nebius · ali-sattari · Jun 22, 2026 · May 29, 2026 · May 29, 2026 · May 22, 2026
diff --git a/VERSION b/VERSION
@@ -1 +1 @@
-4.0.2
+4.1.0
diff --git a/ansible/roles/nc-health-checker/defaults/main.yml b/ansible/roles/nc-health-checker/defaults/main.yml
@@ -1,3 +1,3 @@
 ---
 
-nc_health_checker_version: 1.0.0-178.260115
+nc_health_checker_version: 1.1.8-a3dae386
diff --git a/config/manager/kustomization.yaml b/config/manager/kustomization.yaml
@@ -3,4 +3,4 @@ resources:
 images:
   - name: controller
     newName: cr.eu-north1.nebius.cloud/soperator/slurm-operator
-    newTag: 4.0.2
+    newTag: 4.1.0
diff --git a/config/manager/manager.yaml b/config/manager/manager.yaml
@@ -84,7 +84,7 @@ spec:
             value: "false"
           - name: SLURM_OPERATOR_WATCH_NAMESPACES
             value: "*"
-        image: controller:4.0.2
+        image: controller:4.1.0
         imagePullPolicy: Always
         name: manager
         securityContext:

diff --git a/config/soperatorchecks/kustomization.yaml b/config/soperatorchecks/kustomization.yaml
@@ -4,7 +4,7 @@ resources:
 images:
   - name: controller
     newName: cr.eu-north1.nebius.cloud/soperator/soperatorchecks
-    newTag: 4.0.2
+    newTag: 4.1.0
 patches:
   # Protect the /metrics endpoint by putting it behind auth.
   # If you want your controller-manager to expose the /metrics

diff --git a/docs/metrics-pipeline.md b/docs/metrics-pipeline.md
@@ -162,7 +162,7 @@ curl "http://localhost:8429/api/v1/query?query=up"
 
 #### Remote Write to Nebius Cloud
 - Endpoint: `https://write.monitoring.{region}.nebius.cloud/projects/{projectId}/buckets/soperator/prometheus`
-- Authentication: Bearer token from `/mnt/cloud-metadata/tsa-token`
+- Authentication: Bearer token from `/mnt/imds/tsa-token`, populated and refreshed by the vmagent IMDS token sidecar
 - When: Enabled with `publicEndpointEnabled: true`
 
 ### Visualization

diff --git a/fluxcd/environment/local/helmrelease.yaml b/fluxcd/environment/local/helmrelease.yaml
@@ -8,7 +8,7 @@ spec:
   chart:
     spec:
       chart: helm-soperator-fluxcd
-      version: "4.0.2"
+      version: "4.1.0"
       sourceRef:
         kind: HelmRepository
         name: soperator-fluxcd

diff --git a/fluxcd/environment/local/slurmCluster.yml b/fluxcd/environment/local/slurmCluster.yml
@@ -2,7 +2,7 @@ slurmCluster:
   enabled: true
   interval: 5m
   timeout: 5m
-  version: 4.0.2
+  version: 4.1.0
   namespace: soperator
   releaseName: soperator
   overrideValues:

diff --git a/fluxcd/environment/local/values.yaml b/fluxcd/environment/local/values.yaml
@@ -10,7 +10,7 @@ ns:
 soperator:
   enabled: true
   namespace: soperator-system
-  version: "4.0.2"
+  version: "4.1.0"
   interval: 5m
   timeout: 5m
   releaseName: soperator-controller
@@ -57,7 +57,7 @@ slurmCluster:
   enabled: true
   interval: 5m
   timeout: 5m
-  version: 4.0.2
+  version: 4.1.0
   namespace: soperator
   releaseName: soperator
   overrideValues:
@@ -276,10 +276,10 @@ backup:
 # NFS Server - disabled for local dev
 nfsServer:
   enabled: false
-  version: 4.0.2
+  version: 4.1.0
   overrideValues:
     image:
-      tag: 4.0.2
+      tag: 4.1.0
 # Notifier - disabled for local dev
 notifier:
   enabled: false
diff --git a/fluxcd/environment/nebius-cloud/dev/bootstrap/flux-kustomization.yaml b/fluxcd/environment/nebius-cloud/dev/bootstrap/flux-kustomization.yaml
@@ -11,7 +11,7 @@ spec:
     name: nebius-cloud
   postBuild:
     substitute:
-      soperator_version: 4.0.2
+      soperator_version: 4.1.0
   path: "./fluxcd/enviroment/nebius-cloud/dev"
   prune: true
   timeout: 1m
diff --git a/fluxcd/environment/nebius-cloud/prod/bootstrap/flux-kustomization.yaml b/fluxcd/environment/nebius-cloud/prod/bootstrap/flux-kustomization.yaml
@@ -10,7 +10,7 @@ spec:
     name: nebius-cloud
   postBuild:
     substitute:
-      soperator_version: 4.0.2
+      soperator_version: 4.1.0
   path: "./fluxcd/enviroment/nebius-cloud/prod"
   prune: false
   timeout: 1m
diff --git a/helm/nodeconfigurator/Chart.yaml b/helm/nodeconfigurator/Chart.yaml
@@ -13,9 +13,9 @@ type: application
 # This is the chart version. This version number should be incremented each time you make changes
 # to the chart and its templates, including the app version.
 # Versions are expected to follow Semantic Versioning (https://semver.org/)
-version: 4.0.2
+version: 4.1.0
 # This is the version number of the application being deployed. This version number should be
 # incremented each time you make changes to the application. Versions are not expected to
 # follow Semantic Versioning. They should reflect the version the application is using.
 # It is recommended to use it with quotes.
-appVersion: "4.0.2"
+appVersion: "4.1.0"
diff --git a/helm/nodeconfigurator/values.yaml b/helm/nodeconfigurator/values.yaml
@@ -28,7 +28,7 @@ rebooter:
   env: []
   image:
     repository: "cr.eu-north1.nebius.cloud/soperator/rebooter"
-    tag: "4.0.2"
+    tag: "4.1.0"
     pullPolicy: IfNotPresent
   nodeSelector: {}
   resources: {}

