Merge pull request #2218 from nebius/SCHED-1015/3

Uburro · web-flow · commit 09d453588751 · 2026-02-25T14:01:30.000+01:00
SCHED-1015: add retries values for all Helm Releases (#2215)
diff --git a/helm/soperator-fluxcd/templates/backup-config.yaml b/helm/soperator-fluxcd/templates/backup-config.yaml
@@ -21,15 +21,13 @@ spec:
   dependsOn:
   - name: {{ include "soperator-fluxcd.fullname" . }}-ns
   install:
-    remediation:
-      retries: 3
+    {{- toYaml .Values.backup.config.install | nindent 4 }}
   interval: {{ default .Values.backup.interval .Values.backup.config.interval }}
   timeout: {{ default .Values.backup.timeout .Values.backup.config.timeout }}
   releaseName: {{ .Values.backup.config.releaseName }}
   targetNamespace: {{ .Values.slurmCluster.namespace }}
   upgrade:
-    remediation:
-      retries: 3
+    {{- toYaml .Values.backup.config.upgrade | nindent 4 }}
   values:
   {{- if .Values.backup.config.values }}
   {{- toYaml .Values.backup.config.values | nindent 4 }}
diff --git a/helm/soperator-fluxcd/templates/backup.yaml b/helm/soperator-fluxcd/templates/backup.yaml
@@ -27,11 +27,9 @@ spec:
   - name: {{ include "soperator-fluxcd.fullname" . }}-prometheus-operator-crds
   {{- end }}
   install:
-    crds: CreateReplace
-    remediation:
-      retries: 3
+    {{- toYaml .Values.backup.install | nindent 4 }}
   upgrade:
-    crds: CreateReplace
+    {{- toYaml .Values.backup.upgrade | nindent 4 }}
   interval: {{ .Values.backup.interval }}
   timeout: {{ .Values.backup.timeout }}
   releaseName: {{ .Values.backup.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/cert-manager.yaml b/helm/soperator-fluxcd/templates/cert-manager.yaml
@@ -24,14 +24,9 @@ spec:
   - name: {{ include "soperator-fluxcd.fullname" . }}-prometheus-operator-crds
   {{- end }}
   install:
-    crds: CreateReplace
-    remediation:
-      retries: 3
+    {{- toYaml .Values.certManager.install | nindent 4 }}
   upgrade:
-    crds: CreateReplace
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.certManager.upgrade | nindent 4 }}
   interval: {{ .Values.certManager.interval }}
   targetNamespace: {{ .Values.certManager.namespace }}
   releaseName: {{ .Values.certManager.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/csi-driver-nfs.yaml b/helm/soperator-fluxcd/templates/csi-driver-nfs.yaml
@@ -21,11 +21,9 @@ spec:
   dependsOn:
   - name: {{ include "soperator-fluxcd.fullname" . }}-ns
   install:
-    crds: Skip
-    remediation:
-      retries: 3
+    {{- toYaml .Values.csiDriverNfs.install | nindent 4 }}
   upgrade:
-    crds: Skip
+    {{- toYaml .Values.csiDriverNfs.upgrade | nindent 4 }}
   interval: {{ .Values.csiDriverNfs.interval }}
   timeout: {{ .Values.csiDriverNfs.timeout }}
   releaseName: {{ .Values.csiDriverNfs.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/dcgm-exporter.yaml b/helm/soperator-fluxcd/templates/dcgm-exporter.yaml
@@ -19,18 +19,13 @@ spec:
         name: {{ include "soperator-fluxcd.fullname" . }}-soperator
       version: {{ .Values.observability.dcgmExporter.version }}
   install:
-    crds: CreateReplace
-    remediation:
-      retries: 3
+    {{- toYaml .Values.observability.dcgmExporter.install | nindent 4 }}
   interval: {{ .Values.observability.dcgmExporter.interval }}
   timeout: {{ .Values.observability.dcgmExporter.timeout }}
   releaseName: {{ .Values.observability.dcgmExporter.releaseName }}
   targetNamespace: {{ .Values.observability.dcgmExporter.namespace }}
   upgrade:
