[lustre-discuss] MDT size smaller than expected

Andreas Dilger adilger at whamcloud.com
Tue Jun 26 05:48:31 PDT 2018


On Jun 26, 2018, at 14:21, Steve Barnet <barnet at icecube.wisc.edu> wrote:
> 
> Hi Andreas,
> 
> 
> On 6/25/18 5:47 PM, Andreas Dilger wrote:
>> On Jun 25, 2018, at 20:39, Steve Barnet <barnet at icecube.wisc.edu> wrote:
>>> 
>>> Hi all,
>>> 
>>>  I'm setting up a new lustre filesystem with 2.10.4. Things are
>>> looking OK so far. However, I noticed that when I mount up
>>> my MDT, df reports a smaller size than I expect. The volume is
>>> 2.2TB, but the MDT reports 1.3:
>>> 
>>> 
>>> icecube-lfs6-mds-1 ~ # df -h -t lustre
>>> Filesystem                 Size  Used Avail Use% Mounted on
>>> /dev/mapper/md3420-1-vd-0  1.3T   94M  1.2T   1% /mnt/lustre/lfs6-mdt0000
>>> 
>>> 
>>> It appears that the host sees the correct size for the volume:
>>> 
>>> icecube-lfs6-mds-1 ~ # fdisk -l /dev/mapper/md3420-1-vd-0
>>> 
>>> Disk /dev/mapper/md3420-1-vd-0: 2388.7 GB, 2388672905216 bytes, 4665376768 sectors
>>> Units = sectors of 1 * 512 = 512 bytes
>>> Sector size (logical/physical): 512 bytes / 512 bytes
>>> I/O size (minimum/optimal): 512 bytes / 512 bytes
>>> 
>>> 
>>>  So I am a little confused. Seems to work OK, but I'd like
>>> to understand what might be going on there.
>> About half of the MDT is consumed by the inodes with the default formatting parameters,
>> so this won't ever show up as part of the free space in the filesystem.  The statfs()
>> interface is somewhat limited in what it can show, and the alternative is to show the
>> total blocks as 2.2TB, but there is 1.1TB of "Used" space, which would probably get
>> even more questions on the flip side "why is half of my MDT filesystem used, and how
>> do I get rid of that space usage".
> 
> 
> Thanks much. So just to make sure I'm clear when the
> inevitable question arises: the total size is 2.2TB,
> however, 1.1TB of that is inodes so that is subtracted
> from the 2.2TB and that's what we see in df.
> 
> That makes sense, and I can definitely see why that
> choice would be made. I just wanted to be sure I
> hadn't done something I might regret for the life of
> the filesystem. :-)

No, there is nothing wrong with the system.

---
Andreas Dilger
Principal Lustre Architect
Whamcloud






-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 235 bytes
Desc: Message signed with OpenPGP
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20180626/11422b52/attachment.sig>


More information about the lustre-discuss mailing list