DA Scale change to public repo

datasource.tf

@@ -10,11 +10,17 @@ data "ibm_is_zone" "zone" {
 }
 */
 
+#Fetching Existing VPC CIDR for Security rules:
 data "ibm_is_vpc" "existing_vpc" {
   count = var.vpc_name != null ? 1 : 0
   name  = var.vpc_name
 }
 
+data "ibm_is_vpc_address_prefixes" "existing_vpc_cidr" {
+  count = var.vpc_name != null ? 1 : 0
+  vpc   = data.ibm_is_vpc.existing_vpc[0].id
+}
+
 /*
 data "ibm_is_subnet" "subnet" {
   count      = length(local.subnets)
@@ -28,8 +34,8 @@ data "ibm_is_subnet" "subnet" {
 # }
 
 data "ibm_is_subnet" "existing_cluster_subnets" {
-  count      = var.vpc_name != null && var.cluster_subnet_ids != null ? 1 : 0
-  identifier = var.cluster_subnet_ids
+  count      = var.vpc_name != null && var.cluster_subnet_id != null ? 1 : 0
+  identifier = var.cluster_subnet_id
 }
 
 data "ibm_is_subnet" "existing_storage_subnets" {
@@ -58,6 +64,7 @@ data "ibm_is_ssh_key" "ssh_keys" {
 }
 
 data "ibm_is_subnet" "compute_subnet_crn" {
+  count      = var.vpc_name != null && var.cluster_subnet_id != null ? 1 : 0
   identifier = local.compute_subnet_id
 }
 
@@ -83,7 +90,7 @@ data "ibm_is_instance_profile" "protocol_profile" {
 }
 
 data "ibm_is_subnet_reserved_ips" "protocol_subnet_reserved_ips" {
-  count  = local.scale_ces_enabled == true ? 1 : 0
+  count  = var.enable_deployer == false && local.scale_ces_enabled == true ? 1 : 0
   subnet = local.protocol_subnet_id
 }
 

modules/ansible-roles/roles/cloudmonitoring/tasks/mgmt-cloudmonitoring-configure.yml

@@ -71,8 +71,14 @@
     line: "{{ item.line }}"
     create: yes
   loop:
-    - { regexp: "==ACCESSKEY==", line: "customerid: {{ cloud_monitoring_access_key }}" }
-    - { regexp: "==COLLECTOR==", line: "collector: {{ cloud_monitoring_ingestion_url }}" }
+    - {
+        regexp: "==ACCESSKEY==",
+        line: "customerid: {{ cloud_monitoring_access_key }}",
+      }
+    - {
+        regexp: "==COLLECTOR==",
+        line: "collector: {{ cloud_monitoring_ingestion_url }}",
+      }
     - { regexp: "^tags:", line: "tags: type:management,lsf:true" }
   when: monitoring_enable_for_management | bool
 
@@ -113,14 +119,11 @@
     executable: /bin/bash
   when: monitoring_enable_for_management | bool
 
-- name: Install LSF Prometheus Exporter if not already installed
-  ansible.builtin.shell: |
-    echo "Exporter not found. Cloning and installing...";
-    rm -rf /tmp/lsf_prometheus_exporter
-    git clone --branch wheel_pkg https://[email protected]/platformcomputing/lsf_prometheus_exporter /tmp/lsf_prometheus_exporter
-    {{ pip_executable }} install /tmp/lsf_prometheus_exporter/lsf_prometheus_exporter-1.0.0-py3-none-any.whl
-  args:
-    executable: /bin/bash
+- name: Install LSF Prometheus Exporter using pip
+  ansible.builtin.pip:
+    name: lsf_prometheus_exporter
+    executable: /usr/local/bin/pip3.11
+    extra_args: --no-cache-dir --force-reinstall
   when:
     - monitoring_enable_for_management | bool
     - not exporter_installed.stat.exists
@@ -163,7 +166,7 @@
       exec >> /var/log/lsf_prometheus_exporter.log 2>&1
       source /opt/ibm/lsfsuite/lsf/conf/profile.lsf
       exec /usr/bin/python3 -m lsf_prometheus_exporter
-    mode: '0755'
+    mode: "0755"
     owner: lsfadmin
     group: lsfadmin
   when:
@@ -172,7 +175,7 @@
 - name: Create systemd service for Prometheus Agent
   ansible.builtin.copy:
     dest: /etc/systemd/system/prometheus.service
-    mode: '0644'
+    mode: "0644"
     content: |
       [Unit]
       Description=Prometheus Agent
@@ -198,6 +201,24 @@
     - monitoring_enable_for_management | bool
     - not exporter_installed.stat.exists
 
