BTRFS reporting incorrect disk usage

[jamielsharief]

LXD 4.0.5 with a custom partition formatted as BTRFS and this set as the storage pool.

I noticed that disk space usage was getting lost somewhere, you can reproduce like this.

Create an Ubuntu container, send a patch request to /instances/ubuntu setting a the disk size to 5GB

$ lxc info ubuntu
  Disk usage:
    root: 16.38kB

I then install mysql-server and apache2

Note on a LXD host with a ZFS storage pool the following commands returns usage as 1.25GB

$ lxc info ubuntu
  Disk usage:
    root: 519.21MB

I create a snapshot by sending a post request with the following payload

{
    "stateful": false,
    "name": "ubuntu-20210216-01"
} 

and disk usage drops to the size of the snapshot, I tested on a LXD host with ZFS and this does not happen.

  Disk usage:
    root: 9.37MB

I then got curious, and deleted the snapshot , the disk usage remains the same.

  Disk usage:
    root: 9.37MB

Other observations:

• LXD reports snapshot size for BTRFS storage as 10mb
• LXD reports snapshot size for ZFS storage 1.32mb (after 3 minutes)

[jamielsharief]

Note running du -h -s / within the container on both ZFS an BTRFS hosts return exactly the same results 1.2GB

[jamielsharief]

@stgraber just wondering if you have seen this? Taking a snapshot overwrites the disk usage size and i am not sure why disk space is different when installing the same apps.

[stgraber]

Yeah, I've had that issue in my queue for a little while now.

[tomponline]

Confirmed this is still an issue.

[tomponline]

OK so I figured it out, and am not sure if its actually a bug. This depends on your perspective of what is being asked with lxc info <instance> or more specifically lxc query /1.0/instances/c1?recursion=1.

The reason appears to be that the size show here is only showing the size consumed by the main volume (relative to its earlier snapshots).

[tomponline]

We can see the output from the underlying btrfs tool:

After initial launch:

sudo btrfs qgroup show -e -f  /var/lib/lxd/storage-pools/btrfs/containers/c1
qgroupid         rfer         excl     max_excl 
--------         ----         ----     -------- 
0/274       475.20MiB     16.00KiB         none 

After installing apache2 and stopping the instance:

sudo btrfs qgroup show -e -f  /var/lib/lxd/storage-pools/btrfs/containers/c1
qgroupid         rfer         excl     max_excl 
--------         ----         ----     -------- 
0/274       550.42MiB     82.80MiB         none 

After taking a snapshot:

sudo btrfs qgroup show -e -f  /var/lib/lxd/storage-pools/btrfs/containers/c1
qgroupid         rfer         excl     max_excl 
--------         ----         ----     -------- 
0/274       634.02MiB      9.27MiB         none 

sudo btrfs qgroup show -e -f  /var/lib/lxd/storage-pools/btrfs/containers-snapshots/c1/snap0
qgroupid         rfer         excl     max_excl 
--------         ----         ----     -------- 
0/275       634.02MiB      3.50MiB         none 

One thing that maybe a problem is that the snapshots appear to be accounted under a different qgroup ID. That doesn't seem right.