-template
+data:
+  template.yaml: |
+    apiVersion: v1
+    metadata:
+      namespace: {{ .Release.namespace }}
+    spec:
+      containers:
+      - env: 
+        - name: CLEARML_API_HOST
+          value: "http://{{ include "clearml.fullname" . }}-apiserver:{{ .Values.clearml.apiserver.service.port }}"
+        - name: CLEARML_WEB_HOST
+          value: "http://{{ include "clearml.fullname" . }}-webserver"
+        - name: CLEARML_FILES_HOST
+          value: "http://{{ include "clearml.fullname" . }}-fileserver:{{ .Values.clearml.fileserver.service.port }}"
+        - name: CLEARML_API_ACCESS_KEY
+          valueFrom:
+            secretKeyRef:
+              name: clearml-conf
+              key: apiserver_key
+        - name: CLEARML_API_SECRET_KEY
+          valueFrom:
+            secretKeyRef:
+              name: clearml-conf
+              key: apiserver_secret
+      tolerations:
+        {{- toYaml .Values.agentk8sglue.podTemplate.tolerations | nindent 8 }}
+      nodeSelector:
+        {{- toYaml .Values.agentk8sglue.podTemplate.nodeSelector | nindent 8 }}
+{{- end }}
+

charts/clearml/templates/deployment-agentk8s.yaml

@@ -0,0 +1,62 @@
+{{- if .Values.agentk8sglue.enabled }}
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: "{{ include "clearml.fullname" . }}-k8sagent"
+  labels:
+    app: k8sagent
+spec:
+  replicas: 1
+  selector:
+    matchLabels:
+      app: k8sagent
+  template:
+    metadata:
+      labels:
+        app: k8sagent
+    spec:
+      containers:
+      - name: k8s-glue-container
+        image: "{{ .Values.agentk8sglue.image.repository }}:{{ .Values.agentk8sglue.image.tag }}"
+        imagePullPolicy: Always
+        command: ["/bin/bash", "-c", "export PATH=$PATH:$HOME/bin; source /root/.bashrc && /root/entrypoint.sh"]
+        volumeMounts:
+          - name: k8sagent-pod-template
+            mountPath: /root/template
+        env:
+          - name: CLEARML_API_HOST
+            value: "http://{{ include "clearml.fullname" . }}-apiserver:{{ .Values.clearml.apiserver.service.port }}"
+          - name: CLEARML_WEB_HOST
+            value: "http://{{ include "clearml.fullname" . }}-webserver"
+          - name: CLEARML_FILES_HOST
+            value: "http://{{ include "clearml.fullname" . }}-fileserver:{{ .Values.clearml.fileserver.service.port }}"
+          - name: K8S_GLUE_MAX_PODS
+            value: "{{.Values.agentk8sglue.maxPods}}"
+          - name: K8S_GLUE_QUEUE
+            value: "{{.Values.agentk8sglue.queue}}"
+          - name: K8S_GLUE_EXTRA_ARGS
+            value: "--template-yaml /root/template/template.yaml"
+          - name: CLEARML_API_ACCESS_KEY
+            valueFrom:
+              secretKeyRef:
+                name: clearml-conf
+                key: apiserver_key
+          - name: CLEARML_API_SECRET_KEY
+            valueFrom:
+              secretKeyRef:
+                name: clearml-conf
+                key: apiserver_secret
+          - name: CLEARML_WORKER_ID
+            value: "{{.Values.agentk8sglue.id}}"
+          - name: CLEARML_AGENT_UPDATE_REPO
+            value: ""
+          - name: FORCE_CLEARML_AGENT_REPO
+            value: ""
+          - name: CLEARML_DOCKER_IMAGE
+            value: "{{.Values.agentk8sglue.defaultDockerImage}}"
+      volumes:
+        - name: k8sagent-pod-template
+          configMap: 
+            name: k8sagent-pod-template
+{{- end }}
+

charts/clearml/templates/rbac-agentk8s.yaml

@@ -0,0 +1,25 @@
+{{- if .Values.agentk8sglue.enabled }}
+apiVersion: rbac.authorization.k8s.io/v1
+kind: Role
+metadata:
+  name: k8sagent-pods-access
+rules:
+  - apiGroups:
+      - ""
+    resources:
+      - pods
+    verbs: ["get", "list", "watch", "create", "patch"]
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: RoleBinding
+metadata:
+  name: k8sagent-pods-access
+subjects:
+  - kind: ServiceAccount
+    name: default
+    namespace: {{ .Release.namespace }}
+roleRef:
+  apiGroup: rbac.authorization.k8s.io
+  kind: Role
+  name: k8sagent-pods-access
+{{- end }}

charts/clearml/values.yaml

@@ -294,6 +294,27 @@ agentGroups:
 
     affinity: {}
 
+# This agent will spawn queued experiments in new pods, a good use case is to combine this with
+# GPU autoscaling nodes.
+# https://github.com/allegroai/clearml-agent/tree/master/docker/k8s-glue
+agentk8sglue:
+  enabled: false
+  image:
+    repository: "allegroai/clearml-agent-k8s"
+    tag: "aws-latest-1.21"
+  maxPods: 10
+  defaultDockerImage: nvidia/cuda:11.3.1-cudnn8-runtime-ubuntu18.04  # default docker image that is spawned as new pod
+  queue: aws-instances  # create this queue manually in the UI first for it to work
+  id: k8s-agent
+  podTemplate:
+    tolerations: {}
+      # - key: "nvidia.com/gpu"
+      #   operator: Exists
+      #   effect: "NoSchedule"
+    nodeSelector: {}
+      # fleet: gpu-nodes
+
+
 externalServices:
   # -- Existing ElasticSearch Hostname to use if elasticsearch.enabled is false
   elasticsearchHost: ""