Re: [3.2-rc2] loop device balance_dirty_pages_nr throttling hang

[Dave Chinner <david@fromorbit.com>]

On Tue, Nov 22, 2011 at 08:17:07AM -0500, Theodore Tso wrote:
> 
> On Nov 22, 2011, at 5:49 AM, Dave Chinner wrote:
> 
> > But, I just noticed that the discard that mkfs.ext4 will result
> > in all the extents being discarded in the underlying image file
> > (due to the fact the loopback device now supports hole
> > punching), so the previous state of the image file is getting
> > trashed by the mkfs.ext4 execution, too. I think I already had a
> > ext4 filesystem in some state before I started running this
> > manual prealloc test....
> 
> Which version of mkfs.ext4 are you using?   We've disabled the
> discard by default (unless configured in via a command-line option
> or a /etc/mke2fs.conf setting since no distribution apparently
> wants to be responsible for running supplying a command that
> causes a crap SSD to turning into a brick; personally, I'd blame
> the manufacturer of the crap SSD, but….)

Whatever I've got installed on my test machines. The machien I
reproduced the problem originally on with sparse files is runningi
mke2fs 1.42-WIP (16-Oct-2011) which is the latest in Debian
unstable, the other that is using a real 17TB array is running
mke2fs 1.42-WIP (02-Jul-2011). I saw the discard occurring

BTW, on a 5PB sparse image file I get this error from the latest
mke2fs above:

$ ls -lh /mnt/scratch/scratch.img
-rw------- 1 root root 4.9P Nov 23 06:50 /mnt/scratch/scratch.img
$ sudo mkfs.ext4 /dev/loop0
mke2fs 1.42-WIP (16-Oct-2011)
/dev/loop0: Cannot create filesystem with requested number of inodes while setting up superblock
$

This version does emit that it is discarding blocks:

$ sudo mkfs.ext4 /dev/loop0
mke2fs 1.42-WIP (16-Oct-2011)
Discarding device blocks: done
Filesystem label=
.....


Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com