On Tue, Jan 12, 2016 at 10:45 PM, Markus Stockhausen <stockhausen@collogia.de> wrote:
>> Von: Yaniv Kaul [ykaul@redhat.com]
>> Gesendet: Dienstag, 12. Januar 2016 13:15
>> An: Markus Stockhausen
>> Cc: users@ovirt.org; Mike Hildebrandt
>> Betreff: Re: [ovirt-users] NFS IO timeout configuration
>>
>> On Tue, Jan 12, 2016 at 9:32 AM, Markus Stockhausen stockhausen@collogia.de> wrote:
>> Hi there,
>>
>> we got a nasty situation yesterday in our OVirt 3.5.6 environment.
>> We ran a LSM that failed during the cleanup operation. To be precise
>> when the process deleted an image on the source NFS storage.
>
> Can you share with us your NFS server details?
>Is the NFS connection healthy (can be seen with nfsstat)
>Generally, delete on NFS should be a pretty quick operation.
> Y.

Hi Yaniv,

we usually have no problems with our NFS server. From our observations we
only have issues when deleting files with many extents. This applies to all
OVirt images files. Several of them have more than 50.000 extents, a few
even more than 300.000.

> xfs_bmap 1cb5906f-65d8-4174-99b1-74f5b3cbc537
...
        52976: [629122144..629130335]: 10986198592..10986206783
        52977: [629130336..629130343]: 10986403456..10986403463
        52978: [629130344..629138535]: 10986206792..10986214983
        52979: [629138536..629138543]: 10986411656..10986411663
        52980: [629138544..629145471]: 10986214992..10986221919
        52981: [629145472..629145575]: 10809903560..10809903663
        52982: [629145576..629145599]: 10737615056..10737615079

Our XFS is mounted with:

/dev/mapper/vg00-lvdata on /var/nas4 type xfs (rw,noatime,nodiratime,allocsize=16m)

Why we use allocsize=16M? We once started with allocize=512MB. This
led to sparse files that did not save much bytes. Because a single byte written
resulted in a 512MB allocation. Thin allocation of these files resulted in long runtimes
for formatting disks inside the VMS. So we reduced to 16MB as a balanced config

This works quite well but not for remove operations.

Better ideas?

Sounds like an XFS issue more than NFS.
I've consulted with one of our XFS gurus - here's his reply:

For vm image files, users should set up extent size hints to define
the minimum extent allocation size in a file - allocsize does
nothing for random writes into sparse files. I typically use a hint
of 1MB for all my vm images....`

Y.
 

Markus