feat: Add resource-policy accelerator_topology_mode (#5393)

Author: jamOne-

modules/compute/gke-node-pool/README.md

@@ -280,15 +280,15 @@ limitations under the License.
 |------|---------|
 | <a name="requirement_terraform"></a> [terraform](#requirement\_terraform) | = 1.12.2 |
 | <a name="requirement_google"></a> [google](#requirement\_google) | >= 7.2 |
-| <a name="requirement_google-beta"></a> [google-beta](#requirement\_google-beta) | >= 7.2 |
+| <a name="requirement_google-beta"></a> [google-beta](#requirement\_google-beta) | >= 7.24.0 |
 | <a name="requirement_null"></a> [null](#requirement\_null) | ~> 3.0 |
 
 ## Providers
 
 | Name | Version |
 |------|---------|
 | <a name="provider_google"></a> [google](#provider\_google) | >= 7.2 |
-| <a name="provider_google-beta"></a> [google-beta](#provider\_google-beta) | >= 7.2 |
+| <a name="provider_google-beta"></a> [google-beta](#provider\_google-beta) | >= 7.24.0 |
 | <a name="provider_null"></a> [null](#provider\_null) | ~> 3.0 |
 
 ## Modules
@@ -316,6 +316,7 @@ limitations under the License.
 
 | Name | Description | Type | Default | Required |
 |------|-------------|------|---------|:--------:|
+| <a name="input_accelerator_topology_mode"></a> [accelerator\_topology\_mode](#input\_accelerator\_topology\_mode) | The accelerator topology mode for the resource policy. It accepts values like `PROVISION_ONLY` or `AUTO_CONNECT`. Note that `enable_queued_provisioning` (DWS) is not supported when `accelerator_topology_mode` is set to `PROVISION_ONLY`. | `string` | `null` | no |
 | <a name="input_additional_networks"></a> [additional\_networks](#input\_additional\_networks) | Additional network interface details for GKE, if any. Providing additional networks adds additional node networks to the node pool | <pre>list(object({<br/>    network            = string<br/>    subnetwork         = string<br/>    subnetwork_project = string<br/>    network_ip         = string<br/>    nic_type           = string<br/>    stack_type         = string<br/>    queue_count        = number<br/>    access_config = list(object({<br/>      nat_ip       = string<br/>      network_tier = string<br/>    }))<br/>    ipv6_access_config = list(object({<br/>      network_tier = string<br/>    }))<br/>    alias_ip_range = list(object({<br/>      ip_cidr_range         = string<br/>      subnetwork_range_name = string<br/>    }))<br/>  }))</pre> | `[]` | no |
 | <a name="input_auto_repair"></a> [auto\_repair](#input\_auto\_repair) | Whether the nodes will be automatically repaired. | `bool` | `true` | no |
 | <a name="input_auto_upgrade"></a> [auto\_upgrade](#input\_auto\_upgrade) | Whether the nodes will be automatically upgraded. | `bool` | `false` | no |
@@ -331,7 +332,7 @@ limitations under the License.
 | <a name="input_enable_gcfs"></a> [enable\_gcfs](#input\_enable\_gcfs) | Enable the Google Container Filesystem (GCFS). See [restrictions](https://registry.terraform.io/providers/hashicorp/google/latest/docs/resources/container_cluster#gcfs_config). | `bool` | `false` | no |
 | <a name="input_enable_numa_aware_scheduling"></a> [enable\_numa\_aware\_scheduling](#input\_enable\_numa\_aware\_scheduling) | Enable [NUMA-aware](https://cloud.google.com/kubernetes-engine/distributed-cloud/bare-metal/docs/vm-runtime/numa) scheduling. | `bool` | `false` | no |
 | <a name="input_enable_private_nodes"></a> [enable\_private\_nodes](#input\_enable\_private\_nodes) | Whether nodes have internal IP addresses only. | `bool` | `true` | no |
-| <a name="input_enable_queued_provisioning"></a> [enable\_queued\_provisioning](#input\_enable\_queued\_provisioning) | If true, enables Dynamic Workload Scheduler and adds the cloud.google.com/gke-queued taint to the node pool. | `bool` | `false` | no |
+| <a name="input_enable_queued_provisioning"></a> [enable\_queued\_provisioning](#input\_enable\_queued\_provisioning) | If true, enables Dynamic Workload Scheduler and adds the cloud.google.com/gke-queued taint to the node pool. This cannot be true if `accelerator_topology_mode` is set to `PROVISION_ONLY`. | `bool` | `false` | no |
 | <a name="input_enable_secure_boot"></a> [enable\_secure\_boot](#input\_enable\_secure\_boot) | Enable secure boot for the nodes.  Keep enabled unless custom kernel modules need to be loaded. See [here](https://cloud.google.com/compute/shielded-vm/docs/shielded-vm#secure-boot) for more info. | `bool` | `true` | no |
 | <a name="input_gke_version"></a> [gke\_version](#input\_gke\_version) | GKE version | `string` | n/a | yes |
 | <a name="input_guest_accelerator"></a> [guest\_accelerator](#input\_guest\_accelerator) | List of the type and count of accelerator cards attached to the instance. | <pre>list(object({<br/>    type  = optional(string)<br/>    count = optional(number, 0)<br/>    gpu_driver_installation_config = optional(object({<br/>      gpu_driver_version = string<br/>    }), { gpu_driver_version = "DEFAULT" })<br/>    gpu_partition_size = optional(string)<br/>    gpu_sharing_config = optional(object({<br/>      gpu_sharing_strategy       = string<br/>      max_shared_clients_per_gpu = number<br/>    }))<br/>  }))</pre> | `[]` | no |

