LXC container's disks persistence

[degudejung]

Hi,
first of all, thanks for providing such an amazing piece of software for free. I really enjoy running Proxmox in my homelab and would certainly chose it over TrueNas, OMV again! Now, there is one issue that I am worried about:

Last night I had to restore an LXC container from backup. Backup and restore worked just fine. But now, the three disks I had mounted to the CT are completely wiped and empty. That is particularly ugly, since they are/were the TimeMachine Backup volumes for all the Mac PCs in the house.

Here's what the configuration was and what I did:

/etc/pve/lxc/100.conf

arch: amd64
cmode: shell
cores: 2
features: nesting=1
hostname: ct-webmin
memory: 512
mp0: pool1s1:vm-100-disk-1,mp=/srv/smb_temp,size=200G
mp1: pool1s1:vm-100-disk-2,mp=/srv/smb_tm_n,size=600G
mp2: pool1s1:vm-100-disk-3,mp=/srv/smb_tm_a,size=400G
mp3: /pool1h1/software,mp=/srv/smb_software
[..other mountpoints for all the ZFS datasets I want to share via SMB..]
net0: name=eth0,bridge=vmbr0,firewall=1,gw=192.168.10.1,hwaddr=6E:21:EB:39:18:CB,ip=192.168.10.22/24,type=veth
onboot: 1
ostype: debian
rootfs: pool1s1:vm-100-disk-0,size=3G
startup: order=1,up=30
swap: 512
unprivileged: 1
lxc.idmap: u 0 100000 1000
lxc.idmap: g 0 100000 100
lxc.idmap: u 1000 1000 4
lxc.idmap: g 100 100 1
lxc.idmap: u 1004 101004 64532
lxc.idmap: g 101 100101 64535
lxc.idmap: g 65534 165534 1

Maybe that's relevant: I had to play around with UID and GID mapping quite a lot in order to make the ZFS volumes available they way I want it. That was only recently.

Also, I see in that config, that I do not have a "backup=1" for the disk-1 .. disk-3 set. I did that because I want to perform the backups for those disks separately from the CT-backup. And, yes, I didn't quite get there yet, so no backups for those disks...

When, last night, I wanted to restore the CT from its backup. I could not restore it to the same ID (100), which makes sense. So I removed the CT first and then restored from backup to its original VM-ID.

My understanding was (and kind of is...) that when I create an LV and mount it to a CT as a non-OS file system, then it has "a right exist on its own" rather than being an appendix to a CT that lives and dies with it. Or at least, when I kill the CT I should be asked whether I want to kill the disks with it.

But looking at the output of lvdisplay makes me fear just the worst:

  --- Logical volume ---
  LV Path                /dev/vg1s1/vm-100-disk-0
  LV Name                vm-100-disk-0
  VG Name                vg1s1
  LV UUID                ND69Nn-c5qH-s7Bk-c2U1-AT36-Gawy-ykvf7f
  LV Write Access        read/write
  LV Creation host, time pve1, 2022-10-25 22:18:19 +0200
  LV Pool name           pool1s1
  LV Status              available
  # open                 1
  LV Size                3.00 GiB
  Mapped size            46.92%
  Current LE             768
  Segments               1
  Allocation             inherit
  Read ahead sectors     auto
  - currently set to     2048
  Block device           253:6
  
  --- Logical volume ---
  LV Path                /dev/vg1s1/vm-100-disk-1
  LV Name                vm-100-disk-1
  VG Name                vg1s1
  LV UUID                joBTpP-uJEe-3yXD-xTVP-rZ4s-1v2q-NHRIHr
  LV Write Access        read/write
  LV Creation host, time pve1, 2022-10-25 22:18:20 +0200
  LV Pool name           pool1s1
  LV Status              available
  # open                 1
  LV Size                200.00 GiB
  Mapped size            2.09%
  Current LE             51200
  Segments               1
  Allocation             inherit
  Read ahead sectors     auto
  - currently set to     2048
  Block device           253:7
  
  --- Logical volume ---
  LV Path                /dev/vg1s1/vm-100-disk-2
  LV Name                vm-100-disk-2
  VG Name                vg1s1
  LV UUID                sznZbK-5GlU-tXdY-gNgW-nFoz-uLpS-EUOoux
  LV Write Access        read/write
  LV Creation host, time pve1, 2022-10-25 22:18:23 +0200
  LV Pool name           pool1s1
  LV Status              available
  # open                 1
  LV Size                600.00 GiB
  Mapped size            1.76%
  Current LE             153600
  Segments               1
  Allocation             inherit
  Read ahead sectors     auto
  - currently set to     2048
  Block device           253:16
  
  --- Logical volume ---
  LV Path                /dev/vg1s1/vm-100-disk-3
  LV Name                vm-100-disk-3
  VG Name                vg1s1
  LV UUID                XRFKMJ-6PS3-V20b-avMN-ImtK-cDGf-EbfCNb
  LV Write Access        read/write
  LV Creation host, time pve1, 2022-10-25 22:18:29 +0200
  LV Pool name           pool1s1
  LV Status              available
  # open                 1
  LV Size                400.00 GiB
  Mapped size            1.84%
  Current LE             102400
  Segments               1
  Allocation             inherit
  Read ahead sectors     auto
  - currently set to     2048
  Block device           253:17

So "LV Creation host time" was last night for all the disks. Does that really mean that I accidentally deleted the disk-1 .. disk-3 and all its content with it and then re-created it with the restore process?

Or - I would prefer that so much - did I make the disks content invisible to the guest/CT while playing around with GID/UID mapping?

Please, tell me it's the latter and how I can fix it! I just told my wife she can totally rely on our new backup system that is so much better than the old Apple TimeCapsule...

Thanks!

 

[Dunuin]

See my feature request here: https://forum.proxmox.com/threads/feature-request-advanced-restore-options-in-gui.109707/

When excluding disks from backups, PVE will wipe the excluded disks when doing a restore.

 

[degudejung]

thx @Dunuin I was afraid someone would say that.

But while we're at it... How would I do it right next time?

Again, in my logic, the disk/volume lives on its own and must not die only because it happened to be connected to a VM/CT that got removed. I thought that was the whole meaning of VMs/CTs - that they are basically disposables, while your data persists on volumes that are NOT rooted inside that machine. It works fine with the ZFS datasets I mount into the same CT. I just don't want to use ZFS on an SSD (single drive, consumer grade, not too much RAM available...) but would rather stick with a simpler solution with less overhead, like a LVM-thin LV.

So, I am forced to use ZFS or even bind mount a directory to to VM/CT? That can't be it, right?