-    crds: CreateReplace
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.observability.dcgmExporter.upgrade | nindent 4 }}
   values:
     dcgmHpcJobMappingDir: {{ .Values.observability.dcgmExporter.values.hpcJobMapDir }}
     validateToolkit: {{ .Values.observability.dcgmExporter.values.validateToolkit }}
diff --git a/helm/soperator-fluxcd/templates/kruise.yaml b/helm/soperator-fluxcd/templates/kruise.yaml
@@ -24,14 +24,9 @@ spec:
   - name: {{ include "soperator-fluxcd.fullname" . }}-cert-manager
   {{- end }}
   install:
-    crds: CreateReplace
-    remediation:
-      retries: 3
+    {{- toYaml .Values.soperator.kruise.install | nindent 4 }}
   upgrade:
-    crds: CreateReplace
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.soperator.kruise.upgrade | nindent 4 }}
   interval: {{ .Values.soperator.kruise.interval }}
   timeout: {{ .Values.soperator.kruise.timeout }}
   releaseName: {{ .Values.soperator.kruise.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/mariadb-operator-crds.yaml b/helm/soperator-fluxcd/templates/mariadb-operator-crds.yaml
@@ -19,11 +19,9 @@ spec:
         name: {{ include "soperator-fluxcd.fullname" . }}-mariadb-operator
       version: {{ .Values.mariadbOperator.version }}
   install:
-    crds: CreateReplace
-    remediation:
-      retries: 3
+    {{- toYaml .Values.mariadbOperator.crds.install | nindent 4 }}
   interval: {{ .Values.mariadbOperator.interval }}
   timeout: {{ .Values.mariadbOperator.timeout }}
   upgrade:
-    crds: CreateReplace
+    {{- toYaml .Values.mariadbOperator.crds.upgrade | nindent 4 }}
 {{- end }}
diff --git a/helm/soperator-fluxcd/templates/mariadb-operator.yaml b/helm/soperator-fluxcd/templates/mariadb-operator.yaml
@@ -23,18 +23,13 @@ spec:
   - name: {{ include "soperator-fluxcd.fullname" . }}-mariadb-operator-crds
   - name: {{ include "soperator-fluxcd.fullname" . }}-cert-manager
   install:
-    crds: Skip
-    remediation:
-      retries: 3
+    {{- toYaml .Values.mariadbOperator.install | nindent 4 }}
   interval: {{ .Values.mariadbOperator.interval }}
   timeout: {{ .Values.mariadbOperator.timeout }}
   releaseName: {{ .Values.mariadbOperator.releaseName }}
   targetNamespace: {{ .Values.mariadbOperator.namespace }}
   upgrade:
-    crds: Skip
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.mariadbOperator.upgrade | nindent 4 }}
   values:
   {{- if .Values.mariadbOperator.overrideValues }}
     {{- toYaml .Values.mariadbOperator.overrideValues | nindent 4 }}
diff --git a/helm/soperator-fluxcd/templates/monitoring-dashboards.yaml b/helm/soperator-fluxcd/templates/monitoring-dashboards.yaml
@@ -21,13 +21,9 @@ spec:
   dependsOn:
   - name: {{ include "soperator-fluxcd.fullname" . }}-ns
   install:
-    crds: Skip
-    remediation:
-      retries: 3
+    {{- toYaml .Values.soperator.monitoringDashboards.install | nindent 4 }}
   upgrade:
-    crds: Skip
-    remediation:
-      retries: 3
+    {{- toYaml .Values.soperator.monitoringDashboards.upgrade | nindent 4 }}
   interval: {{ .Values.soperator.monitoringDashboards.interval }}
   timeout: {{ .Values.soperator.monitoringDashboards.timeout }}
   releaseName: {{ .Values.soperator.monitoringDashboards.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/nfs-server.yaml b/helm/soperator-fluxcd/templates/nfs-server.yaml
@@ -22,11 +22,9 @@ spec:
   - name: {{ include "soperator-fluxcd.fullname" . }}-ns
   - name: {{ include "soperator-fluxcd.fullname" . }}-csi-driver-nfs
   install:
