feat(helm,docker-compose): add registry service

.env

@@ -192,3 +192,9 @@ SOCAT_IMAGE=alpine/socat
 SOCAT_VERSION=1.7.4.4-r0
 SOCAT_HOST=socat
 SOCAT_PORT=2375
+
+# registry
+REGISTRY_IMAGE=registry
+REGISTRY_VERSION=2.8.3
+REGISTRY_HOST=registry
+REGISTRY_PORT=5000

Makefile

@@ -47,13 +47,13 @@ HELM_RELEASE_NAME := core
 
 .PHONY: all
 all:			## Launch all services with their up-to-date release version
+	@docker inspect --type=image instill/ray:${RAY_RELEASE_TAG} >/dev/null 2>&1 || printf "\033[1;33mINFO:\033[0m This may take a while due to the enormous size of the Ray server image, but the image pulling process should be just a one-time effort.\n" && sleep 5
 	@make build-release BUILD=${BUILD}
 	@if [ ! -f "$$(echo ${SYSTEM_CONFIG_PATH}/user_uid)" ]; then \
 		mkdir -p ${SYSTEM_CONFIG_PATH} && \
 		docker run --rm --name uuidgen ${INSTILL_CORE_IMAGE_NAME}:${INSTILL_CORE_VERSION} uuidgen > ${SYSTEM_CONFIG_PATH}/user_uid; \
 	fi
 ifeq (${NVIDIA_GPU_AVAILABLE}, true)
-	@docker inspect --type=image instill/ray:${RAY_SERVER_VERSION} >/dev/null 2>&1 || printf "\033[1;33mINFO:\033[0m This may take a while due to the enormous size of the Ray server image, but the image pulling process should be just a one-time effort.\n" && sleep 5
 	@cat docker-compose-nvidia.yml | yq '.services.ray_server.deploy.resources.reservations.devices[0].device_ids |= (strenv(NVIDIA_VISIBLE_DEVICES) | split(",")) | ..style="double"' | \
 		EDITION=$${EDITION:=local-ce} DEFAULT_USER_UID=$$(cat ${SYSTEM_CONFIG_PATH}/user_uid) RAY_RELEASE_TAG=${RAY_RELEASE_TAG} docker compose ${COMPOSE_FILES} -f - up -d --quiet-pull
 else
@@ -62,13 +62,13 @@ endif
 
 .PHONY: latest
 latest:			## Lunch all dependent services with their latest codebase
+	@docker inspect --type=image instill/ray:${RAY_LATEST_TAG} >/dev/null 2>&1 || printf "\033[1;33mINFO:\033[0m This may take a while due to the enormous size of the Ray server image, but the image pulling process should be just a one-time effort.\n" && sleep 5
 	@make build-latest PROFILE=${PROFILE} BUILD=${BUILD}
 	@if [ ! -f "$$(echo ${SYSTEM_CONFIG_PATH}/user_uid)" ]; then \
 		mkdir -p ${SYSTEM_CONFIG_PATH} && \
 		docker run --rm --name uuidgen ${INSTILL_CORE_IMAGE_NAME}:latest uuidgen > ${SYSTEM_CONFIG_PATH}/user_uid; \
 	fi
 ifeq (${NVIDIA_GPU_AVAILABLE}, true)
-	@docker inspect --type=image instill/ray:${RAY_LATEST_TAG} >/dev/null 2>&1 || printf "\033[1;33mINFO:\033[0m This may take a while due to the enormous size of the Ray server image, but the image pulling process should be just a one-time effort.\n" && sleep 5
 	@cat docker-compose-nvidia.yml | yq '.services.ray_server.deploy.resources.reservations.devices[0].device_ids |= (strenv(NVIDIA_VISIBLE_DEVICES) | split(",")) | ..style="double"' | \
 		COMPOSE_PROFILES=${PROFILE} EDITION=$${EDITION:=local-ce:latest} DEFAULT_USER_UID=$$(cat ${SYSTEM_CONFIG_PATH}/user_uid) RAY_LATEST_TAG=${RAY_LATEST_TAG} docker compose ${COMPOSE_FILES} -f docker-compose-latest.yml -f - up -d --quiet-pull
 else

charts/core/templates/_helpers.tpl

@@ -384,6 +384,21 @@ openfga
   {{- printf "%s-openfga" (include "core.fullname" .) -}}
 {{- end -}}
 
