Merge pull request #457 from nebius/hotfix_repopulation

asteny · web-flow · commit 8d2265eebc2c · 2025-02-18T13:40:26.000+01:00
Hotfix populate jail job reconciliation
diff --git a/VERSION b/VERSION
@@ -1 +1 @@
-1.18.0
+1.18.1
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: 1.18.0
+    newTag: 1.18.1
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:1.18.0
+        image: controller:1.18.1
         imagePullPolicy: Always
         name: manager
         securityContext:
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: "1.18.0"
-appVersion: "1.18.0"
+version: "1.18.1"
+appVersion: "1.18.1"
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: "1.18.0"
-appVersion: "1.18.0"
+version: "1.18.1"
+appVersion: "1.18.1"
 kubeVersion: ">=1.29.0-0"
diff --git a/helm/slurm-cluster/values.yaml b/helm/slurm-cluster/values.yaml
@@ -408,14 +408,14 @@ telemetry: {}
 #   otelCollectorPort: 8429
 
 images:
-  slurmctld: "cr.eu-north1.nebius.cloud/soperator/controller_slurmctld:1.18.0-jammy-slurm24.05.5"
-  slurmrestd: "cr.eu-north1.nebius.cloud/soperator/slurmrestd:1.18.0-jammy-slurm24.05.5"
-  slurmd: "cr.eu-north1.nebius.cloud/soperator/worker_slurmd:1.18.0-jammy-slurm24.05.5"
-  sshd: "cr.eu-north1.nebius.cloud/soperator/login_sshd:1.18.0-jammy-slurm24.05.5"
-  munge: "cr.eu-north1.nebius.cloud/soperator/munge:1.18.0-jammy-slurm24.05.5"
-  populateJail: "cr.eu-north1.nebius.cloud/soperator/populate_jail:1.18.0-jammy-slurm24.05.5"
-  ncclBenchmark: "cr.eu-north1.nebius.cloud/soperator/nccl_benchmark:1.18.0-jammy-slurm24.05.5"
-  slurmdbd: "cr.eu-north1.nebius.cloud/soperator/controller_slurmdbd:1.18.0-jammy-slurm24.05.5"
-  exporter: "cr.eu-north1.nebius.cloud/soperator/exporter:1.18.0-jammy-slurm24.05.5"
+  slurmctld: "cr.eu-north1.nebius.cloud/soperator/controller_slurmctld:1.18.1-jammy-slurm24.05.5"
+  slurmrestd: "cr.eu-north1.nebius.cloud/soperator/slurmrestd:1.18.1-jammy-slurm24.05.5"
+  slurmd: "cr.eu-north1.nebius.cloud/soperator/worker_slurmd:1.18.1-jammy-slurm24.05.5"
+  sshd: "cr.eu-north1.nebius.cloud/soperator/login_sshd:1.18.1-jammy-slurm24.05.5"
+  munge: "cr.eu-north1.nebius.cloud/soperator/munge:1.18.1-jammy-slurm24.05.5"
+  populateJail: "cr.eu-north1.nebius.cloud/soperator/populate_jail:1.18.1-jammy-slurm24.05.5"
+  ncclBenchmark: "cr.eu-north1.nebius.cloud/soperator/nccl_benchmark:1.18.1-jammy-slurm24.05.5"
+  slurmdbd: "cr.eu-north1.nebius.cloud/soperator/controller_slurmdbd:1.18.1-jammy-slurm24.05.5"
+  exporter: "cr.eu-north1.nebius.cloud/soperator/exporter:1.18.1-jammy-slurm24.05.5"
   mariaDB: "docker-registry1.mariadb.com/library/mariadb:11.4.3"
   rebooter: "cr.eu-north1.nebius.cloud/soperator/rebooter:1.17.0"
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: 1.18.0
-appVersion: "1.18.0"
+version: 1.18.1
+appVersion: "1.18.1"
 kubeVersion: ">=1.29.0-0"
diff --git a/helm/soperator/Chart.yaml b/helm/soperator/Chart.yaml
@@ -2,6 +2,6 @@ apiVersion: v2
 name: helm-soperator
 description: A Helm chart for Kubernetes
 type: application
-version: 1.18.0
-appVersion: "1.18.0"
+version: 1.18.1
+appVersion: "1.18.1"
 kubeVersion: ">=1.29.0-0"
diff --git a/helm/soperator/values.yaml b/helm/soperator/values.yaml
@@ -37,7 +37,7 @@ controllerManager:
       slurmOperatorWatchNamespaces: '*'
     image:
       repository: cr.eu-north1.nebius.cloud/soperator/slurm-operator
-      tag: 1.18.0
+      tag: 1.18.1
     imagePullPolicy: Always
     resources:
       limits:
diff --git a/internal/consts/version.go b/internal/consts/version.go
@@ -2,5 +2,5 @@
 package consts
 
 const (
-	VersionCR = "1.18.0"
+	VersionCR = "1.18.1"
 )
diff --git a/internal/controller/clustercontroller/populate_job.go b/internal/controller/clustercontroller/populate_job.go
@@ -73,7 +73,7 @@ func (r SlurmClusterReconciler) ReconcilePopulateJail(
 						}
 					}
 
-					if !apierrors.IsNotFound(getErr) && !isMaintenanceStopMode {
+					if getErr != nil && !apierrors.IsNotFound(getErr) && !isMaintenanceStopMode {
 						stepLogger.Error(getErr, "Failed to get")
 						return errors.Wrap(getErr, "getting Populate jail Job")
 					}

Original file line number	Diff line number	Diff line change
`@@ -2,5 +2,5 @@`
`2`	`2`	`package consts`
`3`	`3`
`4`	`4`	`const (`
`5`		`- VersionCR = "1.18.0"`
	`5`	`+ VersionCR = "1.18.1"`
`6`	`6`	`)`