Yes , we have had object-map enabled .
rgds
-----Message d'origine-----
De : Jason Dillaman <jdillama(a)redhat.com>
Envoyé : mardi 5 janvier 2021 15:08
À : Wissem MIMOUNA <wissem.mimouna(a)fiducialcloud.fr>
Cc : ceph-users(a)ceph.io
Objet : Re: [ceph-users] Timeout ceph rbd-nbd mounted image
On Tue, Jan 5, 2021 at 9:01 AM Wissem MIMOUNA <wissem.mimouna(a)fiducialcloud.fr>
wrote:
Hi ,
Thank you for your feedback . It seems the error related to the fstrim run once a week (
default ) .
Do you have object-map enabled? If not, the FS will gladly send huge discard extents
which, if you have a large volume, could result in hundreds of thousands of ops to the
cluster. That's a great way to hang IO.
Do you have more information about the NBD/XFS memeory
pressure issues ?
See [1].
Thanks
-----Message d'origine-----
De : Jason Dillaman <jdillama(a)redhat.com> Envoyé : mardi 5 janvier
2021 14:42 À : Wissem MIMOUNA <wissem.mimouna(a)fiducialcloud.fr> Cc :
ceph-users(a)ceph.io Objet : Re: [ceph-users] Timeout ceph rbd-nbd
mounted image
You can try using the "--timeout X" optional for "rbd-nbd" to
increase the timeout. Some kernels treat the default as infinity, but there were some
>=4.9 kernels that switched behavior and started defaulting to 30 seconds. There is
also known issues with attempting to place XFS file systems on top of NBD due to memory
pressure issues.
On Tue, Jan 5, 2021 at 4:36 AM Wissem MIMOUNA <wissem.mimouna(a)fiducialcloud.fr>
wrote:
Hello ,
Looking for information about a timeout which occur once a week for a ceph rbd image
mounted on a machine using rbd-nbd (Linux Ubuntu machine).
The error found in 'dmseg' is below :
[798016.401469] block nbd0: Connection timed out [798016.401506]
block
nbd0: shutting down sockets
Many Thanks
_______________________________________________
ceph-users mailing list -- ceph-users(a)ceph.io To unsubscribe send an
email to ceph-users-leave(a)ceph.io
--
Jason
[1]
https://urldefense.proofpoint.com/v2/url?u=https-3A__tracker.ceph.com_issue…
--
Jason