+{{/*
+registry
+*/}}
+{{- define "core.registry" -}}
+  {{- printf "%s-registry" (include "core.fullname" .) -}}
+{{- end -}}
+
+{{- define "core.registry.port" -}}
+  {{- printf "5000" -}}
+{{- end -}}
+
+{{- define "core.registry.metricsPort" -}}
+  {{- printf "5001" -}}
+{{- end -}}
+
 {{/*
 internal TLS secret names
 */}}
@@ -446,6 +461,10 @@ Persistent Volume Claims
   {{- printf "%s-ray-conda-data-volume" (include "core.fullname" .) -}}
 {{- end -}}
 
+{{- define "core.registryDataVolume" -}}
+  {{- printf "%s-registry-data-volume" (include "core.fullname" .) -}}
+{{- end -}}
+
 {{- define "core.databaseDataVolume" -}}
   {{- printf "%s-database-data-volume" (include "core.fullname" .) -}}
 {{- end -}}

charts/core/templates/model-backend/configmap.yaml

@@ -95,3 +95,6 @@ data:
     openfga:
       host: {{ template "core.openfga" . }}
       port: 8080
+    registry:
+      host: {{ template "core.registry" . }}
+      port: {{ template "core.registry.port" . }}

charts/core/templates/pvc.yaml

@@ -57,6 +57,37 @@ spec:
     {{- end }}
   {{- end }}
 {{- end }}
+{{- $registry := .Values.persistence.persistentVolumeClaim.registry -}}
+{{- if not $registry.existingClaim }}
+---
+kind: PersistentVolumeClaim
+apiVersion: v1
+metadata:
+  name: {{ template "core.registryDataVolume" . }}
+  annotations:
+  {{- range $key, $value := $registry.annotations }}
+    {{ $key }}: {{ $value | quote }}
+  {{- end }}
+  {{- if eq .Values.persistence.resourcePolicy "keep" }}
+    helm.sh/resource-policy: keep
+  {{- end }}
+  labels:
+    {{- include "core.labels" . | nindent 4 }}
+    app.kubernetes.io/component: registry
+spec:
+  accessModes:
+    - {{ $registry.accessMode }}
+  resources:
+    requests:
+      storage: {{ $registry.size }}
+  {{- if $registry.storageClass }}
+    {{- if eq "-" $registry.storageClass }}
+  storageClassName: ""
+    {{- else }}
+  storageClassName: {{ $registry.storageClass }}
+    {{- end }}
+  {{- end }}
+{{- end }}
 {{- $database := .Values.persistence.persistentVolumeClaim.database -}}
 {{- if and (not $database.existingClaim) .Values.database.enabled }}
 ---

charts/core/templates/ray-service/ray-service.yaml

@@ -55,9 +55,24 @@ spec:
               items:
                 - key: cp_conda_env.sh
                   path: cp_conda_env.sh
+          - name: podman-configmap
+            configMap:
+              name: podman
+              defaultMode: 0666
+              items:
+                - key: registries.conf
+                  path: registries.conf
+                - key: policy.json
+                  path: policy.json
+                - key: storage.conf
+                  path: storage.conf
         containers:
           - name: ray-head
             image: {{ .Values.rayService.image.repository }}:{{ .Values.rayService.image.tag }}
+            securityContext:
+              # for mounting /dev/fuse
+              # TODO: maybe implement a fuse-device-plugin-daemonset
+              privileged: true
             imagePullPolicy: Always
             {{- if .Values.rayService.headGroupSpec.resources }}
             resources:
@@ -70,11 +85,15 @@ spec:
                 value: http://core-grafana:80
               - name: RAY_PROMETHEUS_HOST
                 value: http://core-prometheus:9090
+              - name: RAY_worker_register_timeout_seconds
+                value: "360"
             volumeMounts:
               - mountPath: /ray-conda-pack
                 name: ray-conda
               - mountPath: /home/ray/script
                 name: cp-conda-env-configmap
+              - mountPath: /etc/containers/
+                name: podman-configmap
             ports:
               - containerPort: 6379
                 name: gcs-server
@@ -134,6 +153,15 @@ spec:
                 items:
                   - key: start_ray_serve.sh
                     path: start_ray_serve.sh
+            - name: podman-configmap
+              configMap:
+                name: podman
+                defaultMode: 0666
+                items:
+                  - key: registries.conf
+                    path: registries.conf
+                  - key: policy.json
+                    path: policy.json
           containers:
             - name: ray-worker
               image: {{ $.Values.rayService.image.repository }}:{{ $.Values.rayService.image.tag }}
@@ -162,6 +190,8 @@ spec:
                   name: start-ray-serve-configmap
                 - mountPath: /model-repository
                   name: model-repository
+                - mountPath: /etc/containers/
+                  name: podman-configmap
   {{- end }}
 ---
 apiVersion: v1
@@ -212,3 +242,31 @@ data:
     serve start --http-host=0.0.0.0 --grpc-port 9000 --grpc-servicer-functions ray_pb2_grpc.add_RayServiceServicer_to_server
 
