Merge pull request #1166 from nebius/dev

Soperator release 1.21.3

Author: rdjjke

Makefile

@@ -284,7 +284,8 @@ build: manifests generate fmt vet ## Build manager binary with native toolchain.
 
 .PHONY: run
 run: manifests generate fmt vet ## Run a controller from your host with native toolchain.
-	go run ./cmd/main.go
+	IS_PROMETHEUS_CRD_INSTALLED=true IS_MARIADB_CRD_INSTALLED=true ENABLE_WEBHOOKS=false IS_APPARMOR_CRD_INSTALLED=true go run cmd/main.go \
+	 -log-level=debug -leader-elect=false -operator-namespace=soperator-system --enable-topology-controller=true
 
 .PHONY: docker-build-go-base
 docker-build-go-base: ## Build shared Go base image

VERSION

@@ -1 +1 @@
-1.21.2
+1.21.3

api/v1/slurmcluster_types.go

@@ -188,10 +188,10 @@ type SlurmConfig struct {
 	// It is set automatically to `topology/tree` if SlurmTopologyConfigMapRefName is specified.
 	//
 	// +kubebuilder:validation:Optional
+	// +kubebuilder:default="topology/tree"
 	TopologyPlugin string `json:"topologyPlugin,omitempty"`
 	// TopologyParam is list of comma-separated options identifying network topology options.
 	//
-	// +kubebuilder:default=topology/tree
 	TopologyParam string `json:"topologyParam,omitempty"`
 }
 
@@ -649,6 +649,21 @@ type SlurmRest struct {
 	// +kubebuilder:default=false
 	Enabled bool `json:"enabled,omitempty"`
 
+	// ThreadCount defines the number of threads for slurmrestd
+	//
+	// +kubebuilder:validation:Optional
+	// +kubebuilder:validation:Minimum=2
+	// +kubebuilder:validation:Maximum=1024
+	// +kubebuilder:default=3
+	ThreadCount *int32 `json:"threadCount,omitempty"`
+
+	// MaxConnections defines the maximum number of connections for slurmrestd
+	//
+	// +kubebuilder:validation:Optional
+	// +kubebuilder:validation:Minimum=2
+	// +kubebuilder:default=10
+	MaxConnections *int32 `json:"maxConnections,omitempty"`
+
 	// SlurmRestNode represents the Slurm REST API daemon configuration
 	//
 	// +kubebuilder:validation:Optional

api/v1/zz_generated.deepcopy.go

@@ -26,7 +26,7 @@ type ActiveCheckSpec struct {
 	// Suspend indicates whether the action is suspended.
 	// +kubebuilder:validation:Optional
 	// +kubebuilder:default=true
-	Suspend bool `json:"suspend,omitempty"`
+	Suspend *bool `json:"suspend,omitempty"`
 
 	// ActiveDeadlineSeconds defines the CronJob timeout in seconds
 	//
@@ -48,7 +48,7 @@ type ActiveCheckSpec struct {
 
 	// RunAfterCreation specifies whether the job should run immediately after the CronJob is created.
 	// +kubebuilder:default=true
-	RunAfterCreation bool `json:"runAfterCreation,omitempty"`
+	RunAfterCreation *bool `json:"runAfterCreation,omitempty"`
 
 	// NodeSelector defines the desired selector for the K8s nodes to place Slurm workers on
 	//

api/v1alpha1/activecheck_types.go

@@ -3244,11 +3244,11 @@ spec:
                     description: Additional parameters for the task plugin
                     type: string
                   topologyParam:
-                    default: topology/tree
                     description: TopologyParam is list of comma-separated options
                       identifying network topology options.
                     type: string
                   topologyPlugin:
+                    default: topology/tree
                     description: |-
                       TopologyPlugin identifies the plugin to determine network topology for optimizations.
                       It is set automatically to `topology/tree` if SlurmTopologyConfigMapRefName is specified.
@@ -13515,6 +13515,13 @@ spec:
                           K8sNodeFilterName defines the Kubernetes node filter name associated with the Slurm node.
                           Must correspond to the name of one of [K8sNodeFilter]
                         type: string
+                      maxConnections:
+                        default: 10
+                        description: MaxConnections defines the maximum number of
+                          connections for slurmrestd
+                        format: int32
+                        minimum: 2
+                        type: integer
                       rest:
                         description: SlurmRestNode represents the Slurm REST API daemon
                           configuration
@@ -13576,6 +13583,14 @@ spec:
                         description: Size defines the number of node instances
                         format: int32
                         type: integer
+                      threadCount:
+                        default: 3
+                        description: ThreadCount defines the number of threads for
+                          slurmrestd
+                        format: int32
+                        maximum: 1024
+                        minimum: 2
+                        type: integer
                     required:
                     - k8sNodeFilterName
                     type: object

api/v1alpha1/zz_generated.deepcopy.go

@@ -3,4 +3,4 @@ resources:
 images:
   - name: controller
     newName: cr.eu-north1.nebius.cloud/soperator/slurm-operator
-    newTag: 1.21.2
+    newTag: 1.21.3

config/crd/bases/slurm.nebius.ai_slurmclusters.yaml

@@ -87,7 +87,7 @@ spec:
             value: "false"
           - name: SLURM_OPERATOR_WATCH_NAMESPACES
             value: "*"
-        image: controller:1.21.2
+        image: controller:1.21.3
         imagePullPolicy: Always
         name: manager
         securityContext:

config/manager/kustomization.yaml

@@ -84,6 +84,17 @@ rules:
   verbs:
   - get
   - update
+- apiGroups:
+  - coordination.k8s.io
+  resources:
+  - leases
+  verbs:
+  - create
+  - get
+  - list
+  - patch
+  - update
+  - watch
 - apiGroups:
   - k8s.mariadb.com
   resources: