Define compute groups in terraform via mapping

Author: sjpb

environments/.stackhpc/terraform/main.tf

@@ -22,6 +22,7 @@ variable "cluster_name" {
 variable "os_version" {
   type = string
   description = "RL8 or RL9"
+  default = "RL9"
 }
 
 variable "cluster_image" {
@@ -58,6 +59,11 @@ variable "volume_backed_instances" {
     default = false
 }
 
+data "openstack_images_image_v2" "cluster" {
+    name = var.cluster_image[var.os_version]
+    most_recent = true
+}
+
 module "cluster" {
     source = "../../skeleton/{{cookiecutter.environment}}/terraform/"
 
@@ -66,33 +72,24 @@ module "cluster" {
     cluster_subnet = var.cluster_subnet
     vnic_type = var.vnic_type
     key_pair = "slurm-app-ci"
-    control_node = {
-        flavor: var.control_node_flavor
-        image: var.cluster_image[var.os_version]
-    }
+    cluster_image_id = data.openstack_images_image_v2.cluster.id
+    control_node_flavor = var.control_node_flavor
+
     login_nodes = {
-        login-0: {
-            flavor: var.other_node_flavor
-            image: var.cluster_image[var.os_version]
-        }
+        login-0: var.other_node_flavor
     }
-    compute_types = {
+    compute = {
         standard: { # NB: can't call this default!
+            nodes: ["compute-0", "compute-1"]
             flavor: var.other_node_flavor
-            image: var.cluster_image[var.os_version]
         }
         # Example of how to add another partition:
         # extra: {
+        #     nodes: ["compute-2", "compute-3"]
         #     flavor: var.other_node_flavor
-        #     image: var.cluster_image[var.os_version]
         # }
     }
-    compute_nodes = {
-        compute-0: "standard"
-        compute-1: "standard"
-        # compute-2: "extra"
-        # compute-3: "extra"
-    }
+    
     volume_backed_instances = var.volume_backed_instances
 
     environment_root = var.environment_root

environments/skeleton/{{cookiecutter.environment}}/terraform/.gitignore

@@ -0,0 +1,19 @@
+module "compute" {
+  source = "./compute"
+
+  for_each = var.compute
+
+  nodes = each.value.nodes
+  cluster_name = var.cluster_name
+  cluster_domain_suffix = var.cluster_domain_suffix
+  cluster_net_id = data.openstack_networking_network_v2.cluster_net.id
+  cluster_subnet_id = data.openstack_networking_subnet_v2.cluster_subnet.id
+
+  flavor = each.value.flavor
+  image_id = lookup(each.value, "image_id", var.cluster_image_id)
+  vnic_type = lookup(each.value, "vnic_type", var.vnic_type)
+  vnic_profile = lookup(each.value, "vnic_profile", var.vnic_profile)
+  key_pair = var.key_pair
+  environment_root = var.environment_root
+  security_group_ids = [for o in data.openstack_networking_secgroup_v2.nonlogin: o.id]
+}

environments/skeleton/{{cookiecutter.environment}}/terraform/compute.tf

@@ -0,0 +1,60 @@
+resource "openstack_networking_port_v2" "compute" {
+
+  for_each = toset(var.nodes)
+
+  name = "${var.cluster_name}-${each.key}"
+  network_id = var.cluster_net_id
+  admin_state_up = "true"
+
+  fixed_ip {
+    subnet_id = var.cluster_subnet_id
+  }
+
+  security_group_ids = var.security_group_ids
+
+  binding {
+    vnic_type = var.vnic_type
+    profile = var.vnic_profile
+  }
+}
+
+resource "openstack_compute_instance_v2" "compute" {
+
+  for_each = toset(var.nodes)
+  
+  name = "${var.cluster_name}-${each.key}"
+  image_id = var.image_id
+  flavor_name = var.flavor
+  key_pair = var.key_pair
+
+  dynamic "block_device" {
+    for_each = var.volume_backed_instances ? [1]: []
+    content {
+      uuid = var.image_id
+      source_type  = "image"
+      destination_type = "volume"
+      volume_size = var.root_volume_size
+      boot_index = 0
+      delete_on_termination = true
+    }
+  }
+  
+  network {
+    port = openstack_networking_port_v2.compute[each.key].id
+    access_network = true
+  }
+
+  metadata = {
+    environment_root = var.environment_root
+  }
+
+  user_data = <<-EOF
+    #cloud-config
+    fqdn: ${var.cluster_name}-${each.key}.${var.cluster_name}.${var.cluster_domain_suffix}
+  EOF
+
+}
+
+output "compute_instances" {
+    value = openstack_compute_instance_v2.compute
+}

