Deployment of MicroCloud with a `preseed` files fails when validates node-specific by-id disk paths

[mseralessandri]

When running MicroCloud preseed with different per-node storage paths under /dev/disk/by-id, initialization fails with this error: Failed to find specified disk path: stat /dev/disk/by-id/<peer-disk-id>: no such file or directory. Here is an example of preseed file that reproduces the issue.

initiator: pc7b-ra8-n1
lookup_subnet: 10.20.7.0/24
lookup_timeout: 300

session_passphrase: "CHANGE_ME"
session_timeout: 900

systems:
  - name: pc7b-ra8-n1
    ovn_uplink_interface: bond1.3684
    ovn_underlay_ip: 10.20.7.11
    storage:
      local:
        path: /dev/disk/by-id/nvme-eui.0100000000000000c8d6b731514f0f50
        wipe: true
      ceph:
        - path: /dev/disk/by-id/nvme-eui.0100000000000000c8d6b71d512c0050
          wipe: true
          encrypt: false

  - name: pc7b-ra8-n2
    ovn_uplink_interface: bond1.3684
    ovn_underlay_ip: 10.20.7.12
    storage:
      local:
        path: /dev/disk/by-id/nvme-eui.0100000000000000c8d6b7ae674f0f50
        wipe: true
      ceph:
        - path: /dev/disk/by-id/nvme-eui.0100000000000000c8d6b77eda4e0f50
          wipe: true
          encrypt: false

  - name: pc7b-ra8-n3
    ovn_uplink_interface: bond1.3684
    ovn_underlay_ip: 10.20.7.13
    storage:
      local:
        path: /dev/disk/by-id/nvme-eui.0100000000000000c8d6b729034f0f50
        wipe: true
      ceph:
        - path: /dev/disk/by-id/nvme-eui.0100000000000000c8d6b71fbf4e0f50
          wipe: true
          encrypt: false

ceph:
  cephfs: true
  internal_network: 10.20.7.0/24
  public_network: 10.20.7.0/24

ovn:
  ipv4_gateway: 91.189.94.17/28
  ipv4_range: 91.189.94.18-91.189.94.30
  ipv6_gateway: ""
  dns_servers: ""

Each node has different by-id paths, and those paths exist only on the owning node. Since paths are validated with local os.Stat during iteration over all systems, peer node by-id paths are checked on the wrong machine, causing no such file errors on the initiator. This doesn't happen when manually deploying MicroCloud without a preseed file.

[roosterfish]

Thanks for catching this, that looks to be an annoying one indeed. The reason this isn't caught in the test suite is because all test machines are always setup with the exact same set of disks which results in identical by-id paths present on all of the machines.

The offending commit seems to be ef96113. In my opinion we can simply drop it and don't perform the check. We also don't check the disks again during interactive deployment.

The list of disks is fetched from LXD's /1.0/resources endpoint in the interactive deployment. We should do the same when being in preseed mode just to confirm the existence of the given disk paths.