df output wrong ?

df output wrong ?

[Leen Besselink]

Hi,

Just noticed something about df -h output I didn't expect.

I'm sorry if this has already been answered or is a know problem.

Just see the test-part below to see what I did.

____ setup ____

- installed Fedora 16 Alpha on a KVM-VM with one 8GB HD-image.
- did a yum update
( - got a kernel update, which broke grub setup, worked around it:
kernel I use is: 3.0.0-1.fc16.x86_64)
- installed btrfs-progs
- created a directory /test
- created 2 files with dd from /dev/zero with the size of 2GB in /test
- created 2 loopback devices of the files
- created a btrfs filesystem with RAID 1 data and RAID1 meta-data
- created directory /test/mnt

____ test _____

- mounted the btrfs filesystem on /test/mnt
- created a file of 256MB with dd from /dev/zero in /test/mnt
- created a snapshot of /test/mnt in /test/mnt/snap1
- deleted 256MB file in /test/mnt
- deleted snapshot /test/mnt/snap1
- checked df -h, 256MB wasn't released. The filesystem does not look empty
- unmounted /test/mnt
- mounted /test/mnt
- checked df -h, 256MB was released. The filesystem does look empty

Hope this is helpful.

Have a nice day,
    Leen.

Re: df output wrong ?

[Hugo Mills]

[-- Attachment #1: Type: text/plain, Size: 1708 bytes --]

On Mon, Sep 05, 2011 at 11:50:43AM +0200, Leen Besselink wrote:
> Hi,
> 
> Just noticed something about df -h output I didn't expect.
> 
> I'm sorry if this has already been answered or is a know problem.
> 
> Just see the test-part below to see what I did.
> 
> ____ setup ____
> 
> - installed Fedora 16 Alpha on a KVM-VM with one 8GB HD-image.
> - did a yum update
> ( - got a kernel update, which broke grub setup, worked around it:
> kernel I use is: 3.0.0-1.fc16.x86_64)
> - installed btrfs-progs
> - created a directory /test
> - created 2 files with dd from /dev/zero with the size of 2GB in /test
> - created 2 loopback devices of the files
> - created a btrfs filesystem with RAID 1 data and RAID1 meta-data
> - created directory /test/mnt
> 
> ____ test _____
> 
> - mounted the btrfs filesystem on /test/mnt
> - created a file of 256MB with dd from /dev/zero in /test/mnt
> - created a snapshot of /test/mnt in /test/mnt/snap1
> - deleted 256MB file in /test/mnt
> - deleted snapshot /test/mnt/snap1
> - checked df -h, 256MB wasn't released. The filesystem does not look empty
> - unmounted /test/mnt
> - mounted /test/mnt
> - checked df -h, 256MB was released. The filesystem does look empty
> 
> Hope this is helpful.

   Snapshot/subvolume deletion is asynchronous -- it can take a while
to happen. Try waiting a minute or two after the snapshot deletion and
before looking at df -h.

   Hugo.

-- 
=== Hugo Mills: hugo@... carfax.org.uk | darksatanic.net | lug.org.uk ===
  PGP key: 515C238D from wwwkeys.eu.pgp.net or http://www.carfax.org.uk
      --- If it's December 1941 in Casablanca,  what time is it ---      
                              in New York?                               

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 190 bytes --]

Re: df output wrong ?

[Leen Besselink]

On 09/05/2011 12:19 PM, Hugo Mills wrote:
>
>    Snapshot/subvolume deletion is asynchronous -- it can take a while
> to happen. Try waiting a minute or two after the snapshot deletion and
> before looking at df -h.
>
>    Hugo.
>

As I didn't see any disc activity, I thought it was done.

I just tried it again, it does seem to work the way you mentioned.

Thanks.

Just a quick question: would 'btrfs filesystem sync /path' help speed
that up ?

Like for example if a sys-admin sees that he's running out of space and
needs
to remove some old snapshots immediately.

Or when an automated backup first removes old snapshots to create room
for a new backup just before starting that backup.