+- name: Enable LSF scheduler metrics for Prometheus
+  ansible.builtin.lineinfile:
+    path: "{{ LSF_CONF }}/lsbatch/{{ prefix }}/configdir/lsb.params"
+    insertbefore: "^End Parameters"
+    line: "SCHED_METRIC_ENABLE=Y"
+    state: present
+    backup: yes
+  when:
+    - monitoring_enable_for_management | bool
+
+- name: Restart lsfd service to apply scheduler metric changes
+  ansible.builtin.systemd:
+    name: lsfd
+    state: restarted
+    enabled: yes
+  when:
+    - monitoring_enable_for_management | bool
+
 - name: Reload systemd and start Prometheus Agent
   ansible.builtin.systemd:
     daemon_reload: yes
@@ -226,7 +247,7 @@
 
       [Install]
       WantedBy=multi-user.target
-    mode: '0644'
+    mode: "0644"
   when: monitoring_enable_for_management | bool
 
 - name: Reload systemd and start Prometheus Agent
@@ -240,7 +261,7 @@
 - name: Ensure start script has correct permissions
   ansible.builtin.file:
     path: /opt/ibm/lsfsuite/lsf/start_lsf_prometheus_exporter.sh
-    mode: '0755'
+    mode: "0755"
     owner: lsfadmin
     group: lsfadmin
   when: monitoring_enable_for_management | bool
@@ -262,10 +283,11 @@
       RestartSec=10
       User=lsfadmin
       Group=lsfadmin
+      Restart=always
 
       [Install]
       WantedBy=multi-user.target
-    mode: '0644'
+    mode: "0644"
   when: monitoring_enable_for_management | bool
 
 - name: Reload systemd and start LSF Prometheus Exporter

modules/ansible-roles/roles/cloudmonitoring/vars/main.yml

@@ -1,3 +1,4 @@
 sysdig_config_file: "/opt/draios/etc/dragent.yaml"
 prometheus_config_file: "/opt/prometheus/prometheus.yml"
+LSF_CONF: "/opt/ibm/lsfsuite/lsf/conf"
 PROMETHEUS_VERSION: "2.51.1"

