Add support for Node Health Checks (#654)

sjpb · web-flow · commit eabf59b33650 · 2025-06-20T15:51:03.000+01:00
* add nhc * change nhc configuration from replace to lineinfile * move things around * revert doca move * revert slurm changes * tweak NHC docs * remove debugging stop * rename final playbook and let dnf repos work for post-hook in both site and fatimage * note returntoservice bug * revert slurm playbook changes * revert ordering change for repo disable - see issue 708 * bump CI image * fix exporting NHC config * rememeber compute-init config is synced to /var/tmp * enable NHC during rebuild for stackhpc * fix path to shared compute-init files for NHC * don't write compute-init share into fstab for reliability -is unmounted once synced * bump CI image * change NHC to use templating instead of autoconfiguration * fix nhc task file from compute-init * bump CI image * fix NHC configuration directory * bump CI image * fix NHC nodename/hostname mismatch * remove un-needed nhc conf dir tasks * bump CI image * Revert "bump CI image" 4def7a5 This reverts commit 4def7a5. * Revert "remove un-needed nhc conf dir tasks" c38f9fd This reverts commit c38f9fd. * fix nhc mid-upgrade * bump CI image
diff --git a/ansible/.gitignore b/ansible/.gitignore
@@ -90,3 +90,5 @@ roles/*
 !roles/gateway/**
 !roles/alertmanager/
 !roles/alertmanager/**
+!roles/nhc/
+!roles/nhc/**
diff --git a/ansible/bootstrap.yml b/ansible/bootstrap.yml
@@ -134,6 +134,7 @@
 
 - hosts: dnf_repos
   become: yes
+  tags: dnf_repos
   tasks:
   - name: Check that creds won't be leaked to users
     ansible.builtin.assert:
diff --git a/ansible/disable-repos.yml b/ansible/disable-repos.yml
diff --git a/ansible/extras.yml b/ansible/extras.yml
@@ -58,17 +58,6 @@
     - import_role:
         name: persist_hostkeys
 
-
-- name: Setup NFS export for compute node configuration
-  hosts: compute_init:!builder
-  # NB: has to be after eeesi and os-manila-mount
-  tags: compute_init
-  become: yes
-  tasks:
-    - include_role:
-        name: compute_init
-        tasks_from: export.yml
-
 - name: Install k9s
   become: yes
   hosts: k9s
diff --git a/ansible/fatimage.yml b/ansible/fatimage.yml
@@ -257,7 +257,7 @@
   import_playbook: "{{ hook_path if hook_path | exists else 'noop.yml' }}"
   when: hook_path | exists
 
-- import_playbook: disable-repos.yml
+- import_playbook: final.yml
 
 - hosts: builder
   become: yes
diff --git a/ansible/final.yml b/ansible/final.yml
@@ -0,0 +1,18 @@
+- hosts: dnf_repos
+  become: yes
+  tags: dnf_repos
+  tasks:
+    - name: Disable pulp repos
+      ansible.builtin.include_role:
+        name: dnf_repos
+        tasks_from: disable_repos.yml
+
+- name: Setup NFS export for compute_init
+  hosts: compute_init:!builder
+  # NB: done last so other roles can prepare configuration etc
+  tags: compute_init
+  become: yes
+  tasks:
+    - include_role:
+        name: compute_init
+        tasks_from: export.yml
diff --git a/ansible/roles/compute_init/README.md b/ansible/roles/compute_init/README.md
@@ -84,6 +84,7 @@ it also requires an image build with the role name added to the
 | slurm.yml                | openhpc [10]            | All slurmd functionality        | No                  |
 | slurm.yml                | (set memory limits)     | Fully supported                 | No                  |
 | slurm.yml                | (block ssh)             | Fully supported                 | No                  |
+| slurm.yml                | nhc                     | Fully supported                 | No                  |
 | portal.yml               | (openondemand server)   | Not relevant for compute nodes  | n/a                 |
 | portal.yml               | (openondemand vnc desktop)  | None required - use image build | No              |
 | portal.yml               | (openondemand jupyter server) | None required - use image build | No            |
diff --git a/ansible/roles/compute_init/files/compute-init.yml b/ansible/roles/compute_init/files/compute-init.yml
@@ -19,6 +19,7 @@
     enable_basic_users: "{{ os_metadata.meta.basic_users | default(false) | bool }}"
     enable_eessi: "{{ os_metadata.meta.eessi | default(false) | bool }}"
     enable_chrony: "{{ os_metadata.meta.chrony | default(false) | bool }}"
+    enable_nhc: "{{ os_metadata.meta.nhc | default(false) | bool }}"
 
     # TODO: "= role defaults" - could be moved to a vars_file: on play with similar precedence effects
     resolv_conf_nameservers: []
@@ -63,12 +64,12 @@
         mode: u=rX,g=rwX,o=
 
