Hardlinks accross Datasets?

TrueNAS General
,
25

@gigagames

On an Arch Linux system using a dummy “test” pool, with the following versions, I did a test.

  • Kernel: 6.6.52
  • Coreutils: 9.5
  • ZFS: 2.2.6

Between datasets, I copied a file without any special flags, simply with the cp command and a large 1 GiB non-compressible file comprised of random data.

cp /testpool/mydata/bigfile.dat /testpool/yourdata/

Here are the results. You’ll see that the command-line output is a more accurate representation of what’s going on.

zpool list -o name,size,capacity,alloc,free,bcloneratio,bcloneused,bclonesaved
NAME       SIZE    CAP  ALLOC   FREE  BCLONE_RATIO  BCLONE_USED  BCLONE_SAVED
testpool  3.75G    26%  1.00G  2.75G         2.00x           1G            1G


zfs list -o space
NAME               AVAIL   USED  USEDSNAP  USEDDS  USEDREFRESERV  USEDCHILD
testpool           2.62G  2.00G        0B     25K             0B      2.00G
testpool/mydata    2.62G  1.00G        0B   1.00G             0B         0B
testpool/yourdata  2.62G  1.00G        0B   1.00G             0B         0B

If you add the “USED” of both datasets, it equals 2 GiB, which is reported by the parent dataset. But, according to the pool’s properties, only 1 GiB of space is being used.

See? ZFS math is tricky to work with. :wink:

So if you want to truly know how much space is being used on the pool overall, then you should only rely on the zpool command. Don’t rely on the zfs command or any (parent) dataset properties. Don’t rely on the dashboard or GUI.

Why didn’t my test require --sparse=never? It could be a combination of the ZFS version + kernel version + Coreutils version, or perhaps because my test file is “friendly” with cp’s “sparseyness” heuristics.

1 Like