- add capacity reservations

k8s-training/environment.sh

@@ -52,9 +52,8 @@ NEBIUS_SA_NAME="k8s-training-sa"
 NEBIUS_SA_ID=$(nebius iam service-account get-by-name \
   --parent-id "${NEBIUS_PROJECT_ID}" \
   --name "${NEBIUS_SA_NAME}" \
-  --format json \
-  | jq -r '.metadata.id')
-
+  --format json 2>/dev/null \
+  | jq -r '.metadata.id // empty')
 
 if [ -z "$NEBIUS_SA_ID" ]; then
   NEBIUS_SA_ID=$(nebius iam service-account create \

k8s-training/main.tf

@@ -166,6 +166,10 @@ resource "nebius_mk8s_v1_node_group" "gpu" {
     ] : null
     gpu_cluster  = var.enable_gpu_cluster ? nebius_compute_v1_gpu_cluster.fabric_2[0] : null
     gpu_settings = var.gpu_nodes_driverfull_image ? { drivers_preset = local.device_preset } : null
+    reservation_policy = var.gpu_nodes_reservation_policy != null ? {
+      policy          = var.gpu_nodes_reservation_policy.policy
+      reservation_ids = var.gpu_nodes_reservation_policy.reservation_ids
+    } : null
     preemptible = var.gpu_nodes_preemptible ? {
       on_preemption = "STOP"
       priority      = 3

k8s-training/terraform.tfvars

@@ -39,6 +39,12 @@ gpu_nodes_preset   = "8gpu-128vcpu-1600gb" # GPU nodes preset: 8gpu-128vcpu-1600
 # Infiniband fabrics: https://docs.nebius.com/compute/clusters/gpu#fabrics
 infiniband_fabric = "" # Infiniband fabric name
 
+# gpu_nodes_reservation_policy = {
+#     policy          = "STRICT"
+#     reservation_ids = ["capacityblockgroup-e00xxxxx"]
+# }
+
+
 gpu_nodes_driverfull_image = true
 enable_k8s_node_group_sa   = true
 enable_egress_gateway      = false

k8s-training/variables.tf

@@ -387,6 +387,20 @@ variable "gpu_nodes_preemptible" {
   default     = false
 }
 
+variable "gpu_nodes_reservation_policy" {
+  description = "Reservation policy for GPU node groups. Policy can be AUTO, STRICT, or FORBID."
+  type = object({
+    policy          = string
+    reservation_ids = list(string)
+  })
+  default = null
+
+  validation {
+    condition     = var.gpu_nodes_reservation_policy == null || contains(["AUTO", "STRICT", "FORBID"], var.gpu_nodes_reservation_policy.policy)
+    error_message = "Policy must be one of AUTO, STRICT, or FORBID."
+  }
+}
+
 variable "custom_driver" {
   description = "Use customized driver for the GPU Operator, e.g. to run Cuda 13 on H200"
   type        = bool