Disk full during delayed allocation

Disk full during delayed allocation

[Jan Engelhardt]

Hi,


on space-constrained filesystems, I noticed that extracting lots of 
files [about 10k] bumps the Used count up quickly leading to a disk full 
unless the extraction process (rpm here) is halted, synced, and then 
continued. I believe this is fully within XFS's standard behavior, but I 
would like to learn more how exactly this can happen. My guess is that 
this is due to the "dynamic journal/log" size XFS employs - on a 
filesystem just mkfs'ed, about 4256KB (for a volume of 128MB) are used, 
compared to e.g. reiser3 where the full 32MB for the journal are used 
(according to df) right from the start.
Is this so?

# df; killall -CONT rpm; sleep 1; killall -STOP rpm; df; sync; df;
Filesystem           1K-blocks      Used Available Use% Mounted on
/lo/src.fs               93504     29396     64108  32% /usr/src
/lo/src.fs               93504     49072     44432  53% /usr/src
/lo/src.fs               93504     35632     57872  39% /usr/src

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

Re: Disk full during delayed allocation

[Dave Chinner]

On Tue, Dec 02, 2008 at 09:02:25PM +0100, Jan Engelhardt wrote:
> Hi,
> 
> 
> on space-constrained filesystems, I noticed that extracting lots of 
> files [about 10k] bumps the Used count up quickly leading to a disk full 
> unless the extraction process (rpm here) is halted, synced, and then 
> continued. I believe this is fully within XFS's standard behavior, but I 
> would like to learn more how exactly this can happen.

Outstanding delayed allocation has reservations for metadata blocks
that *may* be required to do the delayed allocation. most delalloc's
don't require metadata blocks to be allocated and hence when the
data is sync'd the reservation is freed and there is "magically"
more space available.

> My guess is that 
> this is due to the "dynamic journal/log" size XFS employs

It is not dynamic. The XFS journal size is fixed at mkfs time - it's
just that it is sized according to the size of the filesystem being
made....

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

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