Merge pull request #1429 from nebius/fix-observability-projects

itechdima · web-flow · commit 83d2e37f676c · 2025-08-15T11:11:48.000+02:00
fix observability projects
diff --git a/VERSION b/VERSION
@@ -1,2 +1,2 @@
-1.21.10
+1.21.11
 
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: 1.21.10
+    newTag: 1.21.11
diff --git a/config/manager/manager.yaml b/config/manager/manager.yaml
@@ -87,7 +87,7 @@ spec:
             value: "false"
           - name: SLURM_OPERATOR_WATCH_NAMESPACES
             value: "*"
-        image: controller:1.21.10
+        image: controller:1.21.11
         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: 1.21.10
+    newTag: 1.21.11
 patches:
   # Protect the /metrics endpoint by putting it behind auth.
   # If you want your controller-manager to expose the /metrics
diff --git a/fluxcd/base/soperator-fluxcd/resources.yaml b/fluxcd/base/soperator-fluxcd/resources.yaml
@@ -46,3 +46,7 @@ spec:
       name: backup-schedule
       valuesKey: values.yaml
       optional: true
+    - kind: ConfigMap
+      name: terraform-opentelemetry-collector
+      optional: true
+      valuesKey: values.yaml
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: 1.21.10
+      soperator_version: 1.21.11
   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: 1.21.10
+      soperator_version: 1.21.11
   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: 1.21.10
+version: 1.21.11
 # 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: "1.21.10"
+appVersion: "1.21.11"
diff --git a/helm/nodeconfigurator/values.yaml b/helm/nodeconfigurator/values.yaml
@@ -22,7 +22,7 @@ rebooter:
   env: []
   image:
     repository: "cr.eu-north1.nebius.cloud/soperator/rebooter"
-    tag: "1.21.10"
+    tag: "1.21.11"
     pullPolicy: IfNotPresent
   nodeSelector: {}
   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: "1.21.10"
-appVersion: "1.21.10"
+version: "1.21.11"
+appVersion: "1.21.11"
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: "1.21.10"
-appVersion: "1.21.10"
+version: "1.21.11"
+appVersion: "1.21.11"
 kubeVersion: ">=1.29.0-0"
diff --git a/helm/slurm-cluster/values.yaml b/helm/slurm-cluster/values.yaml
@@ -512,16 +512,16 @@ telemetry: {}
 #   otelCollectorPort: 8429
 
 images:
-  slurmctld: "cr.eu-north1.nebius.cloud/soperator/controller_slurmctld:1.21.10-jammy-slurm24.11.5"
-  slurmrestd: "cr.eu-north1.nebius.cloud/soperator/slurmrestd:1.21.10-jammy-slurm24.11.5"
-  slurmd: "cr.eu-north1.nebius.cloud/soperator/worker_slurmd:1.21.10-jammy-slurm24.11.5"
-  sshd: "cr.eu-north1.nebius.cloud/soperator/login_sshd:1.21.10-jammy-slurm24.11.5"
-  munge: "cr.eu-north1.nebius.cloud/soperator/munge:1.21.10-jammy-slurm24.11.5"
-  populateJail: "cr.eu-north1.nebius.cloud/soperator/populate_jail:1.21.10-jammy-slurm24.11.5"
-  ncclBenchmark: "cr.eu-north1.nebius.cloud/soperator/nccl_benchmark:1.21.10-jammy-slurm24.11.5"
-  slurmdbd: "cr.eu-north1.nebius.cloud/soperator/controller_slurmdbd:1.21.10-jammy-slurm24.11.5"
-  soperatorExporter: "cr.eu-north1.nebius.cloud/soperator/soperator-exporter:1.21.10-jammy-slurm24.11.5"
-  sConfigController: cr.eu-north1.nebius.cloud/soperator/sconfigcontroller:1.21.10
+  slurmctld: "cr.eu-north1.nebius.cloud/soperator/controller_slurmctld:1.21.11-jammy-slurm24.11.5"
+  slurmrestd: "cr.eu-north1.nebius.cloud/soperator/slurmrestd:1.21.11-jammy-slurm24.11.5"
+  slurmd: "cr.eu-north1.nebius.cloud/soperator/worker_slurmd:1.21.11-jammy-slurm24.11.5"
+  sshd: "cr.eu-north1.nebius.cloud/soperator/login_sshd:1.21.11-jammy-slurm24.11.5"
+  munge: "cr.eu-north1.nebius.cloud/soperator/munge:1.21.11-jammy-slurm24.11.5"
+  populateJail: "cr.eu-north1.nebius.cloud/soperator/populate_jail:1.21.11-jammy-slurm24.11.5"
+  ncclBenchmark: "cr.eu-north1.nebius.cloud/soperator/nccl_benchmark:1.21.11-jammy-slurm24.11.5"
+  slurmdbd: "cr.eu-north1.nebius.cloud/soperator/controller_slurmdbd:1.21.11-jammy-slurm24.11.5"
+  soperatorExporter: "cr.eu-north1.nebius.cloud/soperator/soperator-exporter:1.21.11-jammy-slurm24.11.5"
+  sConfigController: cr.eu-north1.nebius.cloud/soperator/sconfigcontroller:1.21.11
