support additional cloudsql machine types

README.md

@@ -15,6 +15,20 @@ Please review the README.md in the examples for all Pre-deployment and Post-Depl
 
 The default configuration is a large production-sized deployment. Please consider that when testing and adjust the configuration to use smaller sized resources.
 
+## Cloud SQL machine types
+
+The module supports only Cloud SQL Enterprise Plus.
+It selects the disk type from `postgres_machine_type`.
+The default remains `db-perf-optimized-N-8`
+A change to the database machine type causes downtime.
+
+| Machine type | Disk |
+| --- | --- |
+| `db-perf-optimized-N-*` | SSD |
+| `db-c4a-highmem-*` | Hyperdisk Balanced |
+| `db-perf-optimized-C4-*` | Hyperdisk Balanced |
+
+
 ## Development Setup
 
 This section is only relevant if you are a contributor who wants to make changes to this module. All others can skip this section.

examples/braintrust-data-plane/main.tf

@@ -57,8 +57,9 @@ module "braintrust-data-plane" {
   # postgres_machine_type = "db-perf-optimized-N-8"
   # postgres_availability_type = "REGIONAL"
   # postgres_disk_size = 1000
-  # Does this auto expand? how do we handle that?
-  # How do we control disk perf IOPS/etc
+  # Optional Hyperdisk performance values when using C4A or C4 machine types. Null uses Cloud SQL defaults.
+  # postgres_disk_provisioned_iops       = 12000
+  # postgres_disk_provisioned_throughput = 500
 
   ### Redis configuration
   # redis_version = "REDIS_7_2"

examples/braintrust-data-plane/versions.tf

@@ -4,11 +4,11 @@ terraform {
   required_providers {
     google = {
       source  = "hashicorp/google"
-      version = "~> 6.45"
+      version = ">= 8.2.0, < 9.0.0"
     }
     google-beta = {
       source  = "hashicorp/google-beta"
-      version = "~> 6.45"
+      version = ">= 8.2.0, < 9.0.0"
     }
   }
 }

main.tf

@@ -25,18 +25,20 @@ module "kms" {
 module "database" {
   source = "./modules/database"
 
-  deployment_name              = var.deployment_name
-  custom_labels                = var.custom_labels
-  postgres_network             = var.create_vpc ? module.vpc[0].network_self_link : var.existing_network_self_link
-  postgres_kms_cmek_id         = module.kms.kms_key_id
-  postgres_version             = var.postgres_version
-  postgres_availability_type   = var.postgres_availability_type
-  postgres_machine_type        = var.postgres_machine_type
-  postgres_disk_size           = var.postgres_disk_size
-  postgres_enable_seqscan      = var.postgres_enable_seqscan
-  postgres_backup_start_time   = var.postgres_backup_start_time
-  postgres_maintenance_window  = var.postgres_maintenance_window
-  postgres_deletion_protection = var.postgres_deletion_protection
+  deployment_name                      = var.deployment_name
+  custom_labels                        = var.custom_labels
+  postgres_network                     = var.create_vpc ? module.vpc[0].network_self_link : var.existing_network_self_link
+  postgres_kms_cmek_id                 = module.kms.kms_key_id
+  postgres_version                     = var.postgres_version
+  postgres_availability_type           = var.postgres_availability_type
+  postgres_machine_type                = var.postgres_machine_type
+  postgres_disk_provisioned_iops       = var.postgres_disk_provisioned_iops
+  postgres_disk_provisioned_throughput = var.postgres_disk_provisioned_throughput
+  postgres_disk_size                   = var.postgres_disk_size
+  postgres_enable_seqscan              = var.postgres_enable_seqscan
+  postgres_backup_start_time           = var.postgres_backup_start_time
+  postgres_maintenance_window          = var.postgres_maintenance_window
+  postgres_deletion_protection         = var.postgres_deletion_protection
 
   depends_on = [module.vpc]
 }

modules/database/main.tf

@@ -1,4 +1,6 @@
 locals {
+  postgres_uses_hyperdisk = can(regex("^db-(c4a-highmem|perf-optimized-C4)-[0-9]+$", var.postgres_machine_type))
+
   common_labels = merge(var.custom_labels, {
     braintrustdeploymentname = var.deployment_name
   })
@@ -53,11 +55,15 @@ resource "google_sql_database_instance" "braintrust" {
   deletion_protection = var.postgres_deletion_protection
 
   settings {
-    availability_type = var.postgres_availability_type
-    tier              = var.postgres_machine_type
-    disk_type         = "PD_SSD"
-    disk_size         = var.postgres_disk_size
-    disk_autoresize   = true
+    availability_type                = var.postgres_availability_type
+    tier                             = var.postgres_machine_type
+    edition                          = "ENTERPRISE_PLUS"
+    disk_type                        = local.postgres_uses_hyperdisk ? "HYPERDISK_BALANCED" : "PD_SSD"
+    data_disk_provisioned_iops       = var.postgres_disk_provisioned_iops
+    data_disk_provisioned_throughput = var.postgres_disk_provisioned_throughput
+    disk_size                        = var.postgres_disk_size
+    disk_autoresize                  = true
+    disk_autoresize_limit            = 0
 
