Support lustre client

ansible/.gitignore

@@ -58,4 +58,5 @@ roles/*
 !roles/squid/**
 !roles/tuned/
 !roles/tuned/**
-
+!roles/lustre/
+!roles/lustre/**

ansible/fatimage.yml

@@ -25,7 +25,7 @@
 
 - hosts: builder
   become: yes
-  gather_facts: no
+  gather_facts: yes
   tasks:
     # - import_playbook: iam.yml
     - name: Install FreeIPA client
@@ -44,6 +44,11 @@
         name: stackhpc.os-manila-mount
         tasks_from: install.yml
       when: "'manila' in group_names"
+    - name: Install Lustre packages
+      include_role:
+        name: lustre
+        tasks_from: install.yml
+      when: "'lustre' in group_names"
 
 - import_playbook: extras.yml
 
@@ -56,60 +61,76 @@
       include_role:
         name: mysql
         tasks_from: install.yml
+      when: "'mysql' in group_names"
     - name: OpenHPC
       import_role:
         name: stackhpc.openhpc
         tasks_from: install.yml
+      when: "'openhpc' in group_names"
 
     # - import_playbook: portal.yml
     - name: Open Ondemand server (packages)
       include_role:
         name: osc.ood
         tasks_from: install-package.yml
         vars_from: "Rocky/{{ ansible_distribution_major_version }}.yml"
+      when: "'openondemand' in group_names"
     # # FUTURE: install-apps.yml - this is git clones
 
     - name: Open Ondemand server (apps)
       include_role:
         name: osc.ood
         tasks_from: install-apps.yml
         vars_from: "Rocky/{{ ansible_distribution_major_version }}.yml"
+      when: "'openondemand' in group_names"
 
     - name: Open Ondemand remote desktop
       import_role:
         name: openondemand
         tasks_from: vnc_compute.yml
+      when: "'openondemand_desktop' in group_names"
+
     - name: Open Ondemand jupyter node
       import_role:
         name: openondemand
         tasks_from: jupyter_compute.yml
+      when: "'openondemand_jupyter' in group_names"
 
     # - import_playbook: monitoring.yml:
     - import_role:
         name: opensearch
         tasks_from: install.yml
       become: true
+      when: "'opensearch' in group_names"
     # slurm_stats - nothing to do
     - import_role:
         name: filebeat
         tasks_from: install.yml
+      when: "'filebeat' in group_names"
 
     - import_role:
       # can't only run cloudalchemy.node_exporter/tasks/install.yml as needs vars from preflight.yml and triggers service start
       # however starting node exporter is ok
         name: cloudalchemy.node_exporter
+      when: "'node_exporter' in group_names"
 
     - name: openondemand exporter
       dnf:
-        name: ondemand_exporter 
+        name: ondemand_exporter
+      when: "'openondemand' in group_names"
 
     - name: slurm exporter
       import_role:
         name: slurm_exporter
         tasks_from: install
       vars:
         slurm_exporter_state: stopped
+      when: "'slurm_exporter' in group_names"
 
+- hosts: prometheus
+  become: yes
+  gather_facts: yes
+  tasks:
     - import_role:
         name: cloudalchemy.prometheus
         tasks_from: preflight.yml
@@ -162,6 +183,10 @@
         - prometheus
         - promtool
 
+- hosts: grafana
+  become: yes
+  gather_facts: yes
+  tasks:
     - name: Include distribution variables for cloudalchemy.grafana
       include_vars: "{{ appliances_repository_root }}/ansible/roles/cloudalchemy.grafana/vars/redhat.yml"
     - import_role:

ansible/filesystems.yml

@@ -24,3 +24,13 @@
   tasks:
     - include_role:
         name: stackhpc.os-manila-mount
+
+- name: Setup Lustre clients
+  hosts: lustre
+  become: true
+  tags: lustre
+  tasks:
+    - include_role:
+        name: lustre
+        # NB install is ONLY run in builder
+        tasks_from: configure.yml

