Support and test "re-imageable" compute nodes via compute node metadata

.github/workflows/stackhpc.yml

@@ -170,33 +170,22 @@ jobs:
         env:
           DEMO_USER_PASSWORD: ${{ secrets.TEST_USER_PASSWORD }}
 
-      # - name: Build environment-specific compute image
-      #   id: packer_build
-      #   run: |
-      #     . venv/bin/activate
-      #     . environments/.stackhpc/activate
-      #     cd packer/
-      #     packer init
-      #     PACKER_LOG=1 packer build -except openstack.fatimage -on-error=ask -var-file=$PKR_VAR_environment_root/builder.pkrvars.hcl openstack.pkr.hcl
-      #     ../dev/output_manifest.py packer-manifest.json # Sets NEW_COMPUTE_IMAGE_ID outputs
-
-      # - name: Test reimage of compute nodes to new environment-specific image (via slurm)
-      #   run: |
-      #     . venv/bin/activate
-      #     . environments/.stackhpc/activate
-      #     ansible login -v -a "sudo scontrol reboot ASAP nextstate=RESUME reason='rebuild image:${{ steps.packer_build.outputs.NEW_COMPUTE_IMAGE_ID }}' ${TF_VAR_cluster_name}-compute-[0-3]"
-      #     ansible compute -m wait_for_connection -a 'delay=60 timeout=600' # delay allows node to go down
-      #     ansible-playbook -v ansible/ci/check_slurm.yml
-
       - name: Test reimage of login and control nodes (via rebuild adhoc)
         run: |
           . venv/bin/activate
           . environments/.stackhpc/activate
           ansible-playbook -v --limit control,login ansible/adhoc/rebuild.yml
-          ansible all -m wait_for_connection -a 'delay=60 timeout=600' # delay allows node to go down
           ansible-playbook -v ansible/site.yml
           ansible-playbook -v ansible/ci/check_slurm.yml
 
+      - name: Test reimage of compute nodes and compute-init (via rebuild adhoc)
+        run: |
+          . venv/bin/activate
+          . environments/.stackhpc/activate
+          ansible-playbook -v --limit compute ansible/adhoc/rebuild.yml
+          ansible all -m wait_for_connection -a 'delay=60 timeout=600' 
+          ansible-playbook -v ansible/ci/check_slurm.yml
+
       - name: Check sacct state survived reimage
         run: |
           . venv/bin/activate

ansible/roles/compute_init/README.md

@@ -42,10 +42,13 @@ The following roles/groups are currently fully functional:
   node and all compute nodes.
 - `openhpc`: all functionality
 
-# Development/debugging
+The above may be enabled by setting the compute_init_enable property on the
+terraform compute variable.
 
-To develop/debug this without actually having to build an image:
+# Development/debugging
 
+To develop/debug changes to the compute script without actually having to build
+a new image:
 
 1. Deploy a cluster using tofu and ansible/site.yml as normal. This will
    additionally configure the control node to export compute hostvars over NFS.
@@ -103,7 +106,7 @@ as in step 3.
   available v the current approach:
 
     ```
-    [root@rl9-compute-0 rocky]# grep hostvars /mnt/cluster/hostvars/rl9-compute-0/hostvars.yml 
+    [root@rl9-compute-0 rocky]# grep hostvars /mnt/cluster/hostvars/rl9-compute-0/hostvars.yml
         "grafana_address": "{{ hostvars[groups['grafana'].0].api_address }}",
         "grafana_api_address": "{{ hostvars[groups['grafana'].0].internal_address }}",
         "mysql_host": "{{ hostvars[groups['mysql'] | first].api_address }}",

docs/experimental/compute-init.md

@@ -2,112 +2,17 @@
 
 See the role README.md
 
-# Results/progress
+# CI workflow
 
-Without any metadata:
+The compute node rebuild is tested in CI after the tests for rebuilding the
+login and control nodes. The process follows
 
-    [root@rl9-compute-0 rocky]# systemctl status ansible-init
-    ● ansible-init.service
-        Loaded: loaded (/etc/systemd/system/ansible-init.service; enabled; preset: disabled)
-        Active: activating (start) since Fri 2024-12-13 20:41:16 UTC; 1min 45s ago
-    Main PID: 16089 (ansible-init)
-        Tasks: 8 (limit: 10912)
-        Memory: 99.5M
-            CPU: 11.687s
-        CGroup: /system.slice/ansible-init.service
-                ├─16089 /usr/lib/ansible-init/bin/python /usr/bin/ansible-init
-                ├─16273 /usr/lib/ansible-init/bin/python3.9 /usr/lib/ansible-init/bin/ansible-playbook --connection local --inventory 127.0.0.1, /etc/ansible-init/playbooks/1-compute-init.yml
-                ├─16350 /usr/lib/ansible-init/bin/python3.9 /usr/lib/ansible-init/bin/ansible-playbook --connection local --inventory 127.0.0.1, /etc/ansible-init/playbooks/1-compute-init.yml
-                ├─16361 /bin/sh -c "/usr/bin/python3 /root/.ansible/tmp/ansible-tmp-1734122485.9542894-16350-45936546411977/AnsiballZ_mount.py && sleep 0"
-                ├─16362 /usr/bin/python3 /root/.ansible/tmp/ansible-tmp-1734122485.9542894-16350-45936546411977/AnsiballZ_mount.py
-                ├─16363 /usr/bin/mount /mnt/cluster
-                └─16364 /sbin/mount.nfs 192.168.10.12:/exports/cluster /mnt/cluster -o ro,sync
+1. Compute nodes are reimaged:
 
-    Dec 13 20:41:24 rl9-compute-0.rl9.invalid ansible-init[16273]: ok: [127.0.0.1]
-    Dec 13 20:41:24 rl9-compute-0.rl9.invalid ansible-init[16273]: TASK [Report skipping initialization if not compute node] **********************
-    Dec 13 20:41:25 rl9-compute-0.rl9.invalid ansible-init[16273]: skipping: [127.0.0.1]
-    Dec 13 20:41:25 rl9-compute-0.rl9.invalid ansible-init[16273]: TASK [meta] ********************************************************************
-    Dec 13 20:41:25 rl9-compute-0.rl9.invalid ansible-init[16273]: skipping: [127.0.0.1]
-    Dec 13 20:41:25 rl9-compute-0.rl9.invalid ansible-init[16273]: TASK [Ensure the mount directory exists] ***************************************
-    Dec 13 20:41:25 rl9-compute-0.rl9.invalid python3[16346]: ansible-file Invoked with path=/mnt/cluster state=directory owner=root group=root mode=u=rwX,go= recurse=False force=False follow=True modification_time_format=%Y%m%d%H%M.%S access>
-    Dec 13 20:41:25 rl9-compute-0.rl9.invalid ansible-init[16273]: changed: [127.0.0.1]
-    Dec 13 20:41:25 rl9-compute-0.rl9.invalid ansible-init[16273]: TASK [Mount /mnt/cluster] ******************************************************
-    Dec 13 20:41:26 rl9-compute-0.rl9.invalid python3[16362]: ansible-mount Invoked with path=/mnt/cluster src=192.168.10.12:/exports/cluster fstype=nfs opts=ro,sync state=mounted boot=True dump=0 passno=0 backup=False fstab=None
-    [root@rl9-compute-0 rocky]# systemctl status ansible-init
+         ansible-playbook -v --limit compute ansible/adhoc/rebuild.yml
 
-Added metadata via horizon: 
+2. Ansible-init runs against newly reimaged compute nodes
 
-    compute_groups ["compute"]
+3. Run sinfo and check nodes have expected slurm state
 
