Merge pull request #3434 from ighosh98/experimental

[Cherry Pick] Making upgrade settings configurable

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

@@ -348,6 +348,7 @@ limitations under the License.
 | <a name="input_timeout_update"></a> [timeout\_update](#input\_timeout\_update) | Timeout for updating a node pool | `string` | `null` | no |
 | <a name="input_total_max_nodes"></a> [total\_max\_nodes](#input\_total\_max\_nodes) | DEPRECATED: Use autoscaling\_total\_max\_nodes. | `number` | `null` | no |
 | <a name="input_total_min_nodes"></a> [total\_min\_nodes](#input\_total\_min\_nodes) | DEPRECATED: Use autoscaling\_total\_min\_nodes. | `number` | `null` | no |
+| <a name="input_upgrade_settings"></a> [upgrade\_settings](#input\_upgrade\_settings) | Defines node pool upgrade settings. It is highly recommended that you define all max\_surge and max\_unavailable.<br/>If max\_surge is not specified, it would be set to a default value of 0.<br/>If max\_unavailable is not specified, it would be set to a default value of 1. | <pre>object({<br/>    strategy        = string<br/>    max_surge       = optional(number)<br/>    max_unavailable = optional(number)<br/>  })</pre> | <pre>{<br/>  "max_surge": 0,<br/>  "max_unavailable": 1,<br/>  "strategy": "SURGE"<br/>}</pre> | no |
 | <a name="input_zones"></a> [zones](#input\_zones) | A list of zones to be used. Zones must be in region of cluster. If null, cluster zones will be inherited. Note `zones` not `zone`; does not work with `zone` deployment variable. | `list(string)` | `null` | no |
 
 ## Outputs

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

@@ -19,6 +19,14 @@ locals {
   labels = merge(var.labels, { ghpc_module = "gke-node-pool", ghpc_role = "compute" })
 }
 
+locals {
+  upgrade_settings = {
+    strategy        = var.upgrade_settings.strategy
+    max_surge       = coalesce(var.upgrade_settings.max_surge, 0)
+    max_unavailable = coalesce(var.upgrade_settings.max_unavailable, 1)
+  }
+}
+
 locals {
   has_gpu = length(local.guest_accelerator) > 0
   gpu_taint = local.has_gpu ? [{
@@ -60,9 +68,9 @@ resource "google_container_node_pool" "node_pool" {
   }
 
   upgrade_settings {
-    strategy        = "SURGE"
-    max_surge       = 0
-    max_unavailable = 1
+    strategy        = local.upgrade_settings.strategy
+    max_surge       = local.upgrade_settings.max_surge
+    max_unavailable = local.upgrade_settings.max_unavailable
   }
 
   dynamic "placement_policy" {
@@ -262,6 +270,22 @@ resource "google_container_node_pool" "node_pool" {
       )
       error_message = "Shared Extended reservations are not supported by GKE."
     }
+    precondition {
+      condition     = contains(["SURGE"], local.upgrade_settings.strategy)
+      error_message = "Only SURGE strategy is supported"
+    }
+    precondition {
+      condition     = local.upgrade_settings.max_unavailable >= 0
+      error_message = "max_unavailable should be set to 0 or greater"
+    }
+    precondition {
+      condition     = local.upgrade_settings.max_surge >= 0
+      error_message = "max_surge should be set to 0 or greater"
+    }
+    precondition {
+      condition     = local.upgrade_settings.max_unavailable > 0 || local.upgrade_settings.max_surge > 0
+      error_message = "At least one of max_unavailable or max_surge must greater than 0"
+    }
   }
 }
 

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

@@ -391,3 +391,21 @@ variable "node_version" {
   type        = string
   default     = null
 }
+
+variable "upgrade_settings" {
+  description = <<-EOT
+  Defines node pool upgrade settings. It is highly recommended that you define all max_surge and max_unavailable.
+  If max_surge is not specified, it would be set to a default value of 0.
+  If max_unavailable is not specified, it would be set to a default value of 1.  
+  EOT
+  type = object({
+    strategy        = string
+    max_surge       = optional(number)
+    max_unavailable = optional(number)
+  })
+  default = {
+    strategy        = "SURGE"
+    max_surge       = 0
+    max_unavailable = 1
+  }
+}