ansible/roles/lustre/README.md

@@ -0,0 +1,27 @@
+# lustre
+
+Install and configure a Lustre client. This builds RPM packages from source.
+
+**NB:** The `install.yml` playbook in this role should only be run during image build and is not idempotent. This will install the `kernel-devel` package; if not already installed (e.g. from an `ofed` installation), this may require enabling update of DNF packages during build using `update_enable=true`, which will upgrade the kernel as well.
+
+**NB:** Currently this only supports RockyLinux 9.
+
+## Role Variables
+
+- `lustre_version`: Optional str. Version of lustre to build, default `2.15.5` which is the first version with EL9 support
+- `lustre_lnet_label`: Optional str. The "lnet label" part of the host's NID, e.g. `tcp0` or `o2ib1`. Default `tcp`.
+- `lustre_mgs_nid`: Required str. The NID(s) for the MGS, e.g. `192.168.227.11@tcp1` (separate mutiple MGS NIDs using `:`).
+- `lustre_mounts`: Required list. Define Lustre filesystems and mountpoints as a list of dicts with keys:
+    - `fs_name`: Required str. The name of the filesystem to mount
+    - `mount_point`: Required str. Path to mount filesystem at.
+    - `mount_state`: Optional mount state, as for [ansible.posix.mount](https://docs.ansible.com/ansible/latest/collections/ansible/posix/mount_module.html#parameter-state). Default is `lustre_mount_state`.
+    - `mount_options`: Optional mount options. Default is `lustre_mount_options`.
+- `lustre_mount_state`. Optional default mount state for all mounts, as for [ansible.posix.mount](https://docs.ansible.com/ansible/latest/collections/ansible/posix/mount_module.html#parameter-state). Default is `mounted`.
+- `lustre_mount_options`. Optional default mount options. Default values are systemd defaults from [Lustre client docs](http://wiki.lustre.org/Mounting_a_Lustre_File_System_on_Client_Nodes).
+
+The following variables control the package build and and install and should not generally be required:
+- `lustre_build_packages`: Optional list. Prerequisite packages required to build Lustre. See `defaults/main.yml`.
+- `lustre_build_dir`: Optional str. Path to build lustre at, default `/tmp/lustre-release`.
+- `lustre_configure_opts`: Optional list. Options to `./configure` command. Default builds client rpms supporting Mellanox OFED, without support for GSS keys.
+- `lustre_rpm_globs`: Optional list. Shell glob patterns for rpms to install. Note order is important as the built RPMs are not in a yum repo. Default is just the `kmod-lustre-client` and `lustre-client` packages.
+- `lustre_build_cleanup`: Optional bool. Whether to uninstall prerequisite packages and delete the build directories etc. Default `true`.

ansible/roles/lustre/defaults/main.yml

@@ -0,0 +1,36 @@
+lustre_version: '2.15.5' # https://www.lustre.org/lustre-2-15-5-released/
+lustre_lnet_label: tcp
+#lustre_mgs_nid:
+lustre_mounts: []
+lustre_mount_state: mounted
+lustre_mount_options: 'defaults,_netdev,noauto,x-systemd.automount,x-systemd.requires=lnet.service'
+
+# below variables are for build and should not generally require changes
+lustre_build_packages:
+  - "kernel-devel-{{ ansible_kernel }}"
+  - git
+  - gcc
+  - libtool
+  - python3
+  - python3-devel
+  - openmpi
+  - elfutils-libelf-devel
+  - libmount-devel
+  - libnl3-devel
+  - libyaml-devel
+  - rpm-build
+  - kernel-abi-stablelists
+  - libaio
+  - libaio-devel
+lustre_build_dir: /tmp/lustre-release
+lustre_configure_opts:
+  - --disable-server
+  - --with-linux=/usr/src/kernels/*
+  - --with-o2ib=/usr/src/ofa_kernel/default
+  - --disable-maintainer-mode
+  - --disable-gss-keyring
+  - --enable-mpitests=no
+lustre_rpm_globs: # NB: order is important here, as not installing from a repo
+  - "kmod-lustre-client-{{ lustre_version | split('.') | first }}*" # only take part of the version as -RC versions produce _RC rpms
+  - "lustre-client-{{ lustre_version | split('.') | first }}*"
+lustre_build_cleanup: true