modules/compute/gke-node-pool/main.tf

@@ -417,6 +417,10 @@ resource "google_container_node_pool" "node_pool" {
       condition     = var.spot == true ? (var.reservation_affinity.consume_reservation_type == "NO_RESERVATION") : true
       error_message = "Spot consumption option only works with reservation_affinity consume_reservation_type NO_RESERVATION."
     }
+    precondition {
+      condition     = !(var.accelerator_topology_mode == "PROVISION_ONLY" && var.enable_queued_provisioning == true)
+      error_message = "Custom accelerator topology modes (like PROVISION_ONLY) are incompatible with Dynamic Workload Scheduler (queued provisioning)."
+    }
   }
 }
 

modules/compute/gke-node-pool/variables.tf

@@ -450,7 +450,7 @@ variable "run_workload_script" {
 }
 
 variable "enable_queued_provisioning" {
-  description = "If true, enables Dynamic Workload Scheduler and adds the cloud.google.com/gke-queued taint to the node pool."
+  description = "If true, enables Dynamic Workload Scheduler and adds the cloud.google.com/gke-queued taint to the node pool. This cannot be true if `accelerator_topology_mode` is set to `PROVISION_ONLY`."
   type        = bool
   default     = false
 }
@@ -525,3 +525,10 @@ variable "linux_node_config" {
   default  = {}
   nullable = false
 }
+
+
+variable "accelerator_topology_mode" {
+  description = "The accelerator topology mode for the resource policy. It accepts values like `PROVISION_ONLY` or `AUTO_CONNECT`. Note that `enable_queued_provisioning` (DWS) is not supported when `accelerator_topology_mode` is set to `PROVISION_ONLY`."
+  type        = string
+  default     = null
+}

modules/compute/gke-node-pool/versions.tf

@@ -22,7 +22,7 @@ terraform {
     }
     google-beta = {
       source  = "hashicorp/google-beta"
-      version = ">= 7.2"
+      version = ">= 7.24.0"
     }
     null = {
       source  = "hashicorp/null"

modules/compute/resource-policy/README.md

@@ -43,14 +43,14 @@ limitations under the License.
 | Name | Version |
 |------|---------|
 | <a name="requirement_terraform"></a> [terraform](#requirement\_terraform) | = 1.12.2 |
-| <a name="requirement_google-beta"></a> [google-beta](#requirement\_google-beta) | >= 6.29.0 |
+| <a name="requirement_google-beta"></a> [google-beta](#requirement\_google-beta) | >= 7.24.0 |
 | <a name="requirement_random"></a> [random](#requirement\_random) | ~> 3.0 |
 
 ## Providers
 
 | Name | Version |
 |------|---------|
-| <a name="provider_google-beta"></a> [google-beta](#provider\_google-beta) | >= 6.29.0 |
+| <a name="provider_google-beta"></a> [google-beta](#provider\_google-beta) | >= 7.24.0 |
 | <a name="provider_random"></a> [random](#provider\_random) | ~> 3.0 |
 
