Slightly Urgent: XFS No Space Left On Device
Dave Hall
kdhall at binghamton.edu
Thu Apr 2 09:32:54 CDT 2015
Thanks for the help. Rookie error. I didn't set these mount options,
but I see that this option is set for all of the other XFS volumes I have.
I am wondering why XFS would default this way though. Seems like
heuristically you could assume that a large volume on a 64-bit OS would
need 64-bit inodes. At least perhaps put out a message from mkfs.xfs
suggesting the use of inode64 on the mount command?
Thanks.
-Dave
Dave Hall
Binghamton University
kdhall at binghamton.edu
607-760-2328 (Cell)
607-777-4641 (Office)
On 04/01/2015 08:12 PM, Dave Chinner wrote:
> On Wed, Apr 01, 2015 at 03:53:28PM -0400, Dave Hall wrote:
>
>> Please pardon the 'top-post', but here is the additional information
>> requested:
>>
>> This is a Dell R720xd dual 8-core Xeon system with 128GB RAM. The
>> RAID controller is Dell PERC H710 Mini with 12 2TB disks in RAID6.
>>
>> The OS is Debian 6 with kernel 3.2.0-0.bpo.4-amd64 #1 SMP Debian
>> 3.2.65-1+deb7u2~bpo60+1 x86_64.
>>
> So defaults to inode32 allocation....
>
>
>> From /proc/mounts:
>>
>> /dev/sdb1 /data xfs
>> rw,noexec,noatime,attr2,delaylog,allocsize=64k,logbsize=64k,sunit=128,swidth=1280,usrquota,prjquota
>> 0 0
>>
> ... and inode64 is not in the mount options.....
>
>
>> The output from xfs_info was previously included, but is repeated here:
>>
>> # xfs_info /data
>> meta-data=/dev/sdb1 isize=256 agcount=19,agsize=268435440 blks
>>
> Inode allocation requires contiguous free space of 16k aligned to 8k
> boundaries to allocate new inode chunks. Also, 1TB AGs, so with
> inode32, inodes can only be allocated in AG 0.
>
>
>> Here are the more extensive freesp outputs for each of the 19 AGs:
>>
>> # xfs_db -r /dev/sdb1 -c 'freesp -s -a0'
>> from to extents blocks pct
>> 1 1 747 747 19.68
>> 2 3 1045 2496 65.77
>> 4 7 138 552 14.55
>> total free extents 1930
>> total free blocks 3795
>> average free extent size 1.96632
>>
> And that says you have no correctly aligned free 16k extents that
> can be allocated in AG 0. i.e. no more inodes can be allocated, and
> that's where the ENOSPC is coming from.
>
> Unmount, add the inode64 mount option, and you'll be able to
> allocate inodes again as they will be allowed to be allocated in
> any AG, not just AG 0.
>
> Cheers,
>
> Dave.
>
More information about the xfs
mailing list