MSP-3909: Try to support node-local jail submounts

nebius · Jan 8, 2025 · 44b6079 · 44b6079
1 parent 10dfeb6
commit 44b6079
Show file tree

Hide file tree

Showing 8 changed files with 1,269 additions and 20 deletions.
diff --git a/api/v1/slurmcluster_types.go b/api/v1/slurmcluster_types.go
@@ -911,8 +911,13 @@ type NodeVolumeJailSubMount struct {
 	// VolumeSourceName defines the name of the volume source for the sub-mount.
 	// Must correspond to the name of one of [VolumeSource]
 	//
-	// +kubebuilder:validation:Required
-	VolumeSourceName string `json:"volumeSourceName"`
+	// +kubebuilder:validation:Optional
+	VolumeSourceName *string `json:"volumeSourceName"`
+
+	// VolumeClaimTemplateSpec defines the [corev1.PersistentVolumeClaim] template specification
+	//
+	// +kubebuilder:validation:Optional
+	VolumeClaimTemplateSpec *corev1.PersistentVolumeClaimSpec `json:"volumeClaimTemplateSpec,omitempty"`
 }
 
 type Telemetry struct {

diff --git a/api/v1/zz_generated.deepcopy.go b/api/v1/zz_generated.deepcopy.go
diff --git a/config/crd/bases/slurm.nebius.ai_slurmclusters.yaml b/config/crd/bases/slurm.nebius.ai_slurmclusters.yaml
diff --git a/helm/soperator-crds/templates/slurmcluster-crd.yaml b/helm/soperator-crds/templates/slurmcluster-crd.yaml
diff --git a/helm/soperator/crds/slurmcluster-crd.yaml b/helm/soperator/crds/slurmcluster-crd.yaml
diff --git a/internal/render/login/volume.go b/internal/render/login/volume.go
@@ -42,10 +42,19 @@ func renderVolumesAndClaimTemplateSpecs(
 
 	// Jail sub-mounts
 	for _, subMount := range login.JailSubMounts {
-		volumes = append(
-			volumes,
-			common.RenderVolumeFromSource(volumeSources, subMount.VolumeSourceName, subMount.Name),
-		)
+		if v, s, err := common.AddVolumeOrSpec(
+			subMount.VolumeSourceName,
+			func(sourceName string) corev1.Volume {
+				return common.RenderVolumeFromSource(volumeSources, *subMount.VolumeSourceName, subMount.Name)
+			},
+			subMount.VolumeClaimTemplateSpec,
+			subMount.Name,
+		); err != nil {
+			return nil, nil, err
+		} else {
+			volumes = append(volumes, v...)
+			pvcTemplateSpecs = append(pvcTemplateSpecs, s...)
+		}
 	}
 
 	return volumes, pvcTemplateSpecs, nil

diff --git a/internal/render/worker/volume.go b/internal/render/worker/volume.go
@@ -73,10 +73,19 @@ func renderVolumesAndClaimTemplateSpecs(
 
 	// Jail sub-mounts
 	for _, subMount := range worker.JailSubMounts {
-		volumes = append(
-			volumes,
-			common.RenderVolumeFromSource(volumeSources, subMount.VolumeSourceName, subMount.Name),
-		)
+		if v, s, err := common.AddVolumeOrSpec(
+			subMount.VolumeSourceName,
+			func(sourceName string) corev1.Volume {
+				return common.RenderVolumeFromSource(volumeSources, *subMount.VolumeSourceName, subMount.Name)
+			},
+			subMount.VolumeClaimTemplateSpec,
+			subMount.Name,
+		); err != nil {
+			return nil, nil, err
+		} else {
+			volumes = append(volumes, v...)
+			pvcTemplateSpecs = append(pvcTemplateSpecs, s...)
+		}
 	}
 
 	return volumes, pvcTemplateSpecs, nil

diff --git a/internal/values/validate.go b/internal/values/validate.go
@@ -107,11 +107,11 @@ func (c *SlurmCluster) Validate(ctx context.Context) error {
 		}
 		// worker jail sub-mounts
 		for _, subMount := range c.NodeWorker.JailSubMounts {
-			volumeSourceNamesRaw = append(volumeSourceNamesRaw, &subMount.VolumeSourceName)
+			volumeSourceNamesRaw = append(volumeSourceNamesRaw, subMount.VolumeSourceName)
 		}
 		// login jail sub-mounts
 		for _, subMount := range c.NodeLogin.JailSubMounts {
-			volumeSourceNamesRaw = append(volumeSourceNamesRaw, &subMount.VolumeSourceName)
+			volumeSourceNamesRaw = append(volumeSourceNamesRaw, subMount.VolumeSourceName)
 		}
 		for _, volumeSourceName := range volumeSourceNamesRaw {
 			if volumeSourceName == nil {