Automated selection of smallest or biggest disk

[chkpwd]

Selecting the Appropriate Installation Disk Based on Size

In my current cloud configuration, the Kairos system is set to install on the disk designated as /dev/sda, in line with the directive provided in my .install.device configuration.

install:
  auto: true
  device: /dev/sda

This setup would function for me if all my nodes had the same disk configuration. However, what I have is:

Node	Sata Disk	NVME Disk
cattle-k3s-01	256G	1TB
cattle-k3s-02	1TB	256G
cattle-k3s-03	256G	1TB

The installation process would benefit from a more versatile disk selection strategy that can dynamically choose either the smallest or largest disk based on the specified preference. This logic could be integrated into the installation settings:

Proposed Solution for Disk Selection

I'd like to propose an enhancement to the installation configuration to include an option for selecting the smallest available disk automatically. This would require internal logic to identify and select the disk with the smallest capacity for installation. Additionally, a reboot option should be incorporated post-installation.

install:
  auto: true
  select_disk: smallest # Automated selection of smallest or biggest
  reboot: true

If this were to be implemented, it would remove boilerplate configuration for provisioning a secondary disk, as shown below:

stages:
  after-install-chroot:
    - name: "Install Longhorn Prerequisites"
      commands:
        - apt install parted nfs-common open-iscsi

  initramfs:
    - name: "Partition secondary disk if needed"
      if: >-
        [ $(sudo fdisk -l /dev/sdb | grep -q "83 Linux"; echo $?) -ne 0 ]
      commands:
        - parted /dev/sdb --script -- mkpart primary 0 -1

    - name: "Format /dev/sdb1 if needed"
      if: >-
        [ $(sudo lsblk -o FSTYPE /dev/sdb1 | tail -n 1 | wc -l) -eq 0 ]
      commands:
        - mkfs.ext4 -F /dev/sdb1

[tyzbit]

I was just thinking, if sysinfo is available at the time the install options are executed, allowing at least the device field to be Go templated would be a simple way to allow complex functionality. For example, installing to the smallest disk:

install:
  auto: true
  device: >-
    {{- $smallest := 99999 -}}
    {{ $choice := "" }}
    {{- range $disk := .Values.storage -}}
      {{- if $disk.size < $smallest -}}
        {{ $smallest = $disk.size }}
        {{ $choice = $disk.name }}
      {{- end -}}
    {{- end -}}
    {{- cat "/dev/" $choice | nospace }}

Other very useful fields being driver, vendor, model etc. With all of that combined, it should be possible to support the majority of desired configurations, even if it means doing a lot of Go templating.

[jimmykarily]

Even allowing to define an arbitrary shell script that returns the device to be used would be sufficient.

[jimmykarily]

sysinfo is not available in that block because that one is evaluated by kairos-agent itself and not parsed by yip. Making sysinfo work in that block too is another way to fix it.

[chkpwd]

All valid answers. For the time being, what can I do to workaround this?
If I move forward with the below, I could potentially wipe the disk kairos is partitioned on.

stages:
  after-install-chroot:
    - name: "Install Longhorn Prerequisites"
      commands:
        - apt install parted

  initramfs:
    - name: "Get the largest disk"
      commands: [ LARGEST_DISK=$(lsblk -AJ | jq -r '[.blockdevices[] | select(.type == "disk") | {name, size: (.size | rtrimstr("G") | tonumber)}] | max_by(.size) | .name') ]

    - name: "Partition disk"
      if: >-
        [ $(sudo fdisk -l /dev/$LARGEST_DISK | grep -q "83 Linux"; echo $?) -ne 0 ]
      commands:
        - sudo parted /dev/$LARGEST_DISK rm 1
        - parted /dev/$LARGEST_DISK --script -- mkpart primary 0 -1

    - name: "Format disk if needed"
      if: >-
        [ $(sudo lsblk -o FSTYPE /dev/$LARGEST_DISK | tail -n 1 | wc -l) -eq 0 ]
      commands:
        - mkfs.ext4 -F /dev/$LARGEST_DISK

  boot:
    - name: "Mount /dev/$LARGEST_DISK under /mnt/longhorn"
      commands:
        - mount -o rw /dev/sdb1 /mnt/longhorn

[tyzbit]

I think you can technically get the functionality necessary by latching onto kairos-install.pre.before like shown in the config documentation. You'd have a bunch of custom steps with tests (this example installs to the smallest disk between NVMe and SATA):

#cloud-config

install:
  # Make sure the installer won't delete our custom partitions
  no-format: true

stages:
  kairos-install.pre.before:
  - if:  '[ -e /dev/sda ] && [ -e /dev/nvme0n1 ] && [ $(fdisk -l /dev/sda | grep sectors | head -n 1 | cut -d" " -f5) -lt $(fdisk -l /dev/nvme0n1 | grep sectors | head -n 1 | cut -d" " -f5) ]'
    name: "Conditionally partition sda"
    commands:
      - |
        parted --script --machine -- /dev/sda mklabel msdos
    layout:
      device:
        path: /dev/sda
      expand_partition:
        size: 0 # All available space
      add_partitions:
        # all sizes bellow are in MB
        - fsLabel: COS_OEM
          size: 64
          pLabel: oem
        - fsLabel: COS_RECOVERY
          size: 8500
          pLabel: recovery
        - fsLabel: COS_STATE
          size: 18000
          pLabel: state
        - fsLabel: COS_PERSISTENT
          pLabel: persistent
          size: 25000
          filesystem: "ext4"
  - if:  '[ -e /dev/sda ] && [ -e /dev/nvme0n1 ] && [ $(fdisk -l /dev/sda | grep sectors | head -n 1 | cut -d" " -f5) -ge $(fdisk -l /dev/nvme0n1 | grep sectors | head -n 1 | cut -d" " -f5) ]'
    name: "Conditionally partition nvme0n1"
    commands:
      - |
        parted --script --machine -- /dev/nvme0n1 mklabel msdos
    layout:
      device:
        path: /dev/nvme0n1
      expand_partition:
        size: 0 # All available space
      add_partitions:
        # all sizes bellow are in MB
        - fsLabel: COS_OEM
          size: 64
          pLabel: oem
        - fsLabel: COS_RECOVERY
          size: 8500
          pLabel: recovery
        - fsLabel: COS_STATE
          size: 18000
          pLabel: state
        - fsLabel: COS_PERSISTENT
          pLabel: persistent
          size: 25000
          filesystem: "ext4"

---

I think what I suggested above with sysinfo and templating would be one of the lower-effort ways of solving this but another way to solve this is to make that field "smarter" by allowing the following options which behave in the way users would expect:

smallest - picks the disk with the least capacity
largest - picks the disk with the greatest capacity
smallest-sd - picks the disk with the least capacity that also matches /dev/sd*
lagest-sd - same as above but greatest capacity
smallest-nvme - picks the disk with the greatest capacity that alos matches /dev/nvme*
lagest-nvme - same as above but greatest capacity
etc for all reasonable device types...
auto - alias for largest to retain current behavior

And then allow YAML objects in addition to the simple strings above for device, where at least one key is required and Kairos picks the best disk based on comparing the criteria to disks from sysinfo:

install:
  device:
    driver: sd / nvme / auto
    size: smallest / largest / auto / 500
    model: ST950*

model and serial could support regex too.