modules/ansible-roles/roles/lsf_login_config/tasks/login_node_configuration.yml

@@ -0,0 +1,80 @@
+---
+
+- name: Check if LSF logs directory exists
+  stat:
+    path: "{{ LSF_LOGS }}"
+  register: logs_dir_stat
+
+- name: Ensure LSF logs directory exists (recurse only on first creation)
+  file:
+    path: "{{ LSF_LOGS }}"
+    state: directory
+    owner: lsfadmin
+    group: lsfadmin
+    mode: '0755'
+    recurse: "{{ not logs_dir_stat.stat.exists }}"
+
+- name: Ensure LSF conf and work are symlinks via shell
+  shell: |
+    [ -L "{{ LSF_TOP }}/{{ item }}" ] && echo "Symlink exists, skipping." || { \
+      [ -d "{{ LSF_TOP }}/{{ item }}" ] && rm -rf "{{ LSF_TOP }}/{{ item }}"; \
+      ln -s /mnt/lsf/lsf/{{ item }} "{{ LSF_TOP }}/{{ item }}"; }
+  loop:
+    - conf
+    - work
+
+- name: Ensure correct ownership and permissions of /opt/ibm/lsfsuite
+  file:
+    path: "{{ LSF_SUITE }}"
+    owner: lsfadmin
+    group: lsfadmin
+    mode: '0777'
+    recurse: yes
+
+- name: Set login_node_host to first host in login_node group
+  set_fact:
+    login_node_host: "{{ groups['login_node'][0] }}"
+
+- name: Get IPv4 address of the current host
+  shell: "getent ahostsv4 {{ inventory_hostname }} | awk '{ print $1; exit }'"
+  register: ip_result
+  changed_when: false
+
+- name: Ensure login node entry exists in LSF hosts file
+  lineinfile:
+    path: /mnt/lsf/lsf/conf/hosts
+    line: "{{ ip_result.stdout }} {{ login_node_host }}.{{ dns_domain_names }}"
+    state: present
+    insertafter: EOF
+    create: yes
+
+- name: Insert hostname line after "#prune" only once
+  lineinfile:
+    path: "{{ LSF_CLUSTER_FILE }}"
+    insertafter: "^#prune"
+    line: "{{ login_node_host }}.{{ dns_domain_names }} Intel_E5 X86_64 0 ()"
+    state: present
+
+- name: Ensure LSF profile is sourced in root's .bashrc
+  lineinfile:
+    path: "/root/.bashrc"
+    line: "source {{ LSF_CONF }}/profile.lsf"
+    state: present
+
+- name: Ensure LSF profile is sourced in lsfadmin's .bashrc
+  lineinfile:
+    path: "{{ LSFADMIN_DIR }}/.bashrc"
+    line: "source {{ LSF_CONF }}/profile.lsf"
+    state: present
+
+- name: Source current user's .bashrc (only if updated)
+  shell: |
+    grep -q "source {{ LSF_CONF }}/profile.lsf" /root/.bashrc && source /root/.bashrc || true
+  args:
+    executable: /bin/bash
+
+- name: Source lsfadmin's .bashrc (only if updated)
+  shell: |
+    grep -q "source {{ LSF_CONF }}/profile.lsf" "{{ LSFADMIN_DIR }}/.bashrc" && source "{{ LSFADMIN_DIR }}/.bashrc" || true
+  args:
+    executable: /bin/bash

modules/ansible-roles/roles/lsf_login_config/tasks/main.yml

@@ -0,0 +1,4 @@
+---
+
+# Configure Login node
+- import_tasks: login_node_configuration.yml

modules/ansible-roles/roles/lsf_login_config/vars/main.yml

@@ -0,0 +1,8 @@
+LSF_SUITE: "/opt/ibm/lsfsuite"
+LSF_TOP: "{{ LSF_SUITE }}/lsf"
+LSF_CONF: "{{ LSF_TOP }}/conf"
+LSF_WORK: "{{ LSF_TOP }}/work"
+LSF_LOGS: "/opt/ibm/lsflogs"
+LSF_HOSTS_FILE: "{{ LSF_CONF }}/hosts"
+LSF_CLUSTER_FILE: "{{ LSF_CONF }}/lsf.cluster.{{ prefix }}"
+LSFADMIN_DIR: "/home/lsfadmin"