ansible/roles/lustre/tasks/configure.yml

@@ -0,0 +1,47 @@
+- name: Gather Lustre interface info
+  shell:
+    cmd: |
+      ip r get {{ _lustre_mgs_ip }}
+  changed_when: false
+  register: _lustre_ip_r_mgs
+  vars:
+    _lustre_mgs_ip: "{{ lustre_mgs_nid | split('@') | first }}"
+
+- name: Set facts for Lustre interface
+  set_fact:
+    _lustre_interface: "{{ _lustre_ip_r_mgs_info[4] }}"
+    _lustre_ip: "{{ _lustre_ip_r_mgs_info[6] }}"
+  vars:
+    _lustre_ip_r_mgs_info: "{{ _lustre_ip_r_mgs.stdout_lines.0 | split }}"
+    # first line e.g. "10.167.128.1 via 10.179.0.2 dev eth0 src 10.179.3.149 uid 1000"
+
+- name: Write LNet configuration file
+  template:
+    src: lnet.conf.j2
+    dest:  /etc/lnet.conf # exists from package install, expected by lnet service
+    owner: root
+    group: root
+    mode: u=rw,go=r # from package install
+  register: _lnet_conf
+
+- name: Ensure lnet service state
+  systemd:
+    name: lnet
+    state: "{{ 'restarted' if _lnet_conf.changed else 'started' }}"
+
+- name: Ensure mount points exist
+  ansible.builtin.file:
+    path: "{{ item.mount_point }}"
+    state: directory
+  loop: "{{ lustre_mounts }}"
+  when: "(item.mount_state | default(lustre_mount_state)) != 'absent'"
+
+- name: Mount lustre filesystem
+  ansible.posix.mount:
+    fstype: lustre
+    src: "{{ lustre_mgs_nid }}:/{{ item.fs_name }}"
+    path: "{{ item.mount_point }}"
+    state: "{{ (item.mount_state | default(lustre_mount_state)) }}"
+    opts: "{{ item.mount_options | default(lustre_mount_options) }}"
+  loop: "{{ lustre_mounts }}"
+

ansible/roles/lustre/tasks/install.yml

@@ -0,0 +1,70 @@
+- name: Install lustre build prerequisites
+  ansible.builtin.dnf:
+    name: "{{ lustre_build_packages }}"
+  register: _lustre_dnf_build_packages
+
+- name: Clone lustre git repo
+  # https://git.whamcloud.com/?p=fs/lustre-release.git;a=summary
+  ansible.builtin.git:
+    repo: git://git.whamcloud.com/fs/lustre-release.git
+    dest: "{{ lustre_build_dir }}"
+    version: "{{ lustre_version }}"
+
+- name: Prepare for lustre configuration
+  ansible.builtin.command:
+    cmd: sh ./autogen.sh
+    chdir: "{{ lustre_build_dir }}"
+
+- name: Configure lustre build
+  ansible.builtin.command:
+    cmd: "./configure {{ lustre_configure_opts | join(' ') }}"
+    chdir: "{{ lustre_build_dir }}"
+
+- name: Build lustre
+  ansible.builtin.command:
+    cmd: make rpms
+    chdir: "{{ lustre_build_dir }}"
+
+- name: Find rpms
+  ansible.builtin.find:
+    paths: "{{ lustre_build_dir }}"
+    patterns: "{{ lustre_rpm_globs }}"
+    use_regex: false
+  register: _lustre_find_rpms
+
+- name: Check rpms found
+  assert:
+    that: _lustre_find_rpms.files | length
+    fail_msg: "No lustre repos found with lustre_rpm_globs = {{ lustre_rpm_globs }}"
+
+- name: Install lustre rpms
+  ansible.builtin.dnf:
+    name: "{{ _lustre_find_rpms.files | map(attribute='path')}}"
+    disable_gpg_check: yes
+
+- block:
+  - name: Remove lustre build prerequisites
+    # NB Only remove ones this role installed which weren't upgrades
+    ansible.builtin.dnf:
+      name: "{{ _new_pkgs }}"
+      state: absent
+    vars:
+      _installed_pkgs: |
+        {{
+          _lustre_dnf_build_packages.results |
+          select('match', 'Installed:') |
+          map('regex_replace', '^Installed: (.+?)-[0-9].*$', '\1')
+        }}
+      _removed_pkgs: |
+        {{
+          _lustre_dnf_build_packages.results |
+          select('match', 'Removed:') |
+          map('regex_replace', '^Removed: (.+?)-[0-9].*$', '\1')
+        }}
+      _new_pkgs: "{{ _installed_pkgs | difference(_removed_pkgs) }}"
+
+  - name: Delete lustre build dir
+    file:
+      path: "{{ lustre_build_dir }}"
+      state: absent
+  when: lustre_build_cleanup | bool

ansible/roles/lustre/tasks/validate.yml

@@ -0,0 +1,27 @@
+- name: Assert using RockyLinux 9
+  assert:
+    that: ansible_distribution_major_version | int == 9
+    fail_msg: The 'lustre' role requires RockyLinux 9
+
+- name: Check kernel-devel package is installed
+  command: "dnf list --installed kernel-devel-{{ ansible_kernel }}"
+  changed_when: false
+  # NB: we don't check here the kernel will remain the same after reboot etc, see ofed/install.yml
+
+- name: Ensure SELinux in permissive mode
+  assert:
+    that: selinux_state in ['permissive', 'disabled']
+    fail_msg: "SELinux must be permissive for Lustre not '{{ selinux_state }}'; see variable selinux_state"
+
+- name: Ensure lustre_mgs_nid is defined
+  assert:
+    that: lustre_mgs_nid is defined
+    fail_msg: Variable lustre_mgs_nid must be defined
+
+- name: Ensure lustre_mounts entries define filesystem name and mount point
+  assert:
+    that:
+      - item.fs_name is defined
+      - item.mount_point is defined
+    fail_msg: All lustre_mounts entries must specify fs_name and mount_point
+  loop: "{{ lustre_mounts }}"

ansible/roles/lustre/templates/lnet.conf.j2

@@ -0,0 +1,6 @@
+net:
+    - net type: {{ lustre_lnet_label }}
+      local NI(s):
+        - nid: {{ _lustre_ip }}@{{ lustre_lnet_label }}
+          interfaces:
+              0: {{ _lustre_interface }}

ansible/validate.yml

@@ -95,3 +95,11 @@
     - import_role:
         name: freeipa
         tasks_from: validate.yml
+
+- name: Validate lustre configuration
+  hosts: lustre
+  tags: lustre
+  tasks:
+    - import_role:
+        name: lustre
+        tasks_from: validate.yml

environments/common/inventory/groups

@@ -134,4 +134,7 @@ freeipa_client
 # Hosts to run TuneD configuration
 
 [ansible_init]
-# Hosts to run linux-anisble-init
+# Hosts to run linux-anisble-init
+
+[lustre]
+# Hosts to run lustre client

environments/common/layouts/everything

@@ -81,4 +81,7 @@ openhpc
 
 [ansible_init:children]
 # Hosts to run ansible-init
-cluster
+cluster
+
+[lustre]
+# Hosts to run lustre client