   mariaDB: docker-registry1.mariadb.com/library/mariadb:11.4.3
 # Configuration for slurm scripts.
 slurmScripts:
diff --git a/helm/soperator-activechecks/Chart.yaml b/helm/soperator-activechecks/Chart.yaml
@@ -2,5 +2,5 @@ apiVersion: v2
 name: helm-soperator-activechecks
 description: A Helm chart for Kubernetes
 type: application
-version: "1.21.10"
-appVersion: "1.21.10"
+version: "1.21.11"
+appVersion: "1.21.11"
diff --git a/helm/soperator-activechecks/values.yaml b/helm/soperator-activechecks/values.yaml
@@ -41,6 +41,6 @@ activeCheck:
     setCondition: false
     drainSlurmNode: false
 images:
-  slurmJob: "cr.eu-north1.nebius.cloud/soperator/slurm_check_job:1.21.10-jammy-slurm24.11.5"
+  slurmJob: "cr.eu-north1.nebius.cloud/soperator/slurm_check_job:1.21.11-jammy-slurm24.11.5"
   k8sJob: "cr.eu-north1.nebius.cloud/soperator/ubuntu:jammy"
-  munge: "cr.eu-north1.nebius.cloud/soperator/munge:1.21.10-jammy-slurm24.11.5"
+  munge: "cr.eu-north1.nebius.cloud/soperator/munge:1.21.11-jammy-slurm24.11.5"
diff --git a/helm/soperator-crds/Chart.yaml b/helm/soperator-crds/Chart.yaml
@@ -2,6 +2,6 @@ apiVersion: v2
 name: helm-soperator-crds
 description: A Helm chart for Kubernetes
 type: application
-version: 1.21.10
-appVersion: "1.21.10"
+version: 1.21.11
+appVersion: "1.21.11"
 kubeVersion: ">=1.29.0-0"
diff --git a/helm/soperator-dcgm-exporter/Chart.yaml b/helm/soperator-dcgm-exporter/Chart.yaml
@@ -2,5 +2,5 @@ apiVersion: v2
 name: helm-soperator-dcgm-exporter
 description: A Helm chart for Nvidia DCGM Exporter
 type: application
-version: 1.21.10
-appVersion: "1.21.10"
+version: 1.21.11
+appVersion: "1.21.11"
diff --git a/helm/soperator-fluxcd/Chart.yaml b/helm/soperator-fluxcd/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: 1.21.10
+version: 1.21.11
 # 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: "1.21.10"
+appVersion: "1.21.11"
diff --git a/helm/soperator-fluxcd/templates/opentelemetry-collector-events.yaml b/helm/soperator-fluxcd/templates/opentelemetry-collector-events.yaml
@@ -70,7 +70,7 @@ spec:
             initial_interval: 200ms
           timeout: 5s
           headers:
-            iam-container: project-e00h61cxzwnf6zksvdn77
+            iam-container: {{ .Values.observability.logsProjectId }}
           auth:
             authenticator: bearertokenauth
         {{- end }}
diff --git a/helm/soperator-fluxcd/templates/opentelemetry-collector-jail-logs.yaml b/helm/soperator-fluxcd/templates/opentelemetry-collector-jail-logs.yaml
@@ -189,7 +189,7 @@ spec:
             initial_interval: 200ms
           timeout: 5s
           headers:
-            iam-container: project-e00h61cxzwnf6zksvdn77
+            iam-container: {{ .Values.observability.logsProjectId }}
           auth:
             authenticator: bearertokenauth
         {{- end }}
diff --git a/helm/soperator-fluxcd/templates/opentelemetry-collector-logs.yaml b/helm/soperator-fluxcd/templates/opentelemetry-collector-logs.yaml
@@ -63,7 +63,7 @@ spec:
             initial_interval: 200ms
           timeout: 5s
           headers:
-            iam-container: project-e00h61cxzwnf6zksvdn77
+            iam-container: {{ .Values.observability.logsProjectId }}
           auth:
             authenticator: bearertokenauth
         {{- end }}
diff --git a/helm/soperator-fluxcd/templates/vm-stack.yaml b/helm/soperator-fluxcd/templates/vm-stack.yaml
@@ -279,7 +279,7 @@ spec:
           {{- if $hasPublicEndpoint }}
             label:
