fix(terraform): cap EPHEMERAL volume size to reserve disk space for swap partition on worker nodes

terraform/main.tf

@@ -102,6 +102,7 @@ resource "local_file" "worker_configs" {
     gpu_count           = each.value.gpu_count
     extra_disks         = each.value.extra_disks
     swap_size           = each.value.swap_size
+    ephemeral_max_size  = each.value.ephemeral_max_size
 
     # Cluster config
     cluster_id      = var.cluster_id

terraform/templates/worker.tftpl

@@ -44,8 +44,7 @@ machine:
       validSubnets:
         - 192.168.1.0/24
   install:
-    diskSelector:
-      match: disk.transport == "nvme"
+    disk: ${install_disk}
     image: ${factory_image}
     wipe: true
     grubUseUKICmdline: true
@@ -100,6 +99,15 @@ cluster:
 %{ if swap_size != "" ~}
 ---
 apiVersion: v1alpha1
+kind: VolumeConfig
+name: EPHEMERAL
+provisioning:
+  diskSelector:
+    match: disk.transport == "nvme"
+  maxSize: ${ephemeral_max_size}
+  grow: false
+---
+apiVersion: v1alpha1
 kind: SwapVolumeConfig
 name: swap1
 provisioning:

terraform/variables.tf

@@ -133,6 +133,7 @@ variable "worker_configs" {
     gpu_count          = optional(number, 0)
     factory_image      = optional(string)
     swap_size          = optional(string, "")
+    ephemeral_max_size = optional(string, "700GiB")
     extra_disks = optional(list(object({
       device     = string
       mountpoint = string