modules/ansible-roles/roles/lsf_mgmt_config/tasks/app_center_configure.yml

@@ -0,0 +1,73 @@
+---
+# AppCenter HTTPS Configuration
+
+- name: PAC | Check if HTTPS is already enabled
+  ansible.builtin.command: pmcadmin https enable
+  register: https_check
+  changed_when: "'was already enabled' not in https_check.stdout"
+  failed_when: false
+  run_once: true
+
+- name: PAC | Debug HTTPS status
+  ansible.builtin.debug:
+    msg: "HTTPS is already enabled"
+  when: "'was already enabled' in https_check.stdout"
+  run_once: true
+
+- name: PAC | Configure HTTPS for AppCenter
+  block:
+
+    - name: PAC | Set GUI password for lsfadmin
+      ansible.builtin.command: passwd --stdin lsfadmin
+      args:
+        stdin: "{{ app_center_gui_password }}"
+
+    - name: PAC | Enable HTTPS access for AppCenter
+      ansible.builtin.command: >
+        pmcadmin https enable
+        --password {{ app_center_gui_password }}
+        --validhosts localhost
+
+    - name: PAC | Stop pmcadmin service
+      ansible.builtin.command: pmcadmin stop
+
+    - name: PAC | Pause before restarting pmcadmin
+      ansible.builtin.pause:
+        seconds: 5
+
+    - name: PAC | Start pmcadmin service
+      ansible.builtin.command: pmcadmin start
+
+    - name: PAC | Update JS_PAC_SERVER_URL in js.conf
+      ansible.builtin.lineinfile:
+        path: "{{ JS_PAC_SERVER_URL }}"
+        regexp: '^JS_PAC_SERVER_URL='
+        line: "JS_PAC_SERVER_URL=https://{{ lsf_masters[0] }}:8443"
+        backrefs: true
+
+    - name: PAC | Stop ACD (Application Center Daemon) service
+      ansible.builtin.service:
+        name: acd
+        state: stopped
+
+    - name: PAC | Pause before restarting ACD
+      ansible.builtin.pause:
+        seconds: 5
+
+    - name: PAC | Start ACD (Application Center Daemon) service
+      ansible.builtin.service:
+        name: acd
+        state: started
+
+  rescue:
+    - name: PAC | Log error if AppCenter HTTPS configuration fails
+      ansible.builtin.debug:
+        msg: "AppCenter HTTPS configuration block failed. Check previous task results."
+
+  always:
+    - name: PAC | Always log final status of AppCenter HTTPS configuration
+      ansible.builtin.debug:
+        msg: "AppCenter HTTPS configuration block completed (success or failure)."
+
+  when: "'was already enabled' not in https_check.stdout"
+  run_once: true

modules/ansible-roles/roles/lsf_mgmt_config/tasks/configure_dynamic_nodes_templates.yml

@@ -33,13 +33,3 @@
     dest: "{{ LSF_RC_IC_CONF }}/user_data.sh"
     mode: '0644'
   run_once: true
-
-- name: Management Config Templates | Restart lsfd service
-  service:
-    name: lsfd
-    state: restarted
-
-- name: Management Config Templates | Restart NetworkManager
-  service:
-    name: NetworkManager
-    state: restarted

modules/ansible-roles/roles/lsf_mgmt_config/tasks/configure_management_nodes.yml

@@ -103,3 +103,21 @@
     path: "{{ LSF_CLUSTER_FILE }}"
     regexp: '^lsfservers'
     state: absent
+  run_once: true
+
+# Temporary: Remove after new image build includes cleanup
+- name: Temporary Cleanup | Delete all 'sagar-fp-15-new1' folders
+  ansible.builtin.shell: |
+    find "{{ LSF_EXT_CONF }}" -type d -name "sagar-fp-15-new1" -exec rm -rf {} +
+  args:
+    warn: false
+  ignore_errors: true
+  when: inventory_hostname in groups['management_nodes']
+
+# Temporary: Remove after new image build includes cleanup
+- name: Temporary Cleanup | Replace 'sagar-fp-15-new1' with 'lsfservers'
+  ansible.builtin.shell: |
+    grep -rl 'sagar-fp-15-new1' "{{ LSF_EXT_CONF }}"  | xargs sed -i 's/sagar-fp-15-new1/lsfservers/g' || true
+  args:
+    warn: false
+  when: inventory_hostname in groups['management_nodes']

modules/ansible-roles/roles/lsf_mgmt_config/tasks/hosts_file_update.yml

@@ -31,7 +31,7 @@
 
 - name:  Management Config | Aggregate all IPs from all hosts
   set_fact:
-    all_ips: "{{ groups['all'] | map('extract', hostvars, 'host_ip') | list }}"
+    all_ips: "{{ groups['mgmt_compute_nodes'] | map('extract', hostvars, 'host_ip') | list }}"
   run_once: true
 
 - name:  Management Config | Display all resolved IP addresses