CentOS - Jun 2012 - centos 6.2 xfs + nfs space allocation

Centos 6.2 system with xfs filesystem.
I'm sharing this filesystem using nfs.

When I create a 10 gigabyte test file from a nfs client system :

dd if=/dev/zero of=10Gtest bs=1M count=10000
10000+0 records in
10000+0 records out
10485760000 bytes (10 GB) copied, 74.827 s, 140 MB/s

Output from 'ls -al ; du' during this test :

-rw-r--r-- 1 root root 429170688 Jun  8 10:13 10Gtest
654456  10Gtest

-rw-r--r-- 1 root root 1756831744 Jun  8 10:13 10Gtest
2230720 10Gtest

-rw-r--r--   1 root root 2492145664 Jun  8 10:11 10Gtest
4348288 10Gtest

-rw-r--r--   1 root root 4686782464 Jun  8 10:11 10Gtest
8542592 10Gtest

-rw-r--r-- 1 root root 10485760000 Jun  8 10:12 10Gtest
16943552        10Gtest

The file is using almost double the size ?
Even after a few days the file is showing the same disk usage.
Only umounting and remounting the filesystem fixes the problem.
When I do the same test on an ext4 filesystem no issues. (same 
server/client)

Same issue on 2 centos 6.2 servers.
I also tried to reproduce the issue on an rhel 5.8 system with xfs, but 
here the disk usage is ok.
(I don't have a rhel6 systems with the xfs addon subscription)

Stephan

What kernel are you using ?
Are you using inode64 mount option on the Cents server itself for XFS ?

What OS was the NFS client running ? 32bit ?

Just asking, as there seems to be problem with xfs on kernel 2.6.27 when 
using inode64 mount options regarding
32bit nfs....









Stephan van Hienen wrote:
> Centos 6.2 system with xfs filesystem.
> I'm sharing this filesystem using nfs.
>
> When I create a 10 gigabyte test file from a nfs client system :
>
> dd if=/dev/zero of=10Gtest bs=1M count=10000
> 10000+0 records in
> 10000+0 records out
> 10485760000 bytes (10 GB) copied, 74.827 s, 140 MB/s
>
> Output from 'ls -al ; du' during this test :
>
> -rw-r--r-- 1 root root 429170688 Jun  8 10:13 10Gtest
> 654456  10Gtest
>
> -rw-r--r-- 1 root root 1756831744 Jun  8 10:13 10Gtest
> 2230720 10Gtest
>
> -rw-r--r--   1 root root 2492145664 Jun  8 10:11 10Gtest
> 4348288 10Gtest
>
> -rw-r--r--   1 root root 4686782464 Jun  8 10:11 10Gtest
> 8542592 10Gtest
>
> -rw-r--r-- 1 root root 10485760000 Jun  8 10:12 10Gtest
> 16943552        10Gtest
>
> The file is using almost double the size ?
> Even after a few days the file is showing the same disk usage.
> Only umounting and remounting the filesystem fixes the problem.
> When I do the same test on an ext4 filesystem no issues. (same
> server/client)
>
> Same issue on 2 centos 6.2 servers.
> I also tried to reproduce the issue on an rhel 5.8 system with xfs, but
> here the disk usage is ok.
> (I don't have a rhel6 systems with the xfs addon subscription)
>
> Stephan
> _______________________________________________
> CentOS mailing list
> CentOS at centos.org
> http://lists.centos.org/mailman/listinfo/centos

On Mon, 11 Jun 2012, Svavar ?rn Eysteinsson wrote:
>
> What kernel are you using ?
Latest centos 6.2 kernel:
2.6.32-220.17.1.el6.x86_64
> Are you using inode64 mount option on the Cents server itself for XFS ?
fstab for this filesystem :

/dev/mapper/VolGroup01-LogVol00 /raid           xfs     defaults        1 2

/etc/exports for this filesystem :

/raid           192.168.178.5(rw,no_root_squash)

and on the client system /etc/fstab :

server:/raid              /raid                   nfs 
soft,rsize=32768,wsize=32768,vers=3
>
> What OS was the NFS client running ? 32bit ?
Also centos 6.2 x64 with kernel 2.6.32-220.17.1.el6.x86_64.

I also have this problem with a nfs client running 2.6.18-7.4-dm8000 
(dreambox satelite receiver)
And Centos 5.8 x64 with kernel 2.6.18-308.8.1.el5

Stephan

Stephan van Hienen wrote:
> The file is using almost double the size ?
> Even after a few days the file is showing the same disk usage.
> Only umounting and remounting the filesystem fixes the problem.
> When I do the same test on an ext4 filesystem no issues. (same 
> server/client)
What does:

  /usr/sbin/xfs_bmap -pl 10Gtest

output (when run on the server)?

You can also 'recover' the lost space by running (as root on the
server):

  /usr/sbin/xfs_fsr 10Gtest

James Pearson

P.S. you can also see the allocated size of a file using the -s option to ls