[Gluster-users] File size in brick

[kenji kondo]

Hello Brian,

Thank you very much for your advices.
Unfortunately, my gluster version 3.2.7 doesn't support the option.
But I'm planning to update the version in near future.
So is there recommended version although I planed 3.3.2?

Best regard,
Kondo

2013/06/27 1:39、Brian Foster <bfoster at redhat.com> のメッセージ:

> On 06/26/2013 10:55 AM, kenji kondo wrote:
>> Hello everyone,
>>
>> I'm using a stripe volume with 9 bricks. I have a question about file size
>> in bricks.
>> After copy a file of 700GB into the stripe volume, I checked the file size
>> by using "du" and/or "ls -ls" command.
>> But file size increase up to about 5TB. Also I checked the file size in
>> brick. Those file sizes was about 5TB / 9 brick for each brick.
>> In my understanding, the sparse file are created by glusterfsd with XFS
>> file system. So I expected that the actual file size shall be 700 GB / 9
>> because file in brick has any holes according to 'stripe-block-size'. But
>> it seems not correct.
>>
>> This fact is not happy for me because actual available total size decrease.
>> But I feel something wrong for this fact.
>> Is this stripe volume normally working? Or I mistook something about
>> setting?
>
> It's most likely aggressive speculative preallocation in XFS that is
> filling up the space that is expected to remain unallocated. Independent
> workarounds have been made in recent versions of gluster and XFS to
> better deal with this issue. You can try to enable the
> cluster.stripe-coalesce option on your gluster volume to avoid this problem.
>
> Brian
>
>> Best regards,
>> K. Kondo
>>
>>
>>
>> _______________________________________________
>> Gluster-users mailing list
>> Gluster-users at gluster.org
>> http://supercolony.gluster.org/mailman/listinfo/gluster-users
>