BUG - ZFS with virtual sub client - Manual incorrect

1
  1. Virtual subclients names are enclosed within square brackets ‘’
  2. ZFS does not permit square brackets in folder/directory names

This is a known issue that has yet to be fixed.

What this means is; You cannot do different backups (e.g. different times or different paths) to the same client if your server has been configured for ZFS.

Thus; the current manual is incorrect in implying that ZFS is a fully working option. It should (at a minimum) be updated to show the caveats.

I am retired and not a programmer in the C-plus-plus language. However, I would be more than happy to write a plain english caveat - if Uroni were to ask.

I discovered this problem over several days as I tried to move from BTRFS - which was proving unreliable for Urbackup (bit rot creeps in and I can’t afford ECC ram system). BTRFS is fine for semi-static data with regular scrubs on a raid1 FS as the scrubs can usually correct the errors.

I would also submit that the documentation for setting up ZFS with docker on a linux (OMV7) system is unclear/confusing. In that; UID must be zero (root) as well as SYS_ADMIN be set and datasets for images and files must exist.
As above - I would happily assist in updating documentation.

My analysis of the ZFS/sub client problem is:- it has been around for about 8 years or so. It is very difficult to solve. Uroni is aware of it and quite frustrated by it.

I fear that I will have to seek an alternative backup system as the space saving of BTRFS (which I love) just isnt reliable with the way the urbackup server works.

I will continue to follow this forum and contribute anything I can.

I am somewhat of a sedulous person and am averse to giving up.
Sadly, it’s too late for me to learn a couple of new programming languages.

2

I could take this on if you’d like. My idea is to keep the folder names as they are and only change the ZFS dataset name. The snapshot helper would escape [ and ] in the dataset name and create the dataset with an explicit mountpoint, so the backup path stays the same. Existing setups without sub-clients wouldn’t change. Martin, does that fit how you’d want it done, or would you rather change the folder naming on the server side?

3

First I wish to thank you for the kind offer.

I went with ecc ram and updated kernel and btrfs version. Changed to nvme pcie card that holds 4 X nvme cards and doesn’t require bifurcation. Expensive.

However, I have now had many months of error free, clean backups with monthly off-site good old spinning Barracudas.

For me, having got used to the way one can expand btrfs and increase the number of copies in a raid+ config (something like rc53? I cant remember) I will stick to the way I have it.

I am aware that there is quite a following for zfs so you may receive praise from that particular area of fans.

I give you praise not only for your proposal but also for your enthusiasm. I have read a number of your messages helping out people on this forum. Bravo.

David