terraform: Terraform module for AWS (#2503)

terraform/infrastructure/aws/modules/instance_group/main.tf

@@ -0,0 +1,115 @@
+terraform {
+  required_providers {
+    aws = {
+      source  = "hashicorp/aws"
+      version = "5.17.0"
+    }
+    random = {
+      source  = "hashicorp/random"
+      version = "3.5.1"
+    }
+  }
+}
+
+locals {
+  group_uid = random_id.uid.hex
+  name      = "${var.base_name}-${lower(var.role)}-${local.group_uid}"
+}
+
+resource "random_id" "uid" {
+  byte_length = 4
+}
+
+resource "aws_launch_template" "launch_template" {
+  name_prefix   = local.name
+  image_id      = var.image_id
+  instance_type = var.instance_type
+  iam_instance_profile {
+    name = var.iam_instance_profile
+  }
+  vpc_security_group_ids = var.security_groups
+  metadata_options {
+    http_endpoint               = "enabled"
+    http_tokens                 = "required"
+    instance_metadata_tags      = "disabled"
+    http_put_response_hop_limit = 2
+  }
+
+  block_device_mappings {
+    device_name = "/dev/sdb"
+    ebs {
+      volume_size           = var.state_disk_size
+      volume_type           = var.state_disk_type
+      encrypted             = true
+      delete_on_termination = true
+    }
+  }
+
+  # See: https://registry.terraform.io/providers/hashicorp/aws/latest/docs/resources/launch_template#cpu-options
+  cpu_options {
+    # use "enabled" to enable SEV-SNP
+    # use "disabled" to disable SEV-SNP (but still require SNP-capable hardware)
+    # use null to leave the setting unset (allows non-SNP-capable hardware to be used)
+    amd_sev_snp = var.enable_snp ? "enabled" : null
+    # Disable SMT. We are already disabling it inside the image.
+    # Disabling SMT only in the image, not in the Hypervisor creates problems.
+    # Thus, also disable it in the Hypervisor.
+    # TODO(derpsteb): reenable once AWS confirms it's safe to do so.
+    # threads_per_core = 1
+    # When setting threads_per_core we also have to set core_count.
+    # For the currently supported SNP instance families (C6a, M6a, R6a) default_cores
+    # equals the maximum number of available cores.
+    # core_count = data.aws_ec2_instance_type.instance_data.default_cores
+  }
+
+  lifecycle {
+    create_before_destroy = true
+    ignore_changes = [
+      cpu_options,     # required. we cannot change the CPU options of a launch template
+      name_prefix,     # required. Allow legacy scale sets to keep their old names
+      default_version, # required. update procedure creates new versions of the launch template
+      image_id,        # required. update procedure modifies the image id externally
+    ]
+  }
+}
+
+resource "aws_autoscaling_group" "autoscaling_group" {
+  name = local.name
+  launch_template {
+    id = aws_launch_template.launch_template.id
+  }
+  min_size            = 1
+  max_size            = 10
+  desired_capacity    = var.initial_count
+  vpc_zone_identifier = [var.subnetwork]
+  target_group_arns   = var.target_group_arns
+
+  # TODO(msanft): Remove this (to have the 10m default) once AWS SEV-SNP boot problems are resolved.
+  # Set a higher timeout for the ASG to fulfill the desired healthy capcity. Temporary workaround to
+  # long boot times on SEV-SNP machines on AWS.
+  wait_for_capacity_timeout = var.enable_snp ? "20m" : "10m"
+
+  dynamic "tag" {
+    for_each = var.tags
+    content {
+      key                 = tag.key
+      value               = tag.value
+      propagate_at_launch = true
+    }
+  }
+
+  lifecycle {
+    create_before_destroy = true
+    ignore_changes = [
+      name,                      # required. Allow legacy scale sets to keep their old names
+      launch_template.0.version, # required. update procedure creates new versions of the launch template
+      min_size,                  # required. autoscaling modifies the instance count externally
+      max_size,                  # required. autoscaling modifies the instance count externally
+      desired_capacity,          # required. autoscaling modifies the instance count externally
+    ]
+  }
+}
+
+data "aws_ec2_instance_type" "instance_data" {
+  instance_type = var.instance_type
+}