From ee017e829d0dd04b6805bac932756a80cac53923 Mon Sep 17 00:00:00 2001
From: Pieter Neerincx <pieter.neerincx@gmail.com>
Date: Thu, 23 Feb 2023 23:29:43 +0100
Subject: [PATCH 1/2] Improved lustre client config to survive reboots.

---
 roles/lustre_client/tasks/install.yml | 30 ++++++++++++++++++++++-----
 1 file changed, 25 insertions(+), 5 deletions(-)

diff --git a/roles/lustre_client/tasks/install.yml b/roles/lustre_client/tasks/install.yml
index 5fc8677be..bd1b32cc8 100644
--- a/roles/lustre_client/tasks/install.yml
+++ b/roles/lustre_client/tasks/install.yml
@@ -106,15 +106,27 @@
   become: true
 
 - name: 'Create lustre lnet config: part 2.'
+  ansible.builtin.command:
+    cmd: |
+         lnetctl net del --net tcp
+  register: lnetctl_del_status
+  changed_when:
+    - lnetctl_del_status.rc == 0
+  failed_when:
+    - lnetctl_del_status.rc > 0
+    - '"errno: -2" not in lnetctl_del_status.stderr'
+  become: true
+
+- name: 'Create lustre lnet config: part 3.'
   ansible.builtin.command:
     cmd: |
          lnetctl net add --net "{{ item.name }}" --if "{{ item.interface }}"
-  register: lnetctl_status
+  register: lnetctl_add_status
   changed_when:
-    - lnetctl_status.rc == 0
+    - lnetctl_add_status.rc == 0
   failed_when:
-    - lnetctl_status.rc > 0
-    - '"errno: -17" not in lnetctl_status.stderr'
+    - lnetctl_add_status.rc > 0
+    - '"errno: -17" not in lnetctl_add_status.stderr'
   with_items: "{{ lustre_client_networks }}"
   become: true
 
@@ -122,7 +134,7 @@
   ansible.builtin.shell:
     cmd: |
          checksum_old=$(md5sum /etc/lnet.conf)
-         lnetctl export /etc/lnet.conf
+         lnetctl net show > /etc/lnet.conf
          checksum_new=$(md5sum /etc/lnet.conf)
          if [[ "${checksum_old}" != "${checksum_new}" ]]; then
              echo 'lnet.conf has changed.'
@@ -131,4 +143,12 @@
   changed_when:
     - '"lnet.conf has changed." in lnetconf_status.stdout'
   become: true
+
+- name: Start lnet.service.
+  ansible.builtin.systemd:
+    name: lnet.service
+    enabled: true
+    state: started
+    daemon_reload: true
+  become: true
 ...

From 6e159c2ce79f409fb0fab8ac1c77551286ea7135 Mon Sep 17 00:00:00 2001
From: Pieter Neerincx <pieter.neerincx@gmail.com>
Date: Sun, 26 Feb 2023 18:25:29 +0100
Subject: [PATCH 2/2] Improved lustre_client: ensure lnet starts and lustre
 file systems are mounted on (re)boot.

---
 roles/lustre_client/handlers/main.yml | 24 +++++++++++++++++++
 roles/lustre_client/tasks/install.yml | 33 ++++++++++++++++++++++++++-
 2 files changed, 56 insertions(+), 1 deletion(-)

diff --git a/roles/lustre_client/handlers/main.yml b/roles/lustre_client/handlers/main.yml
index 08ee7aa23..fdd4de594 100644
--- a/roles/lustre_client/handlers/main.yml
+++ b/roles/lustre_client/handlers/main.yml
@@ -3,4 +3,28 @@
   ansible.builtin.reboot:
   become: true
   listen: reboot
+
+- name: 'Re-enable lnet.service part 1: disable lnet.service.'
+  ansible.builtin.systemd:
+    name: lnet.service
+    enabled: false
+    daemon_reload: true
+  become: true
+  listen: reenable_lnet
+
+- name: 'Re-enable lnet.service part 2: enable lnet.service.'
+  ansible.builtin.systemd:
+    name: lnet.service
+    enabled: true
+    daemon_reload: true
+  become: true
+  listen: reenable_lnet
+
+- name: Restart lnet.service.
+  ansible.builtin.systemd:
+    name: lnet.service
+    state: restarted
+    daemon_reload: true
+  become: true
+  listen: restart_lnet
 ...
diff --git a/roles/lustre_client/tasks/install.yml b/roles/lustre_client/tasks/install.yml
index bd1b32cc8..8ec990146 100644
--- a/roles/lustre_client/tasks/install.yml
+++ b/roles/lustre_client/tasks/install.yml
@@ -134,7 +134,7 @@
   ansible.builtin.shell:
     cmd: |
          checksum_old=$(md5sum /etc/lnet.conf)
-         lnetctl net show > /etc/lnet.conf
+         lnetctl export -b /etc/lnet.conf
          checksum_new=$(md5sum /etc/lnet.conf)
          if [[ "${checksum_old}" != "${checksum_new}" ]]; then
              echo 'lnet.conf has changed.'
@@ -142,6 +142,37 @@
   register: lnetconf_status
   changed_when:
     - '"lnet.conf has changed." in lnetconf_status.stdout'
+  notify:
+    - restart_lnet
+  become: true
+
+- name: Patch lnet.service file for systemd to ignore failure if a Lustre network ID was already loaded.
+  ansible.builtin.lineinfile:
+    path: /usr/lib/systemd/system/lnet.service
+    backup: true
+    insertafter: '\[Service\]'
+    regexp: '^#?SuccessExitStatus='
+    line: 'SuccessExitStatus=239'  # lustre NID already loaded.
+    owner: root
+    group: root
+    mode: '0644'
+  notify:
+    - restart_lnet
+  become: true
+
+- name: Patch lnet.service file for systemd to start lnet.service before remote-fs.target.
+  ansible.builtin.lineinfile:
+    path: /usr/lib/systemd/system/lnet.service
+    backup: true
+    insertafter: '\[Install\]'
+    regexp: '^#?WantedBy='
+    line: 'WantedBy=multi-user.target remote-fs.target'
+    owner: root
+    group: root
+    mode: '0644'
+  notify:
+    - reenable_lnet
+    - restart_lnet
   become: true
 
 - name: Start lnet.service.