-
-OK:
-
-    [root@rl9-compute-0 rocky]# systemctl status ansible-init
-    ● ansible-init.service
-        Loaded: loaded (/etc/systemd/system/ansible-init.service; enabled; preset: disabled)
-        Active: active (exited) since Fri 2024-12-13 20:43:31 UTC; 33s ago
-        Process: 16089 ExecStart=/usr/bin/ansible-init (code=exited, status=0/SUCCESS)
-    Main PID: 16089 (code=exited, status=0/SUCCESS)
-            CPU: 13.003s
-
-    Dec 13 20:43:31 rl9-compute-0.rl9.invalid ansible-init[16273]: ok: [127.0.0.1] => {
-    Dec 13 20:43:31 rl9-compute-0.rl9.invalid ansible-init[16273]:     "msg": "Skipping compute initialization as cannot mount exports/cluster share"
-    Dec 13 20:43:31 rl9-compute-0.rl9.invalid ansible-init[16273]: }
-    Dec 13 20:43:31 rl9-compute-0.rl9.invalid ansible-init[16273]: TASK [meta] ********************************************************************
-    Dec 13 20:43:31 rl9-compute-0.rl9.invalid ansible-init[16273]: PLAY RECAP *********************************************************************
-    Dec 13 20:43:31 rl9-compute-0.rl9.invalid ansible-init[16273]: 127.0.0.1                  : ok=4    changed=1    unreachable=0    failed=0    skipped=1    rescued=0    ignored=1
-    Dec 13 20:43:31 rl9-compute-0.rl9.invalid ansible-init[16089]: [INFO] executing remote playbooks for stage - post
-    Dec 13 20:43:31 rl9-compute-0.rl9.invalid ansible-init[16089]: [INFO] writing sentinel file /var/lib/ansible-init.done
-    Dec 13 20:43:31 rl9-compute-0.rl9.invalid ansible-init[16089]: [INFO] ansible-init completed successfully
-    Dec 13 20:43:31 rl9-compute-0.rl9.invalid systemd[1]: Finished ansible-init.service.
-
-Now run site.yml, then restart ansible-init again:
-
-
-    [root@rl9-compute-0 rocky]# systemctl status ansible-init
-    ● ansible-init.service
-        Loaded: loaded (/etc/systemd/system/ansible-init.service; enabled; preset: disabled)
-        Active: active (exited) since Fri 2024-12-13 20:50:10 UTC; 11s ago
-        Process: 18921 ExecStart=/usr/bin/ansible-init (code=exited, status=0/SUCCESS)
-    Main PID: 18921 (code=exited, status=0/SUCCESS)
-            CPU: 8.240s
-
-    Dec 13 20:50:10 rl9-compute-0.rl9.invalid ansible-init[19110]: TASK [Report skipping initialization if cannot mount nfs] **********************
-    Dec 13 20:50:10 rl9-compute-0.rl9.invalid ansible-init[19110]: skipping: [127.0.0.1]
-    Dec 13 20:50:10 rl9-compute-0.rl9.invalid ansible-init[19110]: TASK [meta] ********************************************************************
-    Dec 13 20:50:10 rl9-compute-0.rl9.invalid ansible-init[19110]: skipping: [127.0.0.1]
-    Dec 13 20:50:10 rl9-compute-0.rl9.invalid ansible-init[19110]: PLAY RECAP *********************************************************************
-    Dec 13 20:50:10 rl9-compute-0.rl9.invalid ansible-init[19110]: 127.0.0.1                  : ok=3    changed=1    unreachable=0    failed=0    skipped=2    rescued=0    ignored=0
-    Dec 13 20:50:10 rl9-compute-0.rl9.invalid ansible-init[18921]: [INFO] executing remote playbooks for stage - post
-    Dec 13 20:50:10 rl9-compute-0.rl9.invalid ansible-init[18921]: [INFO] writing sentinel file /var/lib/ansible-init.done
-    Dec 13 20:50:10 rl9-compute-0.rl9.invalid ansible-init[18921]: [INFO] ansible-init completed successfully
-    Dec 13 20:50:10 rl9-compute-0.rl9.invalid systemd[1]: Finished ansible-init.service.
-    [root@rl9-compute-0 rocky]# ls /mnt/cluster/host
-    hosts     hostvars/ 
-    [root@rl9-compute-0 rocky]# ls /mnt/cluster/hostvars/rl9-compute-
-    rl9-compute-0/ rl9-compute-1/ 
-    [root@rl9-compute-0 rocky]# ls /mnt/cluster/hostvars/rl9-compute-
-    rl9-compute-0/ rl9-compute-1/ 
-    [root@rl9-compute-0 rocky]# ls /mnt/cluster/hostvars/rl9-compute-0/
-    hostvars.yml
-
-This commit - shows that hostvars have loaded:
-
-    [root@rl9-compute-0 rocky]# systemctl status ansible-init
-    ● ansible-init.service
-        Loaded: loaded (/etc/systemd/system/ansible-init.service; enabled; preset: disabled)
-        Active: active (exited) since Fri 2024-12-13 21:06:20 UTC; 5s ago
-        Process: 27585 ExecStart=/usr/bin/ansible-init (code=exited, status=0/SUCCESS)
-    Main PID: 27585 (code=exited, status=0/SUCCESS)
-            CPU: 8.161s
-
-    Dec 13 21:06:20 rl9-compute-0.rl9.invalid ansible-init[27769]: TASK [Demonstrate hostvars have loaded] ****************************************
-    Dec 13 21:06:20 rl9-compute-0.rl9.invalid ansible-init[27769]: ok: [127.0.0.1] => {
-    Dec 13 21:06:20 rl9-compute-0.rl9.invalid ansible-init[27769]:     "prometheus_version": "2.27.0"
-    Dec 13 21:06:20 rl9-compute-0.rl9.invalid ansible-init[27769]: }
-    Dec 13 21:06:20 rl9-compute-0.rl9.invalid ansible-init[27769]: PLAY RECAP *********************************************************************
-    Dec 13 21:06:20 rl9-compute-0.rl9.invalid ansible-init[27769]: 127.0.0.1                  : ok=5    changed=0    unreachable=0    failed=0    skipped=2    rescued=0    ignored=0
-    Dec 13 21:06:20 rl9-compute-0.rl9.invalid ansible-init[27585]: [INFO] executing remote playbooks for stage - post
-    Dec 13 21:06:20 rl9-compute-0.rl9.invalid ansible-init[27585]: [INFO] writing sentinel file /var/lib/ansible-init.done
-    Dec 13 21:06:20 rl9-compute-0.rl9.invalid ansible-init[27585]: [INFO] ansible-init completed successfully
-    Dec 13 21:06:20 rl9-compute-0.rl9.invalid systemd[1]: Finished ansible-init.service.
+         ansible-playbook -v ansible/ci/check_slurm.yml