     # Braintrust will create a high number of connections to the database. Setting this to an extremely high amount of connections as changing this requires a DB restart.
     database_flags {

modules/database/variables.tf

@@ -35,14 +35,67 @@ variable "postgres_availability_type" {
 
 variable "postgres_machine_type" {
   type        = string
-  description = "Machine size of Cloud SQL for PostgreSQL instance."
+  description = "Enterprise Plus machine type from the N2, C4A, or C4 series. The machine type selects compatible storage settings."
   default     = "db-perf-optimized-N-8"
+
+  validation {
+    condition     = can(regex("^db-(perf-optimized-(N|C4)|c4a-highmem)-[0-9]+$", var.postgres_machine_type))
+    error_message = "Use an Enterprise Plus machine type from the N2, C4A, or C4 series."
+  }
+
+  validation {
+    condition     = !contains(["db-c4a-highmem-2", "db-perf-optimized-C4-2"], var.postgres_machine_type)
+    error_message = "C4A and C4 machine types require at least four vCPUs."
+  }
+}
+
+variable "postgres_disk_provisioned_iops" {
+  type        = number
+  description = "Hyperdisk IOPS. Null uses the Cloud SQL default for the disk size and machine type."
+  default     = null
+
+  validation {
+    condition     = var.postgres_disk_provisioned_iops == null ? true : var.postgres_disk_provisioned_iops >= 3000 && floor(var.postgres_disk_provisioned_iops) == var.postgres_disk_provisioned_iops
+    error_message = "Hyperdisk IOPS must be an integer of at least 3000."
+  }
+
+  validation {
+    condition     = var.postgres_disk_provisioned_iops == null || can(regex("^db-(c4a-highmem|perf-optimized-C4)-[0-9]+$", var.postgres_machine_type))
+    error_message = "Custom IOPS require a C4A or C4 machine type with Hyperdisk Balanced."
+  }
+}
+
+variable "postgres_disk_provisioned_throughput" {
+  type        = number
+  description = "Hyperdisk throughput in MiB/s. Null uses the Cloud SQL default for the disk size and machine type."
+  default     = null
+
+  validation {
+    condition     = var.postgres_disk_provisioned_throughput == null ? true : var.postgres_disk_provisioned_throughput >= 140 && floor(var.postgres_disk_provisioned_throughput) == var.postgres_disk_provisioned_throughput
+    error_message = "Hyperdisk throughput must be an integer of at least 140 MiB/s."
+  }
+
+  validation {
+    condition     = var.postgres_disk_provisioned_throughput == null || can(regex("^db-(c4a-highmem|perf-optimized-C4)-[0-9]+$", var.postgres_machine_type))
+    error_message = "Custom throughput requires a C4A or C4 machine type with Hyperdisk Balanced."
+  }
 }
 
 variable "postgres_disk_size" {
   type        = number
-  description = "Size in GB of PostgreSQL disk."
+  description = "Initial PostgreSQL disk size in GB. Terraform ignores later changes to this value."
   default     = 1000
+  nullable    = false
+
+  validation {
+    condition     = var.postgres_disk_size >= 10 && floor(var.postgres_disk_size) == var.postgres_disk_size
+    error_message = "The initial disk size must be an integer of at least 10 GB."
+  }
+
+  validation {
+    condition     = !can(regex("^db-(c4a-highmem|perf-optimized-C4)-[0-9]+$", var.postgres_machine_type)) || var.postgres_disk_size >= 20
+    error_message = "Hyperdisk Balanced requires a disk size of at least 20 GB."
+  }
 }
 
 variable "postgres_enable_seqscan" {

modules/database/versions.tf

@@ -4,11 +4,11 @@ terraform {
   required_providers {
     google = {
       source  = "hashicorp/google"
-      version = "~> 6.45"
+      version = ">= 8.2.0, < 9.0.0"
     }
     google-beta = {
       source  = "hashicorp/google-beta"
-      version = "~> 6.45"
+      version = ">= 8.2.0, < 9.0.0"
     }
     random = {
       source  = "hashicorp/random"

modules/gke-cluster/main.tf

@@ -123,13 +123,6 @@ resource "google_container_cluster" "braintrust_autopilot" {
     google_kms_crypto_key_iam_member.gke_compute_cmek
   ]
 
