feat: Add support for topologySpreadConstraints

charts/operator-wandb/Chart.lock

@@ -31,7 +31,7 @@ dependencies:
   version: 25.3.5
 - name: etcd
   repository: oci://registry-1.docker.io/bitnamicharts
-  version: 10.6.1
+  version: 10.7.1
 - name: bufstream
   repository: file://charts/bufstream
   version: 0.3.1
@@ -56,5 +56,5 @@ dependencies:
 - name: wandb-base
   repository: file://../wandb-base
   version: 0.1.1
-digest: sha256:1eb284a2d6588f3e3d90293abb99f0e25b02fd62a87758f023d809879a82e31d
-generated: "2024-12-06T11:08:12.69179-08:00"
+digest: sha256:3c02029e1921428d7f45866291a8c42b79894505ee47864c54f34fcd847b2793
+generated: "2024-12-18T18:29:20.054295+05:30"

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: 0.21.8
+version: 0.21.9
 appVersion: 1.0.0
 icon: https://wandb.ai/logo.svg
 

charts/operator-wandb/charts/app/templates/deployment.yaml

@@ -43,6 +43,9 @@ spec:
       {{- include "wandb.nodeSelector" . | nindent 6 }}
       {{- include "wandb.priorityClassName" . | nindent 6 }}
       {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 6 }}
+      {{- if .Values.topologySpreadConstraints.enabled }}
+        {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "app.labels" .))) | nindent 6 }}
+      {{- end }}
       # Extend the pods shutdown grace period from the default of 30s to 60s.
       # This goes in the pod template spec.
       terminationGracePeriodSeconds: 60

charts/operator-wandb/charts/app/values.yaml

@@ -51,6 +51,9 @@ container:
     readOnlyRootFilesystem: false
     privileged: false
 
+topologySpreadConstraints:
+  enabled: false
+
 common:
   labels: {}
 

charts/operator-wandb/charts/console/templates/deployment.yaml

@@ -42,6 +42,9 @@ spec:
       {{- include "wandb.nodeSelector" . | nindent 6 }}
       {{- include "wandb.priorityClassName" . | nindent 6 }}
       {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 6 }}
+      {{- if .Values.topologySpreadConstraints.enabled }}
+        {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "console.labels" .))) | nindent 6 }}
+      {{- end }}
       # Extend the pods shutdown grace period from the default of 30s to 60s.
       # This goes in the pod template spec.
       terminationGracePeriodSeconds: 60

charts/operator-wandb/charts/console/values.yaml

@@ -49,6 +49,9 @@ container:
     readOnlyRootFilesystem: false
     privileged: false
 
+topologySpreadConstraints:
+  enabled: false
+
 service:
   type: ClusterIP
   annotations: {}

charts/operator-wandb/charts/executor/templates/deployment.yaml

@@ -45,6 +45,9 @@ spec:
       {{- include "wandb.nodeSelector" . | nindent 6 }}
       {{- include "wandb.priorityClassName" . | nindent 6 }}
       {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 6 }}
+      {{- if .Values.topologySpreadConstraints.enabled }}
+        {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "executor.labels" .))) | nindent 6 }}
+      {{- end }}
       # needed to ensure ensure Kafka consumers handle pod termination gracefully and avoid data loss
       terminationGracePeriodSeconds: 60
       containers:

charts/operator-wandb/charts/executor/values.yaml

@@ -54,6 +54,9 @@ container:
     readOnlyRootFilesystem: false
     privileged: false
 
+topologySpreadConstraints:
+  enabled: false
+
 common:
   labels: {}
 

charts/operator-wandb/charts/filestream/templates/deployment.yaml

@@ -45,6 +45,9 @@ spec:
       {{- include "wandb.nodeSelector" . | nindent 6 }}
       {{- include "wandb.priorityClassName" . | nindent 6 }}
       {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 6 }}
+      {{- if .Values.topologySpreadConstraints.enabled }}
+        {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "filestream.labels" .))) | nindent 6 }}
+      {{- end }}
       # needed to ensure ensure Kafka consumers handle pod termination gracefully and avoid data loss
       terminationGracePeriodSeconds: 60
       containers:

charts/operator-wandb/charts/filestream/values.yaml

@@ -58,6 +58,9 @@ container:
     readOnlyRootFilesystem: false
     privileged: false
 
+topologySpreadConstraints:
+  enabled: false
+
 common:
   labels: {}
 

charts/operator-wandb/charts/flat-run-fields-updater/templates/deployment.yaml

@@ -45,6 +45,9 @@ spec:
       {{- include "wandb.nodeSelector" . | nindent 6 }}
       {{- include "wandb.priorityClassName" . | nindent 6 }}
       {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 6 }}
+      {{- if .Values.topologySpreadConstraints.enabled }}
+        {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "flat-run-fields-updater.labels" .))) | nindent 6 }}
+      {{- end }}
       # needed to ensure ensure Kafka consumers handle pod termination gracefully and avoid data loss
       terminationGracePeriodSeconds: 60
       containers:

charts/operator-wandb/charts/flat-run-fields-updater/values.yaml

@@ -54,6 +54,9 @@ container:
     readOnlyRootFilesystem: false
     privileged: false
 
