Revise raid startup script

modules/scripts/startup-script/README.md

@@ -298,11 +298,11 @@ No modules.
 | <a name="input_install_docker"></a> [install\_docker](#input\_install\_docker) | Install Docker command line tool and daemon. | `bool` | `false` | no |
 | <a name="input_install_stackdriver_agent"></a> [install\_stackdriver\_agent](#input\_install\_stackdriver\_agent) | Run Google Stackdriver Agent installation script if set to true. Preferred over ops agent for performance. | `bool` | `false` | no |
 | <a name="input_labels"></a> [labels](#input\_labels) | Labels for the created GCS bucket. Key-value pairs. | `map(string)` | n/a | yes |
+| <a name="input_local_ssd_filesystem"></a> [local\_ssd\_filesystem](#input\_local\_ssd\_filesystem) | Format and mount filesystem from local SSD scratch disks, preserving existing data | <pre>object({<br>    enable     = optional(bool, false)<br>    fs_type    = optional(string, "ext4")<br>    mountpoint = optional(string, "/mnt/localssd")<br>  })</pre> | n/a | yes |
 | <a name="input_prepend_ansible_installer"></a> [prepend\_ansible\_installer](#input\_prepend\_ansible\_installer) | DEPRECATED. Use `install_ansible=false` to prevent ansible installation. | `bool` | `null` | no |
 | <a name="input_project_id"></a> [project\_id](#input\_project\_id) | Project in which the HPC deployment will be created | `string` | n/a | yes |
 | <a name="input_region"></a> [region](#input\_region) | The region to deploy to | `string` | n/a | yes |
 | <a name="input_runners"></a> [runners](#input\_runners) | List of runners to run on remote VM.<br>    Runners can be of type ansible-local, shell or data.<br>    A runner must specify one of 'source' or 'content'.<br>    All runners must specify 'destination'. If 'destination' does not include a<br>    path, it will be copied in a temporary folder and deleted after running.<br>    Runners may also pass 'args', which will be passed as argument to shell runners only. | `list(map(string))` | `[]` | no |
-| <a name="input_setup_raid"></a> [setup\_raid](#input\_setup\_raid) | Set up RAID0 for existing local ssds. | `bool` | `false` | no |
 
 ## Outputs
 

modules/scripts/startup-script/files/setup-raid.yml

@@ -18,79 +18,57 @@
   become: true
   hosts: localhost
   vars:
-    DST_MNT: "/mnt/localssd"
-    bashrc: "{{ '/etc/bashrc' if ansible_facts['os_family'] == 'RedHat' else '/etc/bash.bashrc' }}"
-    array_dev: "/dev/md0"
-    mode: 755
+    raid_name: localssd
+    mountpoint: /mnt/{{ raid_name }}
+    array_dev: /dev/md/{{ raid_name }}
+    fstype: ext4
+    interface: nvme
+    mode: '0755'
   tasks:
-  - name: Check if DST_MNT exists
-    ansible.builtin.stat:
-      path: "{{ DST_MNT }}"
-    register: dst_mnt_exists
-
-  - name: Check if {{ array_dev }} exists
-    ansible.builtin.stat:
-      path: "{{ array_dev }}"
-    register: array_dev_exists
-
   - name: Install mdadm
     ansible.builtin.package:
       name: mdadm
       state: present
 
-  - name: Get number of devices
-    ansible.builtin.shell: nvme list | { grep nvme_ || test $? = 1; }  | { grep -v nvme_card-pd || test $? = 1; } | awk '{print $1}' | wc -l
-    register: nvme_device_count
-
-  - name: Get list of devices
-    ansible.builtin.shell: nvme list | { grep nvme_ || test $? = 1; }  | { grep -v nvme_card-pd || test $? = 1; } | awk '{print $1}' | paste -sd ' '
-    register: nvme_device_list
+  - name: Get local SSD devices
+    ansible.builtin.find:
+      file_type: link
+      path: /dev/disk/by-id
+      patterns: google-local-{{ "nvme-" if interface == "nvme" else "" }}ssd-*
+    register: local_ssd_devices
 