               cluster: "%{CLUSTER_NAME}"
-              container_id: {{ .Values.observability.projectId  }}
+              container_id: {{ .Values.observability.metricsProjectId  }}
           {{- if and $hasVmagentSpec $hasVmagentSpec.remoteWriteSettings }}
           {{- toYaml $hasVmagentSpec.remoteWriteSettings | nindent 12 }}
           {{- end }}
@@ -291,7 +291,7 @@ spec:
           {{- $hasRemoteWrite := and $hasVmagentSpec $hasVmagentSpec.remoteWrite }}
           {{- $remoteWriteNotEmpty := and $hasRemoteWrite (gt (len $hasVmagentSpec.remoteWrite) 0) }}
           {{- if $hasPublicEndpoint }}
-            - url: https://write.monitoring.{{ .Values.observability.region }}.nebius.cloud/projects/{{ .Values.observability.projectId }}/buckets/soperator/prometheus
+            - url: https://write.monitoring.{{ .Values.observability.region }}.nebius.cloud/projects/{{ .Values.observability.metricsProjectId }}/buckets/soperator/prometheus
               inlineUrlRelabelConfig:
                 - action: labeldrop
                   regex: feature_node_kubernetes_io_.*|kubernetes_io_.*|nvidia_com_gpu.*
diff --git a/helm/soperator-fluxcd/values.yaml b/helm/soperator-fluxcd/values.yaml
@@ -78,7 +78,8 @@ observability:
   enabled: true
   clusterName: soperator
   publicEndpointEnabled: true
-  projectId: default-project-id
+  logsProjectId: default-project-id
+  metricsProjectId: default-project-id
   region: eu-north1
   opentelemetry:
     enabled: true
@@ -183,7 +184,7 @@ observability:
     enabled: true
     interval: 5m
     timeout: 5m
-    version: 1.21.10
+    version: 1.21.11
     namespace: soperator
     releaseName: soperator-dcgm-exporter
     values:
@@ -202,7 +203,7 @@ slurmCluster:
   enabled: true
   interval: 5m
   timeout: 5m
-  version: 1.21.10
+  version: 1.21.11
   namespace: soperator
   releaseName: soperator
   values: null
@@ -218,7 +219,7 @@ soperator:
   enabled: true
   interval: 5m
   timeout: 5m
-  version: 1.21.10
+  version: 1.21.11
   namespace: "soperator-system"
   releaseName: soperator-controller
   values:
diff --git a/helm/soperator/Chart.yaml b/helm/soperator/Chart.yaml
@@ -2,8 +2,8 @@ apiVersion: v2
 name: helm-soperator
 description: A Helm chart for Kubernetes
 type: application
-version: 1.21.10
-appVersion: "1.21.10"
+version: 1.21.11
+appVersion: "1.21.11"
 kubeVersion: ">=1.29.0-0"
 dependencies:
   - name: kruise
diff --git a/helm/soperator/values.yaml b/helm/soperator/values.yaml
@@ -38,7 +38,7 @@ controllerManager:
       slurmOperatorWatchNamespaces: '*'
     image:
       repository: cr.eu-north1.nebius.cloud/soperator/slurm-operator
-      tag: 1.21.10
+      tag: 1.21.11
     imagePullPolicy: Always
     resources:
       limits:
diff --git a/helm/soperatorchecks/Chart.yaml b/helm/soperatorchecks/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: 1.21.10
+version: 1.21.11
 # 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: "1.21.10"
+appVersion: "1.21.11"
diff --git a/helm/soperatorchecks/values.yaml b/helm/soperatorchecks/values.yaml
@@ -32,7 +32,7 @@ checks:
       slurmOperatorWatchNamespaces: '*'
     image:
       repository: cr.eu-north1.nebius.cloud/soperator/soperatorchecks
-      tag: 1.21.10
+      tag: 1.21.11
     imagePullPolicy: Always
     resources:
       limits:
diff --git a/internal/consts/version.go b/internal/consts/version.go
@@ -2,5 +2,5 @@
 package consts
 
 const (
-	VersionCR = "1.21.10"
+	VersionCR = "1.21.11"
 )