+topologySpreadConstraints:
+  enabled: false
+
 common:
   labels: {}
 

charts/operator-wandb/charts/mysql/templates/statefulset.yaml

@@ -29,10 +29,13 @@ spec:
       {{- include "wandb.nodeSelector" . | nindent 6 }}
       {{- include "wandb.priorityClassName" . | nindent 6 }}
       {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 6 }}
+      {{- if .Values.topologySpreadConstraints.enabled }}
+        {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "mysql.labels" .))) | nindent 6 }}
+      {{- end }}
       containers:
         - name: mysql
           image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"
-          {{- include "wandb.podSecurityContext" .Values.container.securityContext | nindent 10 }}
+          {{- include "wandb.containerSecurityContext" .Values.container.securityContext | nindent 10 }}
           ports:
           - name: mysql
             containerPort: 3306

charts/operator-wandb/charts/mysql/values.yaml

@@ -45,6 +45,9 @@ container:
     readOnlyRootFilesystem: false
     privileged: false
 
+topologySpreadConstraints:
+  enabled: false
+
 resources:
   requests:
     cpu: 500m

charts/operator-wandb/charts/otel/charts/agent/templates/deployment.yaml

@@ -43,6 +43,9 @@ spec:
       {{- include "wandb.nodeSelector" . | nindent 6 }}
       {{- include "wandb.priorityClassName" . | nindent 6 }}
       {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 6 }}
+      {{- if .Values.topologySpreadConstraints.enabled }}
+        {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "otelAgent.labels" .))) | nindent 6 }}
+      {{- end }}
       containers:
         - name: {{ .Chart.Name }}
           image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"

charts/operator-wandb/charts/otel/charts/agent/values.yaml

@@ -51,6 +51,9 @@ container:
     readOnlyRootFilesystem: false
     privileged: false
 
+topologySpreadConstraints:
+  enabled: false
+
 clusterRole:
   annotations: {}
   labels: {}

charts/operator-wandb/charts/otel/values.yaml

@@ -29,6 +29,8 @@ daemonset:
 deployment:
   install: true
   mode: "deployment"
+  topologySpreadConstraints:
+    enabled: false
   presets:
     receivers:
       statsd: true

charts/operator-wandb/charts/parquet/templates/cron.yaml

@@ -42,6 +42,9 @@ spec:
           {{- include "wandb.nodeSelector" . | nindent 10 }}
           {{- include "wandb.priorityClassName" . | nindent 10 }}
           {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 10 }}
+          {{- if .Values.topologySpreadConstraints.enabled }}
+            {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "parquet.labels" .))) | nindent 10 }}
+          {{- end }}
           containers:
             - name: backfill-job
               image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"

charts/operator-wandb/charts/parquet/templates/deployment.yaml

@@ -43,6 +43,9 @@ spec:
       {{- include "wandb.nodeSelector" . | nindent 6 }}
       {{- include "wandb.priorityClassName" . | nindent 6 }}
       {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 6 }}
+      {{- if .Values.topologySpreadConstraints.enabled }}
+        {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "parquet.labels" .))) | nindent 6 }}
+      {{- end }}
       containers:
         - name: {{ .Chart.Name }}
           image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"

charts/operator-wandb/charts/parquet/values.yaml

@@ -55,6 +55,9 @@ container:
     allowPrivilegeEscalation: false
     readOnlyRootFilesystem: false
 
+topologySpreadConstraints:
+  enabled: false
+
 resources:
   # We usually recommend not to specify default resources and to leave this as a
   # conscious choice for the user. This also increases chances charts run on

charts/operator-wandb/charts/weave-trace/templates/deployment.yaml

@@ -40,6 +40,9 @@ spec:
       {{- include "wandb.nodeSelector" . | nindent 6 }}
       {{- include "wandb.priorityClassName" . | nindent 6 }}
       {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 6 }}
+      {{- if .Values.topologySpreadConstraints.enabled }}
+        {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "weaveTrace.labels" .))) | nindent 6 }}
+      {{- end }}
       {{- if not .Values.migration.useHook }}
       initContainers:
         - name: {{ include "weaveTraceMigrate.fullname" . }}

charts/operator-wandb/charts/weave-trace/values.yaml

@@ -46,6 +46,9 @@ container:
     readOnlyRootFilesystem: false
     privileged: false
 
+topologySpreadConstraints:
+  enabled: false
+
 resources:
   requests:
     cpu: 500m

charts/operator-wandb/charts/weave/templates/deployment.yaml

@@ -43,6 +43,9 @@ spec:
       {{- include "wandb.nodeSelector" . | nindent 6 }}
       {{- include "wandb.priorityClassName" . | nindent 6 }}
       {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 6 }}
+      {{- if .Values.topologySpreadConstraints.enabled }}
+        {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "weave.labels" .))) | nindent 6 }}
+      {{- end }}
       containers:
         - name: {{ include "weave.fullname" .  }}
           image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"

charts/operator-wandb/charts/weave/values.yaml

@@ -55,6 +55,9 @@ container:
     readOnlyRootFilesystem: false
     privileged: false
 
