[Users] Re: disk use/empty files question

[Keith Keller]

Hello all, thanks for the quick responses!

On 2012-05-31, Kirill Kolyshkin <kolyshkin at gmail.com> wrote:
>
> Assuming you don't have DISK_QUOTA=no in global config (ie /etc/vz/vz.conf),

That's correct.

> the figures shown come from vzquota, and in order for vzquota to work correctly
> when you want to copy something to container you have to have it mounted (ie
> vzctl status 21 should show the word 'mounted' among others) and copy the
> data to VE_ROOT (ie /vz/root/21) but not to VE_PRIVATE (/vz/private/21).
>
> If you have already done it wrong (I assume you did *), you have to recalculate
> vzquota, the easiest way is to stop container, do vzquota drop 21 and
> start container
> again. This should fix your issue.

You are indeed correct that I originally copied data to /vz/private/21.
But when I attempted to drop the quota, it still reports 0 blocks used.
I wonder if Massimiliano's comment is relevant?

On 2012-05-31, Massimiliano
<massimiliano.sciabica at kiiama.com> wrote:
> Hi,
> I had a similar issue when I first tried to improve performance of VPS 
> with high number of small files.
> When a VPS reports 0% usage it is usually due to a not ext2 family 
> filesystem. What fs where you using?

I am using XFS.  The FAQ mentions that disk quotas do not work with XFS,
so perhaps that's why it isn't displaying quite right (see below).

> I believe this is because of file system crash and the fsck (or journal replay)
> which truncated your files. In other words, this is not directly
> related to what you have described above.

Perhaps--on boot, I didn't notice any unusual messages from fsck, though
I admit I wasn't paying an enormous amount of attention, and the logs
don't have anything interesting to report either.  Does OpenVZ do a lot
of caching of disk writes from within a container?  (It's obviously too
late now to see what xfs_repair thinks of the filesystem, but FWIW it
didn't find anything unusual.)

> Speaking of kernel crashes, it's nice to have some console logger installed,
> such as netconsole so whenever you have an oops you can report the bug.
> See http://wiki.openvz.org/Remote_console_setup

Yes, I just set this up after the first crash--silly oversight not to
have done it right away.  :)

>> # Disk quota parameters (in form of softlimit:hardlimit)
>> DISKSPACE="1000G:2000G"
>
> It looks like you have set disk quota values to more than your really
> have. Since this doesn't make sense
> my question is -- did you meant to disable disk space limit entirely?
> If yes, you can just have
> DISK_QUOTA=no in this config.

Well, I am not entirely sure what I want, to be honest.  If it's true
that having VE_ROOT and VE_PRIVATE on an XFS filesystem means disk
quotas don't work right, then perhaps I should either use ext3 (or
ext4?) on that filesystem, or disable disk quotas for all containers.

As an experiment on the latter, I set DISK_QUOTA=no in vz.conf, and now
I get:

# vzctl exec 21 df -h
Filesystem            Size  Used Avail Use% Mounted on
/dev/simfs            1.0T  332G  693G  33% /
none                   16G  4.0K   16G   1% /dev

But it would be convenient to have disk quotas.  Is there a preference
for ext3 or ext4 for the host filesystem?

--keith


-- 
kkeller at wombat.san-francisco.ca.us