-    crds: Skip
-    remediation:
-      retries: 3
+    {{- toYaml .Values.nfsServer.install | nindent 4 }}
   upgrade:
-    crds: Skip
+    {{- toYaml .Values.nfsServer.upgrade | nindent 4 }}
   interval: {{ .Values.nfsServer.interval }}
   timeout: {{ .Values.nfsServer.timeout }}
   releaseName: {{ .Values.nfsServer.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/nodeconfigurator.yaml b/helm/soperator-fluxcd/templates/nodeconfigurator.yaml
@@ -21,18 +21,13 @@ spec:
   dependsOn:
   - name: {{ include "soperator-fluxcd.fullname" . }}-soperator
   install:
-    crds: Skip
-    remediation:
-      retries: 3
+    {{- toYaml .Values.soperator.nodeConfigurator.install | nindent 4 }}
   interval: {{ .Values.soperator.nodeConfigurator.interval }}
   timeout: {{ .Values.soperator.nodeConfigurator.timeout }}
   releaseName: {{ .Values.soperator.nodeConfigurator.releaseName }}
   targetNamespace: {{ .Values.soperator.namespace }}
   upgrade:
-    crds: Skip
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.soperator.nodeConfigurator.upgrade | nindent 4 }}
   values:
   {{- if .Values.soperator.nodeConfigurator.overrideValues }}
     {{- toYaml .Values.soperator.nodeConfigurator.overrideValues | nindent 4 }}
diff --git a/helm/soperator-fluxcd/templates/nodesets.yaml b/helm/soperator-fluxcd/templates/nodesets.yaml
@@ -26,18 +26,13 @@ spec:
 {{- end }}
 
   install:
-    crds: Skip
-    remediation:
-      retries: 3
+    {{- toYaml .Values.nodesets.install | nindent 4 }}
   interval: {{ .Values.nodesets.interval }}
   timeout: {{ .Values.nodesets.timeout }}
   releaseName: {{ .Values.nodesets.releaseName }}
   targetNamespace: {{ .Values.nodesets.namespace }}
   upgrade:
-    crds: Skip
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.nodesets.upgrade | nindent 4 }}
   values:
   {{- if .Values.nodesets.overrideValues }}
     {{- toYaml .Values.nodesets.overrideValues | nindent 4 }}
diff --git a/helm/soperator-fluxcd/templates/notifier.yaml b/helm/soperator-fluxcd/templates/notifier.yaml
@@ -22,12 +22,9 @@ spec:
     - name: {{ include "soperator-fluxcd.fullname" . }}-ns
     - name: {{ include "soperator-fluxcd.fullname" . }}-vm-logs
   install:
-    remediation:
-      retries: 3
+    {{- toYaml .Values.notifier.install | nindent 4 }}
   upgrade:
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.notifier.upgrade | nindent 4 }}
   interval: {{ .Values.notifier.interval }}
   timeout: {{ .Values.notifier.timeout }}
   releaseName: {{ .Values.notifier.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/opentelemetry-collector-events.yaml b/helm/soperator-fluxcd/templates/opentelemetry-collector-events.yaml
@@ -23,12 +23,9 @@ spec:
   - name: {{ include "soperator-fluxcd.fullname" . }}-ns
   - name: {{ include "soperator-fluxcd.fullname" . }}-vm-logs
   install:
-    remediation:
-      retries: 3
+    {{- toYaml .Values.observability.opentelemetry.events.install | nindent 4 }}
   upgrade:
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.observability.opentelemetry.events.upgrade | nindent 4 }}
   interval: {{ .Values.observability.opentelemetry.events.interval }}
   timeout: {{ .Values.observability.opentelemetry.events.timeout }}
   releaseName: opentelemetry-collector-events
diff --git a/helm/soperator-fluxcd/templates/opentelemetry-collector-jail-logs.yaml b/helm/soperator-fluxcd/templates/opentelemetry-collector-jail-logs.yaml
@@ -25,12 +25,9 @@ spec:
   - name: {{ include "soperator-fluxcd.fullname" . }}-ns