     echo "INFO: Start ray serve"
+---
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: podman
+data:
+  registries.conf: |
+    unqualified-search-registries = ["{{ template "core.registry" . }}:{{ template "core.registry.port" . }}", "docker.io", "quay.io"]
+
+    [[registry]]
+    location = "{{ template "core.registry" . }}:{{ template "core.registry.port" . }}"
+    insecure = true
+  policy.json: |
+    {
+      "default": [
+        {
+          "type": "insecureAcceptAnything"
+        }
+      ],
+      "transports": {
+        "docker-daemon": {
+          "": [{ "type": "insecureAcceptAnything" }]
+        }
+      }
+    }
+  storage.conf: |
+    [storage]
+    driver = "overlay"

charts/core/templates/registry/configmap.yaml

@@ -0,0 +1,28 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: {{ template "core.registry" . }}
+  labels:
+    {{- include "core.labels" . | nindent 4 }}
+    app.kubernetes.io/component: registry
+data:
+  config.yaml: |+
+    version: {{ .Values.registry.config.version }}
+    log:
+      {{- toYaml .Values.registry.config.log | nindent 6 }}
+    storage:
+      {{- toYaml .Values.registry.config.storage | nindent 6 }}
+    http:
+      {{- toYaml .Values.registry.config.http | nindent 6 }}
+    redis:
+      {{- toYaml .Values.registry.config.redis | nindent 6 }}
+    health:
+      storagedriver:
+        enabled: true
+        interval: 10s
+        threshold: 3
+      tcp:
+        - addr: {{ template "core.redis.addr" . }}
+          timeout: 3s
+          interval: 30s
+          threshold: 3

charts/core/templates/registry/cronjob.yaml

@@ -0,0 +1,70 @@
+{{- if .Values.registry.garbageCollect.enabled }}
+apiVersion: batch/v1
+kind: CronJob
+metadata:
+  name: {{ template "core.registry" . }}-garbage-collector
+  labels:
+    {{- include "core.labels" . | nindent 4 }}
+    app.kubernetes.io/component: registry
+spec:
+  concurrencyPolicy: Forbid
+  schedule: {{ .Values.registry.garbageCollect.schedule | quote }}
+  jobTemplate:
+    metadata:
+      labels:
+        {{- include "core.matchLabels" . | nindent 8 }}
+        app.kubernetes.io/component: registry
+      annotations:
+        checksum/config: {{ include (print $.Template.BasePath "/registry/configmap.yaml") . | sha256sum }}
+        {{- with .Values.registry.podAnnotations }}
+        {{- toYaml . | nindent 8 }}
+        {{- end }}
+    spec:
+      template:
+        spec:
+          {{- if .Values.registry.serviceAccountName }}
+          serviceAccountName: {{ .Values.registry.serviceAccountName }}
+          {{- end }}
+          {{- with .Values.imagePullSecrets }}
+          imagePullSecrets:
+            {{- toYaml . | nindent 12 }}
+          {{- end }}
+          securityContext:
+            runAsUser: 65534
+            runAsGroup: 65534
+            fsGroup: 65534
+          containers:
+            - name: garbage-collect
+              image: {{ .Values.registry.image.repository }}:{{ .Values.registry.image.tag }}
+              imagePullPolicy: {{ .Values.registry.image.pullPolicy }}
+              command:
+              - /bin/registry
+              - garbage-collect
+              - --delete-untagged={{ .Values.registry.garbageCollect.deleteUntagged }}
+              - /etc/docker/registry/config.yml
+              securityContext:
+                runAsUser: 65534
+                runAsGroup: 65534
+                fsGroup: 65534
+              volumeMounts:
+                - name: config
+                  mountPath: {{ .Values.registry.configPath }}
+                  subPath: config.yaml
+          restartPolicy: OnFailure
+          {{- with .Values.registry.nodeSelector }}
+          nodeSelector:
+            {{- toYaml . | nindent 8 }}
+          {{- end }}
+          {{- with .Values.registry.affinity }}
+          affinity:
+            {{- toYaml . | nindent 8 }}
+          {{- end }}
+          {{- with .Values.registry.tolerations }}
+          tolerations:
+            {{- toYaml . | nindent 8 }}
+          {{- end }}
+          volumes:
+            - name: config
+              configMap:
+                name: {{ template "core.registry" . }}
+{{- end }}