 ## Modules
@@ -72,11 +72,12 @@ No modules.
 | <a name="input_name"></a> [name](#input\_name) | The resource policy's name. | `string` | n/a | yes |
 | <a name="input_project_id"></a> [project\_id](#input\_project\_id) | The project ID for the resource policy. | `string` | n/a | yes |
 | <a name="input_region"></a> [region](#input\_region) | The region for the the resource policy. | `string` | n/a | yes |
-| <a name="input_workload_policy"></a> [workload\_policy](#input\_workload\_policy) | Describes the workload policy | <pre>object({<br/>    type                  = optional(string, null)<br/>    max_topology_distance = optional(string, null)<br/>    accelerator_topology  = optional(string, null)<br/>  })</pre> | <pre>{<br/>  "accelerator_topology": null,<br/>  "max_topology_distance": null,<br/>  "type": null<br/>}</pre> | no |
+| <a name="input_workload_policy"></a> [workload\_policy](#input\_workload\_policy) | Describes the workload policy | <pre>object({<br/>    type                      = optional(string, null)<br/>    max_topology_distance     = optional(string, null)<br/>    accelerator_topology      = optional(string, null)<br/>    accelerator_topology_mode = optional(string, null)<br/>  })</pre> | <pre>{<br/>  "accelerator_topology": null,<br/>  "accelerator_topology_mode": null,<br/>  "max_topology_distance": null,<br/>  "type": null<br/>}</pre> | no |
 
 ## Outputs
 
 | Name | Description |
 |------|-------------|
+| <a name="output_accelerator_topology_mode"></a> [accelerator\_topology\_mode](#output\_accelerator\_topology\_mode) | The accelerator topology mode for the resource policy. |
 | <a name="output_placement_policy"></a> [placement\_policy](#output\_placement\_policy) | Group placement policy to use for placing VMs or GKE nodes placement. `COMPACT` is the only supported value for `type` currently. `name` is the name of the placement policy.<br/>It is assumed that the specified policy exists. To create a placement policy refer to https://cloud.google.com/sdk/gcloud/reference/compute/resource-policies/create/group-placement.<br/>Note: Placement policies have the [following](https://cloud.google.com/compute/docs/instances/placement-policies-overview#restrictions-compact-policies) restrictions.<br/>The value `tpu_topology` is only used for TPU node pools. The `gke-node-pool` module ensures it is configured appropriately for only TPUs during placement policy mapping. |
 <!-- END OF PRE-COMMIT-TERRAFORM DOCS HOOK -->

modules/compute/resource-policy/main.tf

@@ -31,9 +31,10 @@ resource "google_compute_resource_policy" "policy" {
     for_each = var.workload_policy.type != null ? [1] : []
 
     content {
-      type                  = var.workload_policy.type
-      max_topology_distance = var.workload_policy.max_topology_distance
-      accelerator_topology  = var.workload_policy.accelerator_topology
+      type                      = var.workload_policy.type
+      max_topology_distance     = var.workload_policy.max_topology_distance
+      accelerator_topology      = var.workload_policy.accelerator_topology
+      accelerator_topology_mode = var.workload_policy.accelerator_topology_mode
     }
   }
 
@@ -45,4 +46,11 @@ resource "google_compute_resource_policy" "policy" {
       max_distance = var.group_placement_max_distance
     }
   }
+
+  lifecycle {
+    precondition {
+      condition     = var.workload_policy.accelerator_topology_mode == null || (var.workload_policy.type != null && var.workload_policy.accelerator_topology != null)
+      error_message = "Both workload_policy.type and workload_policy.accelerator_topology must be set when accelerator_topology_mode is specified."
+    }
+  }
 }

modules/compute/resource-policy/outputs.tf

@@ -28,3 +28,8 @@ output "placement_policy" {
     tpu_topology = (var.workload_policy.type != null) ? var.workload_policy.accelerator_topology : null
   }
 }
+
+output "accelerator_topology_mode" {
+  description = "The accelerator topology mode for the resource policy."
+  value       = var.workload_policy.accelerator_topology_mode
+}

modules/compute/resource-policy/variables.tf

@@ -51,14 +51,16 @@ variable "group_placement_max_distance" {
 variable "workload_policy" {
   description = "Describes the workload policy"
   type = object({
-    type                  = optional(string, null)
-    max_topology_distance = optional(string, null)
-    accelerator_topology  = optional(string, null)
+    type                      = optional(string, null)
+    max_topology_distance     = optional(string, null)
+    accelerator_topology      = optional(string, null)
+    accelerator_topology_mode = optional(string, null)
   })
   default = {
-    type                  = null
-    max_topology_distance = null
-    accelerator_topology  = null
+    type                      = null
+    max_topology_distance     = null
+    accelerator_topology      = null
+    accelerator_topology_mode = null
   }
   nullable = false
 }

modules/compute/resource-policy/versions.tf

@@ -18,7 +18,7 @@ terraform {
   required_providers {
     google-beta = {
       source  = "hashicorp/google-beta"
-      version = ">= 6.29.0"
+      version = ">= 7.24.0"
     }
     random = {
       source  = "hashicorp/random"