   - name: {{ include "soperator-fluxcd.fullname" . }}-vm-logs
   install:
-    remediation:
-      retries: 3
+    {{- toYaml .Values.observability.opentelemetry.logs.install | nindent 4 }}
   upgrade:
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.observability.opentelemetry.logs.upgrade | nindent 4 }}
   interval: {{ .Values.observability.opentelemetry.logs.interval }}
   timeout: {{ .Values.observability.opentelemetry.logs.timeout }}
   releaseName: opentelemetry-collector-jail-logs
diff --git a/helm/soperator-fluxcd/templates/opentelemetry-collector-logs.yaml b/helm/soperator-fluxcd/templates/opentelemetry-collector-logs.yaml
@@ -23,12 +23,9 @@ spec:
   - name: {{ include "soperator-fluxcd.fullname" . }}-ns
   - name: {{ include "soperator-fluxcd.fullname" . }}-vm-logs
   install:
-    remediation:
-      retries: 3
+    {{- toYaml .Values.observability.opentelemetry.logs.install | nindent 4 }}
   upgrade:
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.observability.opentelemetry.logs.upgrade | nindent 4 }}
   interval: {{ .Values.observability.opentelemetry.logs.interval }}
   timeout: {{ .Values.observability.opentelemetry.logs.timeout }}
   releaseName: opentelemetry-collector-logs
diff --git a/helm/soperator-fluxcd/templates/security-profiles-operator.yaml b/helm/soperator-fluxcd/templates/security-profiles-operator.yaml
@@ -19,14 +19,9 @@ spec:
         name: {{ include "soperator-fluxcd.fullname" . }}-security-profiles-operator
       version: {{ .Values.securityProfilesOperator.version }}
   install:
-    crds: CreateReplace
-    remediation:
-      retries: 3
+    {{- toYaml .Values.securityProfilesOperator.install | nindent 4 }}
   upgrade:
-    crds: CreateReplace
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.securityProfilesOperator.upgrade | nindent 4 }}
   interval: {{ .Values.securityProfilesOperator.interval }}
   timeout: {{ .Values.securityProfilesOperator.timeout }}
   releaseName: {{ .Values.securityProfilesOperator.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/slurm-cluster-storage.yaml b/helm/soperator-fluxcd/templates/slurm-cluster-storage.yaml
@@ -25,12 +25,9 @@ spec:
   - name: {{ include "soperator-fluxcd.fullname" . }}-soperator
   {{- end }}
   install:
-    remediation:
-      retries: 3
+    {{- toYaml .Values.slurmCluster.slurmClusterStorage.install | nindent 4 }}
   upgrade:
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.slurmCluster.slurmClusterStorage.upgrade | nindent 4 }}
   interval: {{ .Values.slurmCluster.slurmClusterStorage.interval }}
   timeout: {{ .Values.slurmCluster.slurmClusterStorage.timeout }}
   releaseName: {{ .Values.slurmCluster.slurmClusterStorage.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/slurm-cluster.yaml b/helm/soperator-fluxcd/templates/slurm-cluster.yaml
@@ -31,18 +31,13 @@ spec:
 {{- end }}
 
   install:
-    crds: Skip
-    remediation:
-      retries: 3
+    {{- toYaml .Values.slurmCluster.install | nindent 4 }}
   interval: {{ .Values.slurmCluster.interval }}
   timeout: {{ .Values.slurmCluster.timeout }}
   releaseName: soperator
   targetNamespace: soperator
   upgrade:
-    crds: Skip
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.slurmCluster.upgrade | nindent 4 }}
   values:
   {{- if .Values.slurmCluster.overrideValues }}
     {{- toYaml .Values.slurmCluster.overrideValues | nindent 4 }}
diff --git a/helm/soperator-fluxcd/templates/soperator-activechecks.yaml b/helm/soperator-fluxcd/templates/soperator-activechecks.yaml