-  - name: Exit if less than 2 local ssd
+  - name: Exit if zero local ssd found
     ansible.builtin.meta: end_play
-    when: nvme_device_count.stdout < "2"
-
-  - name: Check if there is an existing foreign RAID device
-    ansible.builtin.shell: lsblk | grep -q 'md127'
-    register: raid_check
-    failed_when: false
-    changed_when: false
-
-  - name: Stop and remove foreign RAID device
-    ansible.builtin.shell: mdadm --manage /dev/md127 --stop --remove
-    when: raid_check.rc == 0
-
-  - name: Zero superblock on foreign RAID device
-    ansible.builtin.shell:
-      cmd: sudo mdadm --zero-superblock {{ nvme_device_list.stdout }}
-    when: raid_check.rc == 0
-
-  - name: Create RAID-0 array if none were found
-    ansible.builtin.command: mdadm --create /dev/md0 --level=0 --raid-devices={{ nvme_device_count.stdout }} {{ nvme_device_list.stdout }}
-    when: not array_dev_exists.stat.exists
-
-  - name: Create ext4 filesystem
-    filesystem:
-      fstype: ext4
-      device: /dev/md0
-
-  - name: Tune reserved blocks
-    ansible.builtin.shell:
-      cmd: tune2fs /dev/md0 -r 131072
-
-  - name: Add a label
-    ansible.builtin.shell: e2label /dev/md0 LOCALSSD
-
-  - name: Create mount point
-    ansible.builtin.file:
-      path: "{{ DST_MNT }}"
-      state: directory
-      mode: "{{ mode }}"
+    when: local_ssd_devices.files | length == 0
+
+  - name: Force RAID array if only 1 local SSD
+    ansible.builtin.shell: mdadm --create "{{ array_dev }}" --name="{{ raid_name }}" --homehost=any --level=0 --raid-devices=1 /dev/disk/by-id/google-local-nvme-ssd-* --force
+    args:
+      creates: "{{ array_dev }}"
+    when: local_ssd_devices.files | length == 1
+
+  - name: Create RAID array
+    ansible.builtin.shell: mdadm --create "{{ array_dev }}" --name="{{ raid_name }}" --homehost=any --level=0 --raid-devices="{{ local_ssd_devices.files | length }}" /dev/disk/by-id/google-local-nvme-ssd-*
+    args:
+      creates: "{{ array_dev }}"
+    when: local_ssd_devices.files | length >= 2
+
+  - name: Format filesystem
+    community.general.filesystem:
+      fstype: "{{ fstype }}"
+      device: "{{ array_dev }}"
+      opts: -m 0
 
   - name: Mount RAID array
     ansible.posix.mount:
-      src: /dev/md0
-      path: "{{ DST_MNT }}"
-      fstype: ext4
+      src: "{{ array_dev }}"
+      path: "{{ mountpoint }}"
+      fstype: "{{ fstype }}"
       opts: discard,defaults,nofail
       state: mounted
+
+  - name: Set mount permissions
+    ansible.builtin.file:
+      path: "{{ mountpoint }}"
+      state: directory
+      mode: "{{ mode }}"

modules/scripts/startup-script/main.tf

@@ -98,16 +98,20 @@ locals {
     },
   ]
 
-  raid_setup = !var.setup_raid ? [] : [
+  raid_setup = var.local_ssd_filesystem == {} ? [] : [
     {
       type        = "ansible-local"
       destination = "setup-raid.yml"
       content     = file("${path.module}/files/setup-raid.yml")
+      args = join(" ", [
+        "-e mountpoint=${var.local_ssd_filesystem.mountpoint}",
+        "-e fs_type=${var.local_ssd_filesystem.fs_type}",
+      ])
     },
   ]
 
   supplied_ansible_runners = anytrue([for r in var.runners : r.type == "ansible-local"])
-  has_ansible_runners      = anytrue([local.supplied_ansible_runners, local.configure_ssh, var.install_docker, var.setup_raid])
+  has_ansible_runners      = anytrue([local.supplied_ansible_runners, local.configure_ssh, var.install_docker, var.local_ssd_filesystem.enable])
   install_ansible          = coalesce(var.install_ansible, local.has_ansible_runners)
   ansible_installer = local.install_ansible ? [{
     type        = "shell"

modules/scripts/startup-script/variables.tf

@@ -126,10 +126,13 @@ variable "install_docker" {
   nullable    = false
 }
 
-variable "setup_raid" {
-  description = "Set up RAID0 for existing local ssds."
-  type        = bool
-  default     = false
+variable "local_ssd_filesystem" {
+  description = "Format and mount filesystem from local SSD scratch disks, preserving existing data"
+  type = object({
+    enable     = optional(bool, false)
+    fs_type    = optional(string, "ext4")
+    mountpoint = optional(string, "/mnt/localssd")
+  })
 }
 
 variable "install_cloud_ops_agent" {