project quota: df/reservations

project quota: df/reservations

[Jerzy Borkowski]

Hello,

1) Fancy a project with quota set to 10GB, and filled
with 5GB of data. When I type

  df /path/to/project/tree

it shows 5GB free space. Now, the root user
fills up the disk (outside project tree), so the
XFS filesystem is really 100% full. The df command
still shows 5GB free space. I guess, this is
intentional. I propose to add some switch/flag
(either during mount or project creation),
which would tell XFS to report in statfs() calls
the free space as: min(project_free_space, filesystem_free_space).

2) is there any way to have space reservations for projects (like in ZFS) ?
For instance, a project with 10GB quota and 7GB of preallocated
space, so even root, writing outside the project tree, cannot
completely fill up the disk.

Thanks,

Jurek


_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: project quota: df/reservations

[Dave Chinner]

On Fri, Feb 04, 2011 at 01:25:53PM +0100, Jerzy Borkowski wrote:
> Hello,
> 
> 1) Fancy a project with quota set to 10GB, and filled
> with 5GB of data. When I type
> 
>   df /path/to/project/tree
> 
> it shows 5GB free space. Now, the root user
> fills up the disk (outside project tree), so the
> XFS filesystem is really 100% full. The df command
> still shows 5GB free space. I guess, this is
> intentional. I propose to add some switch/flag
> (either during mount or project creation),
> which would tell XFS to report in statfs() calls
> the free space as: min(project_free_space, filesystem_free_space).

If you just run "df" or "df /path/to/mtpt" rather than asking
specifically for the project tree path, you should see the real
filesystem usage.

Think of project quotas like thin provisioned filesystems - they can
still have space available when the underlying storage runs out and
it is up to the admin to manage this situation.

> 2) is there any way to have space reservations for projects (like in ZFS) ?
> For instance, a project with 10GB quota and 7GB of preallocated
> space, so even root, writing outside the project tree, cannot
> completely fill up the disk.

No. Quotas can only be used to limit space usage, not reserve it.
It could probably be done, but it's not a simple retrofit...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

inode64

[Jerzy Borkowski]

Hi,

If an 800GB XFS filesystem is mounted with inode64 option,
is it guaranteed that all inodes in the filesystem are in
the range 0-0xFFFFFFFF ?
Kernel is 3.17.x

Thanks,

Jurek

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: inode64

[Carlos Maiolino]

Hi,

If you use inode64, the inodes will be spread all along the disk. The inode
numbers are created based on its location in the disk, so, if you have a disk
which is larger than 1TB, you can have inode numbers larger than 32bit which you
mentioned.

If you want to keep inodes in the range you mentioned, you must need inode32
options, so no inode will be allocated beyond the 32 bits limit which you
mentioned.

But, since you have a 800GB filesystem, and not a filesystem larger than 1TB,
you can use inode64 and you will still have inodes in the 32bit range, but, if
in any circunstance you grow the filesystem, you'll have inodes being allocated
beyond that.

if you must keep inode allocations in the 32bit range, I suggest you using the
inode32 option, to avoid any future problems in case of a grow of the
filesystem.


On Tue, May 19, 2015 at 04:53:07PM +0200, Jerzy Borkowski wrote:
> Hi,
> 
> If an 800GB XFS filesystem is mounted with inode64 option,
> is it guaranteed that all inodes in the filesystem are in
> the range 0-0xFFFFFFFF ?
> Kernel is 3.17.x
> 
> Thanks,
> 
> Jurek
> 
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs

-- 
Carlos

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs