Merge 680444e into c47b082

wandb · Jan 13, 2025 · 02cda80 · 02cda80
2 parents c47b082 + 680444e
commit 02cda80
Show file tree

Hide file tree

Showing 20 changed files with 688 additions and 114 deletions.
diff --git a/charts/operator-wandb/Chart.lock b/charts/operator-wandb/Chart.lock
@@ -2,6 +2,9 @@ dependencies:
 - name: app
   repository: file://charts/app
   version: 0.1.0
+- name: wandb-base
+  repository: file://../wandb-base
+  version: 0.3.0
 - name: console
   repository: file://charts/console
   version: 0.1.0
@@ -55,6 +58,9 @@ dependencies:
   version: 0.1.0
 - name: wandb-base
   repository: file://../wandb-base
-  version: 0.1.1
-digest: sha256:49d1e4088513e55b9824284c9f5a80d3c70c9fab0195e6cc5650731fb2670a06
-generated: "2025-01-13T09:43:24.543619-08:00"
+  version: 0.3.0
+- name: wandb-base
+  repository: file://../wandb-base
+  version: 0.3.0
+digest: sha256:c0bbd695dffdd8b72a2453d6eb113bc048f7c89a31d2b929baa77cf5e77f6d63
+generated: "2025-01-13T12:35:58.707616-06:00"
diff --git a/charts/operator-wandb/Chart.yaml b/charts/operator-wandb/Chart.yaml
@@ -2,7 +2,7 @@ apiVersion: v2
 name: operator-wandb
 description: A Helm chart for deploying W&B to Kubernetes
 type: application
-version: 0.22.5
+version: 0.23.0
 appVersion: 1.0.0
 icon: https://wandb.ai/logo.svg
 
@@ -16,6 +16,11 @@ dependencies:
     version: "*.*.*"
     repository: file://charts/app
     condition: app.install
+  - name: wandb-base
+    alias: api
+    condition: global.beta.api.enabled
+    repository: file://../wandb-base
+    version: "*.*.*"
   - name: console
     version: "*.*.*"
     repository: file://charts/console
@@ -89,3 +94,8 @@ dependencies:
     condition: global.beta.glue.enabled
     repository: file://../wandb-base
     version: "*.*.*"
+  - name: wandb-base
+    alias: settingsMigrationHook
+    condition: settingsMigrationHook.install
+    repository: file://../wandb-base
+    version: "*.*.*"
diff --git a/charts/operator-wandb/charts/app/templates/deployment.yaml b/charts/operator-wandb/charts/app/templates/deployment.yaml
@@ -108,6 +108,13 @@ spec:
                   resource: limits.memory
             - name: GLUE_ENABLED
               value: "{{ not .Values.global.beta.glue.enabled }}"
+            ## TODO: Turn this on when we are ready to remove the gorilla service from wandb/local
+            # {{- if .Values.global.beta.api.enabled }}
+            # - name: GORILLA_ENABLED
+            #   value: "{{ not .Values.global.beta.api.enabled }}"
+            # - name: PROXY_PASS_BACKEND_HOST
+            #   value: "$({{.Release.Name | upper | replace "-" "_" }}_API_SERVICE_HOST):$({{.Release.Name | upper | replace "-" "_" }}_API_SERVICE_PORT)"
+            # {{- end }}
             - name: BUCKET_ACCESS_KEY
               valueFrom:
                 secretKeyRef:
@@ -309,6 +316,8 @@ spec:
 
             {{- include "app.extraEnv" (dict "global" $.Values.global "local" .Values) | nindent 12 }}
             {{- include "wandb.extraEnvFrom" (dict "root" $ "local" .) | nindent 12 }}
+          ## TODO: If we want to remove the api service from wandb/local, we can uncomment this if else block
+          # {{- if not .Values.global.beta.api.enabled }}
           livenessProbe:
             httpGet:
               path: /healthz
@@ -326,6 +335,28 @@ spec:
             initialDelaySeconds: 20
             periodSeconds: 5
             failureThreshold: 120
+          # {{- else }}
+          # # If we disabled the api, we still want to have a healthcheck fror the frontend/nginx
+          # # We also can't just disable the migrations which would make the health endpint return 200
+          # # because glue, parquet, and filemeta are still running in the local container.
+          # livenessProbe:
+          #   httpGet:
+          #     path: /proxy/healthz
+          #     port: http
+          # readinessProbe:
+          #   httpGet:
+          #     path: /proxy/healthz
+          #     port: http
+          #   initialDelaySeconds: 20
+          #   periodSeconds: 5
+          # startupProbe:
+          #   httpGet:
+          #     path: /proxy/healthz
+          #     port: http
+          #   initialDelaySeconds: 20
+          #   periodSeconds: 5
+          #   failureThreshold: 120
+          # {{- end }}
           # Increase the sleep before SIGTERM to 25s. I had this as 5s previously and it wasn't enough.
           lifecycle:
             preStop:

diff --git a/charts/operator-wandb/templates/_ingress.tpl b/charts/operator-wandb/templates/_ingress.tpl
@@ -54,11 +54,27 @@ It expects a dictionary with two entries:
       port: 
         number: 8080
       {{- end }}
+{{- if .Values.global.beta.api.enabled }}
+- pathType: Prefix
+  path: /api
+  backend:
+    service:
+      name: {{ $.Release.Name }}-api
+      port: 
+        number: 8080
+- pathType: Prefix
+  path: /graphql
+  backend:
+    service:
+      name: {{ $.Release.Name }}-api
+      port: 
+        number: 8080
+{{- end }}
 - pathType: Prefix
   path: /console
   backend:
     service:
       name: {{ $.Release.Name }}-console
       port:
         number: 8082
-{{- end }}
+{{- end }}
diff --git a/charts/operator-wandb/templates/bucket.yaml b/charts/operator-wandb/templates/bucket.yaml
@@ -4,6 +4,12 @@ metadata:
   name: {{ include "wandb.bucket.config" . }}
   labels:
     {{- include "wandb.commonLabels" . | nindent 4 }}
+  {{- if .Values.settingsMigrationHook.install }}
+  {{- with .Values.global.settingsMigrationHook.helmHookAnnotations }}
+  annotations:
+    {{- toYaml . | nindent 4 }}
+  {{- end }}
+  {{- end }}
 data:
 {{- with include "wandb.bucket" . | fromYaml }}
   BUCKET_NAME: "{{ .name }}"
@@ -19,6 +25,12 @@ metadata:
   name: {{ include "wandb.bucket.secret" . }}
   labels:
     {{- include "wandb.commonLabels" . | nindent 4 }}
+  {{- if .Values.settingsMigrationHook.install }}
+  {{- with .Values.global.settingsMigrationHook.helmHookAnnotations }}
+  annotations:
+    {{- toYaml . | nindent 4 }}
+  {{- end }}
+  {{- end }}
 data:
 {{- with include "wandb.bucket" . | fromYaml }}
   {{- if .accessKey }}

diff --git a/charts/operator-wandb/templates/gorilla.yaml b/charts/operator-wandb/templates/gorilla.yaml
@@ -10,10 +10,10 @@ stringData:
   {{- else }}
   GORILLA_EMAIL_SINK: "https://api.wandb.ai/email/dispatch"
   {{- end }}
-  SLACK_SECRET: {{ default "" .Values.global.slack.secret | b64enc }}
+  SLACK_SECRET: {{ default "" .Values.global.slack.secret }}
   {{- if and (not .Values.global.licenseSecret.name) (not .Values.global.licenseSecret.key) .Values.global.license }}
-  LICENSE: {{ .Values.global.license | b64enc }}
-  GORILLA_LICENSE: {{ .Values.global.license | b64enc }}
+  LICENSE: {{ .Values.global.license }}
+  GORILLA_LICENSE: {{ .Values.global.license }}
   {{- end }}
   {{- if ne .Values.global.auth.oidc.clientId ""  }}
   OIDC_CLIENT_SECRET: {{ .Values.global.auth.oidc.secret }}
@@ -86,7 +86,3 @@ data:
   GORILLA_TASK_QUEUE: "noop://"
   GORILLA_TASK_QUEUE_WORKER_ENABLED: "false"
   {{- end }}
-  {{- with include "wandb.bucket" . | fromYaml }}
-  GORILLA_FILE_STORE: "{{ .url }}"
-  GORILLA_STORAGE_BUCKET: "{{ .url }}"
-  {{- end }}
diff --git a/charts/operator-wandb/templates/session-key.yaml b/charts/operator-wandb/templates/session-key.yaml
@@ -0,0 +1,16 @@
+apiVersion: v1
+kind: Secret
+metadata:
+  name: {{ .Release.Name }}-gorilla-session-key
+  annotations:
+    "helm.sh/resource-policy": "keep"
+  labels:
+    {{- include "wandb.commonLabels" . | nindent 4 }}
+type: Opaque
+data:
+  # Retrieve the secret data using lookup function and when not exists, return an empty dictionary / map as result
+  {{- $secretObj := (lookup "v1" "Secret" .Release.Namespace (printf "%s-gorilla-session-key" .Release.Name)) | default dict }}
+  {{- $secretData := (get $secretObj "data") | default dict }}
+  # Set $gorillaSessionKey to existing secret data or generate a random one when not exists
+  {{- $gorillaSessionKey := (get $secretData "GORILLA_SESSION_KEY") | default (randAlphaNum 32 | b64enc) }}
+  GORILLA_SESSION_KEY: {{ $gorillaSessionKey | quote }}
diff --git a/charts/operator-wandb/templates/settings-migration-hook.yaml b/charts/operator-wandb/templates/settings-migration-hook.yaml
@@ -0,0 +1,20 @@
+{{- if .Values.settingsMigrationHook.install }}
+apiVersion: v1
+kind: Secret
+metadata:
+  name: {{ .Release.Name }}-smh-secret
+  labels:
+    {{- include "wandb.commonLabels" . | nindent 4 }}
+  {{- with .Values.global.settingsMigrationHook.helmHookAnnotations }}
+  annotations:
+    {{- toYaml . | nindent 4 }}
+  {{- end }}
+stringData:
+  SMH_DEBUG: {{ .Values.global.settingsMigrationHook.debug | quote }}
+  SMH_DRY_RUN: {{ .Values.global.settingsMigrationHook.dryRun | quote }}
+  {{- with include "wandb.bucket" . | fromYaml }}
+  AWS_REGION: "{{ .region }}"
+  AWS_S3_KMS_ID: "{{ .kmsKey }}"
+  {{- end }}
+{{- end }}
+