     - name: Mount /mnt/cluster
-      mount:
+      ansible.posix.mount:
         path: /mnt/cluster
         src: "{{ server_node_ip }}:/exports/cluster"
         fstype: nfs
         opts: ro,sync
-        state: mounted
+        state: ephemeral # will be unmounted after sync, don't want it in fstab
       register: _mount_mnt_cluster
       ignore_errors: true
       # exits from playbook if this failed below, allowing ansible-init to
@@ -350,6 +351,11 @@
         enabled: true
         state: started
 
+    - name: Provide NHC configuration
+      ansible.builtin.include_role:
+        name: nhc
+        tasks_from: boot.yml
+      when: enable_nhc
 
     - name: Ensure node is resumed
       # TODO: consider if this is always safe for all job states?
diff --git a/ansible/roles/compute_init/tasks/export.yml b/ansible/roles/compute_init/tasks/export.yml
@@ -98,3 +98,9 @@
     name: sshd
     tasks_from: export.yml
   when: "'sshd' in group_names"
+
+- name: Export generated NHC config
+  import_role:
+    name: nhc
+    tasks_from: export.yml
+  when: "'nhc' in group_names"
diff --git a/ansible/roles/compute_init/tasks/install.yml b/ansible/roles/compute_init/tasks/install.yml
@@ -49,6 +49,8 @@
       dest: roles/
     - src: ../../lustre
       dest: roles/
+    - src: ../../nhc
+      dest: roles/
 
 - name: Add filter_plugins to ansible.cfg
   lineinfile:
diff --git a/ansible/roles/nhc/README.md b/ansible/roles/nhc/README.md
@@ -0,0 +1,66 @@
+# Node Health Checks (nhc)
+
+Deploys and configures the LBNL [Node Health Check](https://github.com/mej/nhc)
+(NHC) which will put nodes in `DOWN` state if they fail periodic checks on
+various aspects.
+
+Due to the integration with Slurm this is tightly linked to the configuration
+for the [stackhpc.openhpc](../stackhpc.openhpc/README.md) role.
+
+## Enabling
+
+By [default](../../../environments/common/inventory/group_vars/all/openhpc.yml)
+the required `nhc-ohpc` packages are installed in all images.
+
+To enable node health checks, ensure the `nhc` group contains the `compute` group:
+
+```yaml
+# environments/site/inventory/groups:
+[nhc:children]
+# Hosts to configure for node health checks
+compute
+```
+
+When the `anisble/site.yml` playbook is run this will automatically:
+1. Add NHC-related configuration to the `slurm.conf` Slurm configuration file.
+   The default configuration is defined in `openhpc_config_nhc`
+   (see [environments/common/inventory/group_vars/all/openhpc.yml](../../../environments/common/inventory/group_vars/all/openhpc.yml)).
+   It will run healthchecks on all `IDLE` nodes which are not `DRAINED` or
+   `NOT_RESPONDING` every 300 seconds. See [slurm.conf parameters](https://slurm.schedmd.com/slurm.conf.html)
+   `HealthCheckInterval`, `HealthCheckNodeState`, `HealthCheckProgram`. These
+   may be overriden if required by redefining `openhpc_config_nhc` in e.g.
+   `environments/site/inventory/group_vars/nhc/yml`.
+
+2. Template out node health check rules using Ansible facts for each compute
+   node. Currently these check:
+    - Filesystem mounts
+    - Ethernet interfaces
+
+    See `/etc/nhc/nhc.conf` on a compute node for the full configuration.
+
+If a node healthcheck run fails, Slurm will mark the node `DOWN`. With the
+default [alerting configuration](../../../docs/alerting.md) this will trigger
+an alert.
+
+## Role Variables
+
+- `nhc_config_template`: Template to use. Default is the in-role template
+  providing rules described above.
+- `nhc_config_extra`: Possibly multiline string defining [additional rules](https://github.com/mej/nhc/blob/master/README.md) to
+  add. Jinja templating may be used. Default is empty string.
+
+## Structure
+
+This role contains 3x task files, which run at different times:
+- `main.yml`: Runs from `site.yml` -> `slurm.yml`. Templates health check
+  configuration to nodes.
+- `export.yml`: Runs from `site.yml` -> `final.yml` via role `compute_init`
+  tasks `export.yml`. Templates health check configuration to the cluster NFS
+  share for compute-init. 
+- `boot.yml`: Runs on boot via `compute_init/files/compute-init.yml`. Copies
+  the node's generated health check configuration from the cluster share to
+  local disk.
+
+Note that the `stackhpc.openhpc` role:
+- Installs the required package
+- Configures slurm.conf parameterss
diff --git a/ansible/roles/nhc/defaults/main.yml b/ansible/roles/nhc/defaults/main.yml
@@ -0,0 +1,2 @@
+nhc_config_template: nhc.conf.j2
+nhc_config_extra: ''
diff --git a/ansible/roles/nhc/tasks/boot.yml b/ansible/roles/nhc/tasks/boot.yml
@@ -0,0 +1,8 @@
+- name: Copy stored NHC configuration to active location
+  ansible.builtin.copy:
+    remote_src: true
+    src: "/var/tmp/cluster/hostconfig/{{ ansible_hostname }}/nhc.conf"
+    dest: /etc/nhc/nhc.conf
+    owner: root
+    group: root
+    mode: u=rw,go=
diff --git a/ansible/roles/nhc/tasks/export.yml b/ansible/roles/nhc/tasks/export.yml
@@ -0,0 +1,6 @@
+# Used for compute-init
+- name: Template out host specific NHC config
+  ansible.builtin.template:
+    src: "{{ nhc_config_template }}"
+    dest: "/exports/cluster/hostconfig/{{ inventory_hostname }}/nhc.conf"
+  delegate_to: "{{ groups['control'] | first }}"
diff --git a/ansible/roles/nhc/tasks/main.yml b/ansible/roles/nhc/tasks/main.yml
@@ -0,0 +1,20 @@
+
+- name: Ensure NHC configuration directory exists
+  # When running site.yml after login/control upgrade, nhc group might be
+  # enabled in repo, but as the compute nodes have not yet been upgraded they
+  # do not have the package and hence this directory
+  ansible.builtin.file:
+    path: /etc/nhc
+    state: directory
+    # to match nhc-ohpc install:
+    owner: root
+    group: root
+    mode: u=rwX,go=
+
+- name: Template out NHC configuration
+  ansible.builtin.template:
+    src: "{{ nhc_config_template }}"
+    dest: /etc/nhc/nhc.conf
+    owner: root
+    group: root
+    mode: u=rw,go= # HealthChecProgram is run by root
diff --git a/ansible/roles/nhc/templates/nhc.conf.j2 b/ansible/roles/nhc/templates/nhc.conf.j2
@@ -0,0 +1,18 @@
+# {{ ansible_managed }}
+
+# Nodes have long FQDN hostnames but short nodenames
+* || HOSTNAME="$HOSTNAME_S"
+
+## Filesystem checks
+{% for mount in ansible_mounts %}
+{% set mount_mode = 'rw' if 'rw' in mount.options.split(',') else 'ro' %}
+{{ ansible_fqdn }} || check_fs_mount_{{ mount_mode }} -t "{{ mount.fstype }}" -s "{{ mount.device }}" -f "{{ mount.mount }}"
+{% endfor %}
+
+## Ethernet interface checks
+{% for iface in ansible_interfaces | select('match', 'eth') %}
+{{ ansible_fqdn }} || check_hw_eth {{ iface }}
+{% endfor %}
+
+## Site-specific checks
+{{ nhc_config_extra }}
diff --git a/ansible/site.yml b/ansible/site.yml
@@ -27,7 +27,7 @@
 - import_playbook: slurm.yml
 - import_playbook: portal.yml
 - import_playbook: monitoring.yml
-- import_playbook: disable-repos.yml
+- import_playbook: final.yml
 
 - name: Run post.yml hook
   vars:
@@ -37,12 +37,4 @@
   import_playbook: "{{ hook_path if hook_path | exists else 'noop.yml' }}"
   when: hook_path | exists
 
-- name: Clean up and shutdown Packer VM
-  hosts: builder
-  gather_facts: no
-  become: yes
-  tasks:
-    - import_tasks: cleanup.yml
-    - community.general.shutdown:
-
 ...
diff --git a/ansible/slurm.yml b/ansible/slurm.yml
@@ -63,3 +63,12 @@
     - include_role:
         name: stackhpc.openhpc
         tasks_from: "{{ 'runtime.yml' if appliances_mode == 'configure' else 'main.yml' }}"
+
+- name: Setup Node Health Checks
+  # Has to be done here as it requires openhpc repos etc for installation
+  hosts: nhc:!builder
+  become: yes
+  tags: nhc
+  tasks:
+    - include_role:
+        name: nhc
diff --git a/docs/production.md b/docs/production.md
@@ -166,3 +166,5 @@ and referenced from the `site` and `production` environments, e.g.:
   is properly tuned, again, demonstrated through testing.
 
 - Enable alertmanager if Slack is available - see [docs/alerting.md](./alerting.md).
+
+- Enable node health checks - see [ansible/roles/nhc/README.md](../ansible/roles/nhc/README.md).
diff --git a/environments/.stackhpc/tofu/cluster_image.auto.tfvars.json b/environments/.stackhpc/tofu/cluster_image.auto.tfvars.json
@@ -1,6 +1,6 @@
 {
     "cluster_image": {
-        "RL8": "openhpc-RL8-250617-1558-2065722e",
-        "RL9": "openhpc-RL9-250617-1557-2065722e"
+        "RL8": "openhpc-RL8-250620-1251-d606a45c",
+        "RL9": "openhpc-RL9-250620-1251-d606a45c"
     }
 }
diff --git a/environments/.stackhpc/tofu/main.tf b/environments/.stackhpc/tofu/main.tf
@@ -78,7 +78,7 @@ module "cluster" {
         standard = { # NB: can't call this default!
             nodes = ["compute-0", "compute-1"]
             flavor = var.other_node_flavor
-            compute_init_enable = ["compute", "chrony", "etc_hosts", "nfs", "basic_users", "eessi", "tuned", "cacerts"]
+            compute_init_enable = ["compute", "chrony", "etc_hosts", "nfs", "basic_users", "eessi", "tuned", "cacerts", "nhc"]
             ignore_image_changes = true
         }
         # Normally-empty partition for testing:
diff --git a/environments/common/inventory/group_vars/all/openhpc.yml b/environments/common/inventory/group_vars/all/openhpc.yml
@@ -37,26 +37,51 @@ openhpc_packages_default:
   - slurm-libpmi-ohpc # to allow intel mpi to work properly
   - ohpc-gnu12-openmpi4-perf-tools # for hpctests
   - openblas-gnu12-ohpc # for hpctests (HPL)
+  - nhc-ohpc # node health checks
   # EPEL packages:
   - apptainer
   - podman-compose
 openhpc_packages_extra: []
 openhpc_packages: "{{ (openhpc_packages_default + openhpc_packages_extra) | select | list }}"
 openhpc_munge_key: "{{ vault_openhpc_mungekey | b64decode }}"
 openhpc_login_only_nodes: login
+openhpc_state_save_location: "{{ appliances_state_dir + '/slurmctld' if appliances_state_dir is defined else '/var/spool' }}"
+
+# additional site/environment-specific slurm.conf parameters:
+# NB: override in environments/site/inventory/group_vars/all/openhpc.yml, not here:
+openhpc_config_extra: {}
+
+# default additional slurm.conf parameters for the appliance:
 openhpc_config_default:
   SlurmctldParameters:
     - enable_configless
   TaskPlugin: task/cgroup,task/affinity
-  ReturnToService: 2 # is stackhpc.openhpc default, but templating bug means it is needed here too
+  ReturnToService: 2 # workaround for templating bug TODO: Remove once on stackhpc.openhpc v1.2.0
+
+# default additional slurm.conf parameters when "rebuild" enabled:
 openhpc_config_rebuild:
   RebootProgram: /opt/slurm-tools/bin/slurm-openstack-rebuild
   SlurmctldParameters:
     - reboot_from_controller
   ResumeTimeout: 300
-openhpc_config_extra: {}
-openhpc_config: "{{ openhpc_config_default | combine(openhpc_config_rebuild if groups['rebuild'] | length > 0 else {}, openhpc_config_extra, list_merge='append') }}"
-openhpc_state_save_location: "{{ appliances_state_dir + '/slurmctld' if appliances_state_dir is defined else '/var/spool' }}"
+
+# default additional slurm.conf parameters when "nhc" enabled:
+openhpc_config_nhc:
+  HealthCheckProgram: /usr/sbin/nhc
+  HealthCheckInterval: 300
+  HealthCheckNodeState: NONDRAINED_IDLE
+
+# indirection to allow automatic construction of slurm.conf parameters:
+openhpc_config_groups:
+  - enabled: "{{ groups['rebuild'] | length > 0 }}"
+    config: "{{ openhpc_config_rebuild }}"
+  - enabled: "{{ groups['nhc'] | length > 0 }}"
+    config: "{{ openhpc_config_nhc }}"
+  - enabled: true
+    config: "{{ openhpc_config_extra }}"
+
+# constructed slurm.conf parameters for stackpc.openhpc role var:
+openhpc_config: "{{ openhpc_config_default | combine(openhpc_config_groups | selectattr('enabled') | map(attribute='config'), list_merge='append') }}"
 
 openhpc_install_type: ohpc # 'ohpc' or 'generic', see https://github.com/stackhpc/ansible-slurm-appliance/pull/326
 
diff --git a/environments/common/inventory/groups b/environments/common/inventory/groups
@@ -182,3 +182,6 @@ extra_packages
 
 [gateway]
 # Add builder to this group to install gateway ansible-init playbook into image
+
+[nhc]
+# Hosts to configure for node health checks - either entire 'compute' group or empty
diff --git a/environments/common/layouts/everything b/environments/common/layouts/everything
@@ -125,3 +125,7 @@ builder
 [gateway:children]
 # Add builder to this group to install gateway ansible-init playbook into image
 builder
+
+[nhc:children]
+# Hosts to configure for node health checks
+compute