@@ -22,18 +22,13 @@ spec:
   - name: {{ include "soperator-fluxcd.fullname" . }}-slurm-cluster
   - name: {{ include "soperator-fluxcd.fullname" . }}-soperatorchecks  
   install:
-    crds: Skip
-    remediation:
-      retries: 3
+    {{- toYaml .Values.soperatorActiveChecks.install | nindent 4 }}
   interval: {{ .Values.soperatorActiveChecks.interval }}
   timeout: {{ .Values.soperatorActiveChecks.timeout }}
   releaseName: {{ .Values.soperatorActiveChecks.releaseName }}
   targetNamespace: {{ .Values.soperatorActiveChecks.namespace }}
   upgrade:
-    crds: Skip
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.soperatorActiveChecks.upgrade | nindent 4 }}
   values:
   {{- if .Values.soperatorActiveChecks.overrideValues }}
     {{- toYaml .Values.soperatorActiveChecks.overrideValues | nindent 4 }}
diff --git a/helm/soperator-fluxcd/templates/soperator-custom-configmaps.yaml b/helm/soperator-fluxcd/templates/soperator-custom-configmaps.yaml
@@ -21,13 +21,9 @@ spec:
   dependsOn:
   - name: {{ include "soperator-fluxcd.fullname" . }}-ns
   install:
-    crds: Skip
-    remediation:
-      retries: 3
+    {{- toYaml .Values.customConfigmaps.install | nindent 4 }}
   upgrade:
-    crds: Skip
-    remediation:
-      retries: 3
+    {{- toYaml .Values.customConfigmaps.upgrade | nindent 4 }}
   interval: {{ .Values.customConfigmaps.interval }}
   timeout: {{ .Values.customConfigmaps.timeout }}
   releaseName: {{ .Values.customConfigmaps.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/soperator.yaml b/helm/soperator-fluxcd/templates/soperator.yaml
@@ -25,14 +25,9 @@ spec:
   - name: {{ include "soperator-fluxcd.fullname" . }}-cert-manager
   {{- end }}
   install:
-    crds: CreateReplace
-    remediation:
-      retries: 3
+    {{- toYaml .Values.soperator.install | nindent 4 }}
   upgrade:
-    crds: CreateReplace
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.soperator.upgrade | nindent 4 }}
   interval: {{ .Values.soperator.interval }}
   timeout: {{ .Values.soperator.timeout }}
   releaseName: {{ .Values.soperator.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/soperatorchecks.yaml b/helm/soperator-fluxcd/templates/soperatorchecks.yaml
@@ -21,12 +21,9 @@ spec:
   dependsOn:
   - name: {{ include "soperator-fluxcd.fullname" . }}-soperator
   install:
-    remediation:
-      retries: 3
+    {{- toYaml .Values.soperator.soperatorChecks.install | nindent 4 }}
   upgrade:
-    remediation:
-      retries: 3
-      remediateLastFailure: true
+    {{- toYaml .Values.soperator.soperatorChecks.upgrade | nindent 4 }}
   interval: {{ .Values.soperator.soperatorChecks.interval }}
   timeout: {{ .Values.soperator.soperatorChecks.timeout }}
   releaseName: {{ .Values.soperator.soperatorChecks.releaseName }}
diff --git a/helm/soperator-fluxcd/templates/storageclasses.yaml b/helm/soperator-fluxcd/templates/storageclasses.yaml
diff --git a/helm/soperator-fluxcd/templates/tailscale.yaml b/helm/soperator-fluxcd/templates/tailscale.yaml
diff --git a/helm/soperator-fluxcd/templates/victoria-metrics-operator-crds.yaml b/helm/soperator-fluxcd/templates/victoria-metrics-operator-crds.yaml
diff --git a/helm/soperator-fluxcd/templates/vm-logs.yaml b/helm/soperator-fluxcd/templates/vm-logs.yaml
diff --git a/helm/soperator-fluxcd/templates/vm-stack.yaml b/helm/soperator-fluxcd/templates/vm-stack.yaml
diff --git a/helm/soperator-fluxcd/values.yaml b/helm/soperator-fluxcd/values.yaml
