Weird behaviour of mkfs.xfs

Weird behaviour of mkfs.xfs

[Emmanuel Florac]

Hi all,
I'm testing NVMe storage, and it works fine so far. However here's
something bizarre: I've made a RAID-0 array of 2 NVMe drives with
mdadm, and mkfs.xfs /dev/md0 starts an endless, full speed write of the
array:

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           0,21    0,00    0,17   12,39    0,00   87,24

Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s    wMB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sda               0,00     0,00    0,00    0,00     0,00     0,00     0,00     0,00    0,00    0,00    0,00   0,00   0,00
nvme0n1           0,00     0,00    0,00 5152,33     0,00  1288,03   511,98  1019,53  197,87    0,00  197,87   0,19 100,00
nvme1n1           0,00     0,00    0,00 5154,67     0,00  1288,67   512,00    50,62    9,82    0,00    9,82   0,19 100,00
dm-0              0,00     0,00    0,00    0,00     0,00     0,00     0,00     0,00    0,00    0,00    0,00   0,00   0,00
md0               0,00     0,00    0,00 5152,00     0,00  2576,00  1024,00     0,00    0,00    0,00    0,00   0,00   0,00


I'm running a plain vanilla 4.4.6 kernel; the xfs_progs are only
slightly older, being version 4.2.

Nothing will stop the crazy mkfs.xfs, but  a hard reboot. Any idea?

-- 
------------------------------------------------------------------------
Emmanuel Florac     |   Direction technique
                    |   Intellique
                    |	<eflorac@intellique.com>
                    |   +33 1 78 94 84 02
------------------------------------------------------------------------

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

Re: Weird behaviour of mkfs.xfs

[Emmanuel Florac]

Le Sat, 26 Mar 2016 14:50:37 +0100
Emmanuel Florac <eflorac@intellique.com> écrivait:

> I'm running a plain vanilla 4.4.6 kernel; the xfs_progs are only
> slightly older, being version 4.2.
> 
> Nothing will stop the crazy mkfs.xfs, but  a hard reboot. Any idea?
> 

Actually I was too impatient; it finally ended avec 30 minutes of
burning bits to the flash. I don't understand the behaviour, though.
I'm used to mkfs.xfs making its magic extremely quickly, even on
humongous devices. Here it's a very fast array of only 3.2 TB...

-- 
------------------------------------------------------------------------
Emmanuel Florac     |   Direction technique
                    |   Intellique
                    |	<eflorac@intellique.com>
                    |   +33 1 78 94 84 02
------------------------------------------------------------------------

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

Re: Weird behaviour of mkfs.xfs

[Christoph Hellwig]

On Sat, Mar 26, 2016 at 03:46:19PM +0100, Emmanuel Florac wrote:
> Actually I was too impatient; it finally ended avec 30 minutes of
> burning bits to the flash. I don't understand the behaviour, though.
> I'm used to mkfs.xfs making its magic extremely quickly, even on
> humongous devices. Here it's a very fast array of only 3.2 TB...

Trey doing a mkfs.xfs -K, without that it diascards the whole device.
I've seen some NVMe device misbehaving under discard storms, up to
the point of resetting the controller..

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

Re: Weird behaviour of mkfs.xfs

[Emmanuel Florac]

Le Sat, 26 Mar 2016 09:36:42 -0700
Christoph Hellwig <hch@infradead.org> écrivait:

> On Sat, Mar 26, 2016 at 03:46:19PM +0100, Emmanuel Florac wrote:
> > Actually I was too impatient; it finally ended avec 30 minutes of
> > burning bits to the flash. I don't understand the behaviour, though.
> > I'm used to mkfs.xfs making its magic extremely quickly, even on
> > humongous devices. Here it's a very fast array of only 3.2 TB...  
> 
> Trey doing a mkfs.xfs -K, without that it diascards the whole device.
> I've seen some NVMe device misbehaving under discard storms, up to
> the point of resetting the controller..

Ah, the discard is a nice touch, good to know :) It doesn't seem to be
mentioned in the man  page (except for the -K option), it would be a
good information to know.

-- 
------------------------------------------------------------------------
Emmanuel Florac     |   Direction technique
                    |   Intellique
                    |	<eflorac@intellique.com>
                    |   +33 1 78 94 84 02
------------------------------------------------------------------------

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