Optionally ignore image changes in TF (#545)

* add lifecycle ignore changes to skeleton cookiecutter config

* add docs to experimental/compute-init

* refine docs

* account for transitory down state in ci/check_slurm.yml

* Update environments/skeleton/{{cookiecutter.environment}}/tofu/variables.tf

Co-authored-by: Steve Brasier <[email protected]>

---------

Co-authored-by: Steve Brasier <[email protected]>

Author: bertiethorpe

ansible/ci/check_slurm.yml

@@ -6,7 +6,7 @@
       shell: 'sinfo --noheader --format="%N %P %a %l %D %t" | sort' # using --format ensures we control whitespace: Partition,partition_state,max_jobtime,num_nodes,node_state,node_name
       register: sinfo
       changed_when: false
-      until: not ("boot" in sinfo.stdout or "idle*" in sinfo.stdout)
+      until: not ("boot" in sinfo.stdout or "idle*" in sinfo.stdout or "down" in sinfo.stdout)
       retries: 10
       delay: 5
     - name: Check nodes have expected slurm state

docs/experimental/compute-init.md

@@ -2,6 +2,18 @@
 
 See the role README.md
 
+# Changes to image / tofu state
+
+When a compute group has the `ignore_image_changes` parameter set to true,
+changes to the `image_id` parameter (which defaults to `cluster_image_id`) are
+ignored by OpenTofu.
+
+Regardless of whether `ignore_image_changes` is set, OpenTofu templates out the
+`image_id` into the Ansible inventory for each compute node. The `compute_init`
+role templates out hostvars to the control node, which means the "target" image
+ID is then available on the control node. Subsequent work will use this to
+rebuild the node via slurm.
+
 # CI workflow
 
 The compute node rebuild is tested in CI after the tests for rebuilding the

environments/.stackhpc/tofu/main.tf

@@ -81,6 +81,7 @@ module "cluster" {
             nodes: ["compute-0", "compute-1"]
             flavor: var.other_node_flavor
             compute_init_enable: ["compute", "etc_hosts", "nfs", "basic_users", "eessi"]
+            # ignore_image_changes: true
         }
         # Example of how to add another partition:
         # extra: {

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

@@ -21,6 +21,7 @@ module "compute" {
   extra_volumes = lookup(each.value, "extra_volumes", {})
 
   compute_init_enable = lookup(each.value, "compute_init_enable", [])
+  ignore_image_changes = lookup(each.value, "ignore_image_changes", false)
 
   key_pair = var.key_pair
   environment_root = var.environment_root

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

@@ -9,6 +9,9 @@ locals {
   # this is a mapping with
   # keys "compute-0-vol-a", "compute-0-vol-b" ...
   # values which are a mapping e.g. {"node"="compute-0", "volume"="vol-a"}
+
+  # Workaround for lifecycle meta-argument only taking static values
+  compute_instances = var.ignore_image_changes ? openstack_compute_instance_v2.compute_fixed_image : openstack_compute_instance_v2.compute
 }
 
 resource "openstack_blockstorage_volume_v3" "compute" {
@@ -24,7 +27,7 @@ resource "openstack_compute_volume_attach_v2" "compute" {
 
   for_each = local.all_compute_volumes
 
-  instance_id = openstack_compute_instance_v2.compute["${each.value.node}"].id
+  instance_id = local.compute_instances["${each.value.node}"].id
   volume_id  = openstack_blockstorage_volume_v3.compute["${each.key}"].id
 }
 
@@ -48,9 +51,57 @@ resource "openstack_networking_port_v2" "compute" {
   }
 }
 
+resource "openstack_compute_instance_v2" "compute_fixed_image" {
+
+  for_each = var.ignore_image_changes ? 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 = merge(
+    {
+        environment_root = var.environment_root
+        k3s_token          = var.k3s_token
+        control_address    = var.control_address
+    },
+    {for e in var.compute_init_enable: e => true}
+  )
+
+  user_data = <<-EOF
+    #cloud-config
+    fqdn: ${var.cluster_name}-${each.key}.${var.cluster_name}.${var.cluster_domain_suffix}
+  EOF
+
+  lifecycle {
+    ignore_changes = [
+      image_id,
+    ]
+  }
+
+}
+
 resource "openstack_compute_instance_v2" "compute" {
 
-  for_each = toset(var.nodes)
+  for_each = var.ignore_image_changes ? [] : toset(var.nodes)
 
   name = "${var.cluster_name}-${each.key}"
   image_id = var.image_id
@@ -91,5 +142,5 @@ resource "openstack_compute_instance_v2" "compute" {
 }
 
 output "compute_instances" {
-    value = openstack_compute_instance_v2.compute
+    value = local.compute_instances
 }

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

@@ -93,4 +93,10 @@ variable "compute_init_enable" {
     type = list(string)
     description = "Groups to activate for ansible-init compute rebuilds"
     default = []
+}
+
+variable "ignore_image_changes" {
+    type = bool
+    description = "Whether to ignore changes to the image_id parameter"
+    default = false
 }

environments/skeleton/{{cookiecutter.environment}}/tofu/inventory.tpl

@@ -28,6 +28,7 @@ ${cluster_name}_${group_name}:
         ${ node.name }:
             ansible_host: ${node.access_ip_v4}
             instance_id: ${ node.id }
+            image_id: ${ node.image_id }
 %{ endfor ~}
 %{ endfor ~}
 

environments/skeleton/{{cookiecutter.environment}}/tofu/variables.tf

@@ -53,6 +53,7 @@ variable "compute" {
             vnic_type: Overrides variable vnic_type
             vnic_profile: Overrides variable vnic_profile
             compute_init_enable: Toggles compute-init rebuild (see compute-init role docs)
+            ignore_image_changes: Ignore changes to the image_id parameter (see docs/experimental/compute-init.md)
             volume_backed_instances: Overrides variable volume_backed_instances
             root_volume_size: Overrides variable root_volume_size
             extra_volumes: Mapping defining additional volumes to create and attach