-  lifecycle {
-    # GKE Autopilot may report ALL_OBJECTS_ENCRYPTION_ENABLED after create, but
-    # the Terraform provider currently accepts only ENCRYPTED/DECRYPTED as input.
-    ignore_changes = [
-      database_encryption[0].state,
-    ]
-  }
 }
 
 #----------------------------------------------------------------------------------------------

modules/gke-cluster/versions.tf

@@ -4,11 +4,7 @@ terraform {
   required_providers {
     google = {
       source  = "hashicorp/google"
-      version = "~> 6.45"
-    }
-    random = {
-      source  = "hashicorp/random"
-      version = ">= 3.7.2"
+      version = ">= 8.2.0, < 9.0.0"
     }
   }
 }

modules/gke-iam/versions.tf

@@ -4,15 +4,7 @@ terraform {
   required_providers {
     google = {
       source  = "hashicorp/google"
-      version = "~> 6.45"
-    }
-    google-beta = {
-      source  = "hashicorp/google-beta"
-      version = "~> 6.45"
-    }
-    random = {
-      source  = "hashicorp/random"
-      version = ">= 3.7.2"
+      version = ">= 8.2.0, < 9.0.0"
     }
   }
 }

modules/kms/versions.tf

@@ -4,7 +4,7 @@ terraform {
   required_providers {
     google = {
       source  = "hashicorp/google"
-      version = "~> 6.45"
+      version = ">= 8.2.0, < 9.0.0"
     }
     random = {
       source  = "hashicorp/random"

modules/redis/versions.tf

@@ -4,11 +4,7 @@ terraform {
   required_providers {
     google = {
       source  = "hashicorp/google"
-      version = "~> 6.45"
-    }
-    random = {
-      source  = "hashicorp/random"
-      version = ">= 3.7.2"
+      version = ">= 8.2.0, < 9.0.0"
     }
   }
 }

modules/storage/versions.tf

@@ -4,7 +4,7 @@ terraform {
   required_providers {
     google = {
       source  = "hashicorp/google"
-      version = "~> 6.45"
+      version = ">= 8.2.0, < 9.0.0"
     }
     random = {
       source  = "hashicorp/random"

modules/vpc/versions.tf

@@ -4,15 +4,7 @@ terraform {
   required_providers {
     google = {
       source  = "hashicorp/google"
-      version = "~> 6.45"
-    }
-    google-beta = {
-      source  = "hashicorp/google-beta"
-      version = "~> 6.45"
-    }
-    random = {
-      source  = "hashicorp/random"
-      version = ">= 3.7.2"
+      version = ">= 8.2.0, < 9.0.0"
     }
   }
 }

tests/sql_disk_options.tftest.hcl

@@ -0,0 +1,50 @@
+mock_provider "google" {}
+mock_provider "google-beta" {}
+mock_provider "random" {}
+
+variables {
+  deployment_name      = "sql-disk-test"
+  postgres_network     = "projects/test/global/networks/test"
+  postgres_kms_cmek_id = "projects/test/locations/us-central1/keyRings/test/cryptoKeys/test"
+}
+
+run "default_growth" {
+  command = plan
+  module { source = "./modules/database" }
+  assert {
+    condition     = google_sql_database_instance.braintrust.settings[0].disk_autoresize && google_sql_database_instance.braintrust.settings[0].disk_autoresize_limit == 0
+    error_message = "Automatic disk growth must remain enabled without a custom limit."
+  }
+}
+
+run "reject_small_disk" {
+  command = plan
+  module { source = "./modules/database" }
+  variables {
+    postgres_disk_size = 9
+  }
+  expect_failures = [var.postgres_disk_size]
+}
+
+run "n2_accepts_ten_gb_disk" {
+  command = plan
+  module { source = "./modules/database" }
+  variables { postgres_disk_size = 10 }
+  assert {
+    condition     = google_sql_database_instance.braintrust.settings[0].disk_size == 10
+    error_message = "N2 must retain support for a 10 GB SSD."
+  }
+}
+
+run "c4a_accepts_twenty_gb_disk" {
+  command = plan
+  module { source = "./modules/database" }
+  variables {
+    postgres_machine_type = "db-c4a-highmem-8"
+    postgres_disk_size    = 20
+  }
+  assert {
+    condition     = google_sql_database_instance.braintrust.settings[0].disk_size == 20
+    error_message = "C4A must accept the 20 GB Hyperdisk minimum."
+  }
+}