Add machine type- availability zone precondition. Validate infra is p…

…rovisioned only if the machine exists in one of the zones specified

Update tools/cloud-build/daily-tests/tests/hcls.yml

Co-authored-by: Tom Downes <[email protected]>

Update tools/cloud-build/daily-tests/tests/hcls.yml

Co-authored-by: Tom Downes <[email protected]>

Use europe-west1-d in reservations.tf

community/modules/compute/schedmd-slurm-gcp-v6-nodeset/README.md

@@ -130,14 +130,15 @@ modules. For support with the underlying modules, see the instructions in the
 
 | Name | Version |
 |------|---------|
-| <a name="requirement_terraform"></a> [terraform](#requirement\_terraform) | >= 1.3 |
+| <a name="requirement_terraform"></a> [terraform](#requirement\_terraform) | >= 1.4 |
 | <a name="requirement_google"></a> [google](#requirement\_google) | >= 5.11 |
 
 ## Providers
 
 | Name | Version |
 |------|---------|
 | <a name="provider_google"></a> [google](#provider\_google) | >= 5.11 |
+| <a name="provider_terraform"></a> [terraform](#provider\_terraform) | n/a |
 
 ## Modules
 
@@ -147,7 +148,9 @@ No modules.
 
 | Name | Type |
 |------|------|
+| [terraform_data.machine_type_zone_validation](https://registry.terraform.io/providers/hashicorp/terraform/latest/docs/resources/data) | resource |
 | [google_compute_image.slurm](https://registry.terraform.io/providers/hashicorp/google/latest/docs/data-sources/compute_image) | data source |
+| [google_compute_machine_types.machine_types_by_zone](https://registry.terraform.io/providers/hashicorp/google/latest/docs/data-sources/compute_machine_types) | data source |
 | [google_compute_reservation.reservation](https://registry.terraform.io/providers/hashicorp/google/latest/docs/data-sources/compute_reservation) | data source |
 | [google_compute_zones.available](https://registry.terraform.io/providers/hashicorp/google/latest/docs/data-sources/compute_zones) | data source |
 

community/modules/compute/schedmd-slurm-gcp-v6-nodeset/main.tf

@@ -75,7 +75,7 @@ locals {
     gpu                    = one(local.guest_accelerator)
 
     labels           = local.labels
-    machine_type     = var.machine_type
+    machine_type     = terraform_data.machine_type_zone_validation.output
     metadata         = local.metadata
     min_cpu_platform = var.min_cpu_platform
 
@@ -171,3 +171,28 @@ data "google_compute_reservation" "reservation" {
     # Add a validation that if reservation.project != var.project_id it should be a shared reservation
   }
 }
+
+data "google_compute_machine_types" "machine_types_by_zone" {
+  for_each = local.zones
+  filter   = format("name = \"%s\"", var.machine_type)
+  zone     = each.value
+}
+
+locals {
+  machine_types_by_zone   = data.google_compute_machine_types.machine_types_by_zone
+  zones_with_machine_type = [for k, v in local.machine_types_by_zone : k if length(v.machine_types) > 0]
+}
+
+resource "terraform_data" "machine_type_zone_validation" {
+  input = var.machine_type
+  lifecycle {
+    precondition {
+      condition     = length(local.zones_with_machine_type) > 0
+      error_message = <<-EOT
+        machine type ${var.machine_type} is not available in any of the zones ${jsonencode(local.zones)}". To list zones in which it is available, run:
+
+        gcloud compute machine-types list --filter="name=${var.machine_type}"
+        EOT
+    }
+  }
+}

community/modules/compute/schedmd-slurm-gcp-v6-nodeset/versions.tf

@@ -15,7 +15,7 @@
 */
 
 terraform {
-  required_version = ">= 1.3"
+  required_version = ">= 1.4"
 
   required_providers {
     google = {

tools/cloud-build/daily-tests/tests/hcls.yml

@@ -19,7 +19,7 @@ deployment_name: "hcls-v6-{{ build }}"
 # No non-alphanumerical characters in the slurm cluster name - they will be
 # removed by Cluster Toolkit slurm wrappers, which will break the playbook
 slurm_cluster_name: "hclsv6{{ build[0:4] }}"
-zone: europe-west1-d
+zone: europe-west1-c
 workspace: /workspace
 blueprint_yaml: "{{ workspace }}/examples/hcls-blueprint.yaml"
 network: "{{ test_name }}-net"

tools/cloud-build/daily-tests/tests/hpc-enterprise-slurm.yml

@@ -18,12 +18,12 @@ deployment_name: "enter-{{ build }}"
 # Manually adding the slurm_cluster_name for use in node names, which filters
 # non-alphanumeric chars and is capped at 10 chars.
 slurm_cluster_name: "enter{{ build[0:5] }}"
-zone: europe-west1-d
+zone: europe-west4-c
 cli_deployment_vars:
   network_name: "{{ network }}"
-  region: europe-west1
+  region: europe-west4
   zone: "{{ zone }}"
-  gpu_zones: "[europe-west1-b,europe-west1-c,europe-west1-d]"
+  gpu_zones: "[europe-west4-a,europe-west4-b,europe-west4-c]"
 workspace: /workspace
 blueprint_yaml: "{{ workspace }}/examples/hpc-enterprise-slurm.yaml"
 network: "{{ test_name }}-net"

tools/cloud-build/daily-tests/tests/slurm-v6-rocky8.yml

@@ -22,10 +22,10 @@ slurm_cluster_name: "rock8{{ build[0:5] }}"
 
 cli_deployment_vars:
   network_name: "{{ network }}"
-  region: us-west4
-  zone: us-west4-c
+  region: us-central1
+  zone: us-central1-a
 
-zone: us-west4-c
+zone: us-central1-a
 workspace: /workspace
 blueprint_yaml: "{{ workspace }}/examples/hpc-slurm.yaml"
 network: "{{ test_name }}-net"

tools/cloud-build/provision/reservations.tf

@@ -30,8 +30,8 @@ resource "google_compute_reservation" "c2standard60_us_west4_c" {
 }
 
 resource "google_compute_reservation" "n1standard8_with_tesla_t4_europe_west1_d" {
-  name        = "n1standard8-with-tesla-t4-europe-west1-d"
-  zone        = "europe-west1-d"
+  name        = "n1standard8-with-tesla-t4-europe-west1-c"
+  zone        = "europe-west1-c"
   description = local.reservation_description
 
   specific_reservation {