diff --git a/helm/nodesets/Chart.yaml b/helm/nodesets/Chart.yaml
@@ -13,9 +13,9 @@ type: application
 # This is the chart version. This version number should be incremented each time you make changes
 # to the chart and its templates, including the app version.
 # Versions are expected to follow Semantic Versioning (https://semver.org/)
-version: 4.0.2
+version: 4.1.0
 # This is the version number of the application being deployed. This version number should be
 # incremented each time you make changes to the application. Versions are not expected to
 # follow Semantic Versioning. They should reflect the version the application is using.
 # It is recommended to use it with quotes.
-appVersion: "4.0.2"
+appVersion: "4.1.0"
diff --git a/helm/nodesets/values.yaml b/helm/nodesets/values.yaml
@@ -25,13 +25,13 @@ priorityClasses:
 images:
   munge:
     repository: "cr.eu-north1.nebius.cloud/soperator/munge"
-    tag: "4.0.2-slurm25.11.3"
+    tag: "4.1.0-slurm25.11.3"
   sssd:
     repository: "cr.eu-north1.nebius.cloud/soperator/sssd"
     tag: "2.9.4-1-63"
   slurmd:
     repository: "cr.eu-north1.nebius.cloud/soperator/worker_slurmd"
-    tag: "4.0.2-slurm25.11.3"
+    tag: "4.1.0-slurm25.11.3"
 # NodeSets configuration
 # Each NodeSet defines a group of worker nodes with specific characteristics
 nodesets:
@@ -98,7 +98,7 @@ nodesets:
       # Optional
       image:
         repository: "cr.eu-north1.nebius.cloud/soperator/worker_slurmd"
-        tag: "4.0.2-slurm25.11.3"
+        tag: "4.1.0-slurm25.11.3"
         pullPolicy: "IfNotPresent"
       customEnv:
         - name: "NVIDIA_DRIVER_CAPABILITIES"
@@ -188,7 +188,7 @@ nodesets:
       image:
         repository: "cr.eu-north1.nebius.cloud/soperator/munge"
         # Optional
-        tag: "4.0.2-slurm25.11.3"
+        tag: "4.1.0-slurm25.11.3"
         pullPolicy: "IfNotPresent"
       # Munge container resources
       resources:

diff --git a/helm/slurm-cluster-storage/Chart.yaml b/helm/slurm-cluster-storage/Chart.yaml
@@ -2,5 +2,5 @@ apiVersion: v2
 name: helm-slurm-cluster-storage
 description: A Helm chart for Kubernetes
 type: application
-version: "4.0.2"
-appVersion: "4.0.2"
+version: "4.1.0"
+appVersion: "4.1.0"
diff --git a/helm/slurm-cluster/Chart.yaml b/helm/slurm-cluster/Chart.yaml
@@ -2,6 +2,6 @@ apiVersion: v2
 name: helm-slurm-cluster
 description: A Helm chart for Kubernetes
 type: application
-version: "4.0.2"
-appVersion: "4.0.2"
+version: "4.1.0"
+appVersion: "4.1.0"
 kubeVersion: ">=1.29.0-0"
diff --git a/helm/slurm-cluster/slurm_scripts/alloc_gpus_busy.drain.sh.json b/helm/slurm-cluster/slurm_scripts/alloc_gpus_busy.drain.sh.json
@@ -1,7 +1,7 @@
 {
   "name": "alloc_gpus_busy",
   "command": "./alloc_gpus_busy.drain.sh",
-  "platforms": ["8xGPU", "1xGPU"],
+  "platforms": ["8xGPU", "1xGPU", "4xGPU"],
   "skip_for_cpu_jobs": true,
   "skip_for_partial_gpu_jobs": false,
   "skip_for_reservation_prefixes": [],

diff --git a/helm/slurm-cluster/slurm_scripts/alloc_gpus_busy.undrain.sh.json b/helm/slurm-cluster/slurm_scripts/alloc_gpus_busy.undrain.sh.json
@@ -1,7 +1,7 @@
 {
   "name": "alloc_gpus_busy",
   "command": "./alloc_gpus_busy.undrain.sh",
-  "platforms": ["8xGPU", "1xGPU"],
+  "platforms": ["8xGPU", "1xGPU", "4xGPU"],
   "skip_for_cpu_jobs": true,
   "skip_for_partial_gpu_jobs": false,
   "skip_for_reservation_prefixes": [],