+topologySpreadConstraints:
+  enabled: false
+
 resources:
   # We usually recommend not to specify default resources and to leave this as a
   # conscious choice for the user. This also increases chances charts run on

charts/operator-wandb/charts/yace/templates/deployment.yaml

@@ -42,6 +42,9 @@ spec:
       {{- include "wandb.nodeSelector" . | nindent 6 }}
       {{- include "wandb.priorityClassName" . | nindent 6 }}
       {{- include "wandb.podSecurityContext" .Values.pod.securityContext | nindent 6 }}
+      {{- if .Values.topologySpreadConstraints.enabled }}
+        {{- include "wandb.topologySpreadConstraints" (dict "matchLabels" (fromYaml (include "yace.labels" .))) | nindent 6 }}
+      {{- end }}
       containers:
         - name: {{ .Chart.Name }}
           image: "{{ .Values.image.registry }}/{{ .Values.image.repository }}:{{ .Values.image.tag }}"

charts/operator-wandb/charts/yace/values.yaml

@@ -113,6 +113,9 @@ container:
     readOnlyRootFilesystem: false
     privileged: false
 
+topologySpreadConstraints:
+  enabled: false
+
 deployment: {}
 
 serviceAccount:

charts/operator-wandb/templates/_pods.tpl

@@ -88,3 +88,27 @@ securityContext:
 {{- end }}
 {{- end }}
 
+{{/*
+Return a simplified topologySpreadConstraints definition with dynamic labelSelector and matchLabels passed as a map.
+*/}}
+{{- define "wandb.topologySpreadConstraints" -}}
+{{- $matchLabels := .matchLabels -}}
+topologySpreadConstraints:
+  - maxSkew: 1
+    topologyKey: kubernetes.io/hostname
+    whenUnsatisfiable: ScheduleAnyway
+    labelSelector:
+      matchLabels:
+        {{- range $key, $value := $matchLabels }}
+        {{ $key }}: {{ $value | quote }}
+        {{- end }}
+  - maxSkew: 1
+    topologyKey: topology.kubernetes.io/zone
+    whenUnsatisfiable: ScheduleAnyway
+    labelSelector:
+      matchLabels:
+        {{- range $key, $value := $matchLabels }}
+        {{ $key }}: {{ $value | quote }}
+        {{- end }}
+{{- end }}
+

charts/operator-wandb/values.yaml

@@ -208,6 +208,8 @@ parquet:
     tag: latest
   envFrom:
     "{{ .Release.Name }}-bucket-configmap": "configMapRef"
+  topologySpreadConstraints:
+    enabled: false
 
 app:
   install: true
@@ -216,6 +218,8 @@ app:
     tag: latest
   envFrom:
     "{{ .Release.Name }}-bucket-configmap": "configMapRef"
+  topologySpreadConstraints:
+    enabled: false
 
 nginx:
   install: false
@@ -225,6 +229,8 @@ weave:
   image:
     repository: wandb/local
     tag: latest
+  topologySpreadConstraints:
+    enabled: false
 
 weave-trace:
   install: false
@@ -233,12 +239,16 @@ weave-trace:
     tag: latest
   datadog:
     enabled: false
+  topologySpreadConstraints:
+    enabled: false
 
 console:
   install: true
   image:
     repository: wandb/console
     tag: latest
+  topologySpreadConstraints:
+    enabled: false
 
 flat-run-fields-updater:
   install: false
@@ -249,6 +259,8 @@ flat-run-fields-updater:
     subscription: ""
   envFrom:
     "{{ .Release.Name }}-bucket-configmap": "configMapRef"
+  topologySpreadConstraints:
+    enabled: false
 
 filestream:
   install: false
@@ -259,6 +271,8 @@ filestream:
     subscription: ""
   envFrom:
     "{{ .Release.Name }}-bucket-configmap": "configMapRef"
+  topologySpreadConstraints:
+    enabled: false
 
 executor:
   install: false
@@ -267,12 +281,16 @@ executor:
     tag: latest
   envFrom:
     "{{ .Release.Name }}-bucket-configmap": "configMapRef"
+  topologySpreadConstraints:
+    enabled: false
 
 mysql:
   install: false
   persistence:
     size: 20Gi
     storageClass: ""
+  topologySpreadConstraints:
+    enabled: false
 
 yace:
   install: false
@@ -283,6 +301,8 @@ yace:
       prometheus.io/port: "5000"
       prometheus.io/path: "/metrics"
       prometheus.io/scheme: http
+  topologySpreadConstraints:
+    enabled: false
 
 redis:
   install: true
@@ -334,6 +354,8 @@ otel:
   #   promethus endpoint which causes duplicated metrics that get repoted.
   deployment:
     install: false
+    topologySpreadConstraints:
+      enabled: false
 
 #   # cannot use install because schema validation will fail.
 #   enabled: true