environments/skeleton/{{cookiecutter.environment}}/terraform/compute/nodes.tf

@@ -0,0 +1,69 @@
+variable "nodes" {
+    type = list(string)
+    description = "list of node names for partition"
+}
+
+variable "flavor" {
+    type = string
+    description = "Name of flavor for partition"
+}
+
+variable "cluster_name" {
+    type = string
+}
+
+variable "cluster_domain_suffix" {
+    type = string
+    default = "invalid"
+}
+
+variable "cluster_net_id" {
+    type = string
+}
+
+variable "cluster_subnet_id" {
+    type = string
+}
+
+variable "key_pair" {
+    type = string
+    description = "Name of an existing keypair in OpenStack"
+}
+
+variable "image_id" {
+    type = string
+    description = "ID of image for the partition"
+}
+
+variable "environment_root" {
+    type = string
+    description = "Path to environment root, automatically set by activate script"
+}
+
+variable "vnic_type" {
+    type = string
+    description = "VNIC type, see https://registry.terraform.io/providers/terraform-provider-openstack/openstack/latest/docs/resources/networking_port_v2#vnic_type"
+    default = "normal"
+}
+
+variable "vnic_profile" {
+    type = string
+    description = "VNIC binding profile as json string, see https://registry.terraform.io/providers/terraform-provider-openstack/openstack/latest/docs/resources/networking_port_v2#profile."
+    default = "{}"
+}
+
+variable "volume_backed_instances" {
+    description = "Whether to use volumes for root disks"
+    type = bool
+    default = false
+}
+
+variable "root_volume_size" {
+    description = "Size of volume for root volumes if using volume backed instances, in Gb"
+    type = number
+    default = 40
+}
+
+variable "security_group_ids" {
+    type = list
+}

environments/skeleton/{{cookiecutter.environment}}/terraform/compute/variables.tf

@@ -5,19 +5,16 @@ resource "local_file" "hosts" {
                             "cluster_domain_suffix": var.cluster_domain_suffix,
                             "control_instances": openstack_compute_instance_v2.control
                             "login_instances": openstack_compute_instance_v2.login
-                            "compute_instances": openstack_compute_instance_v2.compute
-                            "state_dir": var.state_dir,
-                            "compute_types": var.compute_types,
-                            "compute_nodes": var.compute_nodes,
+                            "compute_groups": module.compute
                           },
                           )
-  filename = "../inventory/hosts"
+  filename = "../inventory/hosts.yml"
 }
 
 resource "local_file" "partitions" {
     content  = templatefile("${path.module}/partitions.tpl",
                             {
-                              "compute_types": var.compute_types,
+                              "compute_groups": module.compute,
                             },
     )
     filename = "../inventory/group_vars/all/partitions.yml" # as all/ is created by skeleton

environments/skeleton/{{cookiecutter.environment}}/terraform/inventory.tf

@@ -1,32 +1,36 @@
-[all:vars]
-openhpc_cluster_name=${cluster_name}
-cluster_domain_suffix=${cluster_domain_suffix}
+all:
+    vars:
+        openhpc_cluster_name: ${cluster_name}
+        cluster_domain_suffix: ${cluster_domain_suffix}
 
-[control]
+control:
+    hosts:
 %{ for control in control_instances ~}
-${ control.name } ansible_host=${[for n in control.network: n.fixed_ip_v4 if n.access_network][0]} node_fqdn=${ control.name }.${cluster_name}.${cluster_domain_suffix}
+        ${ control.name }:
+            ansible_host: ${[for n in control.network: n.fixed_ip_v4 if n.access_network][0]}
+            instance_id: ${ control.id }
 %{ endfor ~}
 
-[control:vars]
-# NB needs to be set on group not host otherwise it is ignored in packer build!
-appliances_state_dir=${state_dir}
-
-[login]
+login:
+    hosts:
 %{ for login in login_instances ~}
-${ login.name } ansible_host=${[for n in login.network: n.fixed_ip_v4 if n.access_network][0]} node_fqdn=${ login.name }.${cluster_name}.${cluster_domain_suffix}
+        ${ login.name }:
+            ansible_host: ${[for n in login.network: n.fixed_ip_v4 if n.access_network][0]}
+            instance_id: ${ login.id }
 %{ endfor ~}
 
-[compute]
-%{ for compute in compute_instances ~}
-${ compute.name } ansible_host=${[for n in compute.network: n.fixed_ip_v4 if n.access_network][0]} node_fqdn=${ compute.name }.${cluster_name}.${cluster_domain_suffix}
+%{ for group_name in keys(compute_groups) ~}
+${cluster_name}_${group_name}:
+    hosts:
+%{ for node in compute_groups[group_name]["compute_instances"] ~}
+        ${ node.name }:
+            ansible_host: ${node.access_ip_v4}
+            instance_id: ${ node.id }
+%{ endfor ~}
 %{ endfor ~}
 
-# Define groups for slurm parititions:
-%{~ for type_name, type_descr in compute_types}
-[${cluster_name}_${type_name}]
-    %{~ for node_name, node_type in compute_nodes ~}
-        %{~ if node_type == type_name ~}
-${ compute_instances[node_name].name }
-        %{~ endif ~}
-%{~ endfor ~}
+compute:
+    children:
+%{ for group_name in keys(compute_groups) ~}
+        ${cluster_name}_${group_name}:
 %{ endfor ~}