environments/.stackhpc/terraform/main.tf

@@ -82,6 +82,7 @@ module "cluster" {
         standard: { # NB: can't call this default!
             nodes: ["compute-0", "compute-1"]
             flavor: var.other_node_flavor
+            compute_init_enable: ["compute", "etc_hosts", "nfs", "basic_users", "eessi"]
         }
         # Example of how to add another partition:
         # extra: {

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

@@ -18,4 +18,6 @@ module "compute" {
   k3s_token = var.k3s_token
   control_address = [for n in openstack_compute_instance_v2.control["control"].network: n.fixed_ip_v4 if n.access_network][0]
   security_group_ids = [for o in data.openstack_networking_secgroup_v2.nonlogin: o.id]
+
+  compute_init_enable = each.value.compute_init_enable
 }

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

@@ -44,11 +44,14 @@ resource "openstack_compute_instance_v2" "compute" {
     access_network = true
   }
 
-  metadata = {
-    environment_root = var.environment_root
-    k3s_token = var.k3s_token
-    control_address = var.control_address
-  }
+  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

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

@@ -76,3 +76,9 @@ variable "control_address" {
     description = "Name/address of control node"
     type = string
 }
+
+variable "compute_init_enable" {
+    type = list(string)
+    description = "Groups to activate for ansible-init compute rebuilds"
+    default = []
+}

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

@@ -52,6 +52,7 @@ variable "compute" {
             image_id: Overrides variable cluster_image_id
             vnic_type: Overrides variable vnic_type
             vnic_profile: Overrides variable vnic_profile
+            compute_init_enable: Toggles compute-init rebuild (see compute-init role docs)
     EOF
 }
 
@@ -135,4 +136,4 @@ variable "root_volume_size" {
 variable "k3s_token" {
     description = "K3s cluster authentication token, set automatically by Ansible"
     type = string
-}
+}