Merge branch 'main' into fix/unauthorized-kms

* main:
  fix: remove old variables from node groups
  fix: revert instance type
  fix: test utils code
  ref: following suggestion in issue 3068
  fix: run precommit
  REF: update based on code review
  fix: revert qhub changes
  Include functionality for spot instances on AWS

Author: satra

src/_nebari/stages/infrastructure/__init__.py

@@ -188,6 +188,7 @@ class AWSNodeGroupInputVars(schema.Base):
     max_size: int
     single_subnet: bool
     permissions_boundary: Optional[str] = None
+    spot: bool = False
     ami_type: Optional[AWSAmiTypes] = None
     launch_template: Optional[AWSNodeLaunchTemplate] = None
     node_taints: list[dict]
@@ -521,6 +522,7 @@ class AWSNodeGroup(NodeGroup):
     gpu: bool = False
     single_subnet: bool = False
     permissions_boundary: Optional[str] = None
+    spot: bool = False
     # Disabled as part of 2024.11.1 until #2832 is resolved
     # launch_template: Optional[AWSNodeLaunchTemplate] = None
 
@@ -959,6 +961,7 @@ def input_vars(self, stage_outputs: Dict[str, Dict[str, Any]]):
                             gpu_enabled=node_group.gpu,
                             launch_template=None,
                         ),
+                        spot=node_group.spot,
                     )
                     for name, node_group in self.config.amazon_web_services.node_groups.items()
                 ],

src/_nebari/stages/infrastructure/template/aws/modules/kubernetes/main.tf

@@ -92,6 +92,7 @@ resource "aws_eks_node_group" "main" {
   subnet_ids      = var.node_groups[count.index].single_subnet ? [element(var.cluster_subnets, 0)] : var.cluster_subnets
 
   instance_types = [var.node_groups[count.index].instance_type]
+  capacity_type  = var.node_groups[count.index].spot ? "SPOT" : "ON_DEMAND"
   ami_type       = var.node_groups[count.index].ami_type
   disk_size      = var.node_groups[count.index].launch_template == null ? 50 : null
 

src/_nebari/stages/infrastructure/template/aws/modules/kubernetes/variables.tf

@@ -46,6 +46,7 @@ variable "node_groups" {
   type = list(object({
     name            = string
     instance_type   = string
+    spot            = bool
     gpu             = bool
     min_size        = number
     desired_size    = number

src/_nebari/stages/infrastructure/template/aws/variables.tf

@@ -33,6 +33,7 @@ variable "node_groups" {
   type = list(object({
     name            = string
     instance_type   = string
+    spot            = bool
     gpu             = bool
     min_size        = number
     desired_size    = number

tests/common/config_mod_utils.py

@@ -130,13 +130,13 @@ def add_preemptible_node_group(config, cloud="aws"):
     node_group = None
     if cloud == "aws":
         cloud_name = "amazon_web_services"
-        # TODO: how to make preemptible?
         node_group = AWSNodeGroup(
             instance="m5.xlarge",
             min_nodes=1,
             max_nodes=5,
             single_subnet=False,
             taints=DEFAULT_NODE_GROUP_TAINTS,
+            spot=True,
         )
     elif cloud == "gcp":
         cloud_name = "google_cloud_platform"