NFS and pass discards\unmap question
I've upgraded to 4.1 release, it have great feature "Pass discards", that now can be used without vdsm hooks, After upgrade I've tested it with NFS 4.1 storage, exported from netapp, but unfortunately found out, that it's not working, after some investigation, I've found, that NFS implementation(even 4.1) in Centos 7 doesn't support sparse files and fallocate(FALLOC_FL_PUNCH_HOLE), that quemu uses for file storage, it was added only in kernel 3.18, and sparse files is also announced feature of upcoming NFS4.2, sparsify also not working on this data domains(runs, but nothing happens). This test also shows, that FALLOC_FL_PUNCH_HOLE not working, it was executed on centos ovirt host with mounted nfs share: # truncate -s 1024 test1 # fallocate -p -o 0 -l 1024 test1 fallocate: keep size mode (-n option) unsupported Is there any plans to backport this feature to node-ng, or centos? or we should wait for RHEL 8? NFS is more and more popular, so discards is VERY useful feature. I'm also planning to test fallocate on latest fedora with 4.x kernel and mounted nfs. Thanks for your work! --
On Thu, Feb 2, 2017 at 11:45 PM, Sergey Kulikov <serg_k@msm.ru> wrote:
I've upgraded to 4.1 release, it have great feature "Pass discards", that now can be used without vdsm hooks, After upgrade I've tested it with NFS 4.1 storage, exported from netapp, but unfortunately found out, that it's not working, after some investigation, I've found, that NFS implementation(even 4.1) in Centos 7 doesn't support sparse files and fallocate(FALLOC_FL_PUNCH_HOLE), that quemu uses for file storage, it was added only in kernel 3.18, and sparse files is also announced feature of upcoming NFS4.2, sparsify also not working on this data domains(runs, but nothing happens).
This test also shows, that FALLOC_FL_PUNCH_HOLE not working, it was executed on centos ovirt host with mounted nfs share: # truncate -s 1024 test1 # fallocate -p -o 0 -l 1024 test1 fallocate: keep size mode (-n option) unsupported
Is there any plans to backport this feature to node-ng, or centos? or we should wait for RHEL 8?
Interesting, I was under the impression it was fixed some time ago, for 7.2[1] (kernel-3.10.0-313.el7) Perhaps you are not mounted with 4.2? Y. [1] https://bugzilla.redhat.com/show_bug.cgi?id=1079385
NFS is more and more popular, so discards is VERY useful feature. I'm also planning to test fallocate on latest fedora with 4.x kernel and mounted nfs.
Thanks for your work!
--
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
Hm... maybe I need to set any options, is there any way to force ovirt to mount with this extension, or version 4.2 there is only 4.1 selection in "New Domain" menu. Current mount options: type nfs4 (rw,relatime,vers=4.1,rsize=65536,wsize=65536,namlen=255,soft,nosharecache,proto=tcp,port=0,timeo=600,retrans=6,sec=sys,local_lock=none) it should work only if forced option vers=4.2 ? I thought it's implemented as feature to older version, not 4.2, there is few info about this. -- Friday, February 3, 2017, 14:45:43:
On Thu, Feb 2, 2017 at 11:45 PM, Sergey Kulikov <serg_k@msm.ru> wrote:
I've upgraded to 4.1 release, it have great feature "Pass discards", that now can be used without vdsm hooks, After upgrade I've tested it with NFS 4.1 storage, exported from netapp, but unfortunately found out, that it's not working, after some investigation, I've found, that NFS implementation(even 4.1) in Centos 7 doesn't support sparse files and fallocate(FALLOC_FL_PUNCH_HOLE), that quemu uses for file storage, it was added only in kernel 3.18, and sparse files is also announced feature of upcoming NFS4.2, sparsify also not working on this data domains(runs, but nothing happens).
This test also shows, that FALLOC_FL_PUNCH_HOLE not working, it was executed on centos ovirt host with mounted nfs share: # truncate -s 1024 test1 # fallocate -p -o 0 -l 1024 test1 fallocate: keep size mode (-n option) unsupported
Is there any plans to backport this feature to node-ng, or centos? or we should wait for RHEL 8?
Interesting, I was under the impression it was fixed some time ago, for 7.2[1] (kernel-3.10.0-313.el7) Perhaps you are not mounted with 4.2?
Y. [1] https://bugzilla.redhat.com/show_bug.cgi?id=1079385 NFS is more and more popular, so discards is VERY useful feature. I'm also planning to test fallocate on latest fedora with 4.x kernel and mounted nfs.
Thanks for your work!
--
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
On Fri, Feb 3, 2017 at 2:29 PM, Sergey Kulikov <serg_k@msm.ru> wrote:
Hm... maybe I need to set any options, is there any way to force ovirt to mount with this extension, or version 4.2 there is only 4.1 selection in "New Domain" menu. Current mount options: type nfs4 (rw,relatime,vers=4.1,rsize=65536,wsize=65536,namlen=255,soft,nosharecache,proto=tcp,port=0,timeo=600,retrans=6,sec=sys,local_lock=none)
it should work only if forced option vers=4.2 ? I thought it's implemented as feature to older version, not 4.2, there is few info about this.
Looks like ovirt engine does not allow nfs version 4.2. We have this RFE: https://bugzilla.redhat.com/1406398 So practically, both sparsify and pass discard with NFS are useless in the current version. I think this should be fix for next 4.1 build. Nir
--
Friday, February 3, 2017, 14:45:43:
On Thu, Feb 2, 2017 at 11:45 PM, Sergey Kulikov <serg_k@msm.ru> wrote:
I've upgraded to 4.1 release, it have great feature "Pass discards", that now can be used without vdsm hooks, After upgrade I've tested it with NFS 4.1 storage, exported from netapp, but unfortunately found out, that it's not working, after some investigation, I've found, that NFS implementation(even 4.1) in Centos 7 doesn't support sparse files and fallocate(FALLOC_FL_PUNCH_HOLE), that quemu uses for file storage, it was added only in kernel 3.18, and sparse files is also announced feature of upcoming NFS4.2, sparsify also not working on this data domains(runs, but nothing happens).
This test also shows, that FALLOC_FL_PUNCH_HOLE not working, it was executed on centos ovirt host with mounted nfs share: # truncate -s 1024 test1 # fallocate -p -o 0 -l 1024 test1 fallocate: keep size mode (-n option) unsupported
Is there any plans to backport this feature to node-ng, or centos? or we should wait for RHEL 8?
Interesting, I was under the impression it was fixed some time ago, for 7.2[1] (kernel-3.10.0-313.el7) Perhaps you are not mounted with 4.2?
Y. [1] https://bugzilla.redhat.com/show_bug.cgi?id=1079385
NFS is more and more popular, so discards is VERY useful feature. I'm also planning to test fallocate on latest fedora with 4.x kernel and mounted nfs.
Thanks for your work!
--
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
Unfortunately I can't browse this bug: "You are not authorized to access bug #1079385." Can you email me details on ths bug? I think that's the reason I can't find this fix for rhel\centos in google) -- Friday, February 3, 2017, 14:45:43:
On Thu, Feb 2, 2017 at 11:45 PM, Sergey Kulikov <serg_k@msm.ru> wrote:
I've upgraded to 4.1 release, it have great feature "Pass discards", that now can be used without vdsm hooks, After upgrade I've tested it with NFS 4.1 storage, exported from netapp, but unfortunately found out, that it's not working, after some investigation, I've found, that NFS implementation(even 4.1) in Centos 7 doesn't support sparse files and fallocate(FALLOC_FL_PUNCH_HOLE), that quemu uses for file storage, it was added only in kernel 3.18, and sparse files is also announced feature of upcoming NFS4.2, sparsify also not working on this data domains(runs, but nothing happens).
This test also shows, that FALLOC_FL_PUNCH_HOLE not working, it was executed on centos ovirt host with mounted nfs share: # truncate -s 1024 test1 # fallocate -p -o 0 -l 1024 test1 fallocate: keep size mode (-n option) unsupported
Is there any plans to backport this feature to node-ng, or centos? or we should wait for RHEL 8?
Interesting, I was under the impression it was fixed some time ago, for 7.2[1] (kernel-3.10.0-313.el7) Perhaps you are not mounted with 4.2?
Y. [1] https://bugzilla.redhat.com/show_bug.cgi?id=1079385 NFS is more and more popular, so discards is VERY useful feature. I'm also planning to test fallocate on latest fedora with 4.x kernel and mounted nfs.
Thanks for your work!
--
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
participants (3)
-
Nir Soffer -
Sergey Kulikov -
Yaniv Kaul