Sheepdog storage not thin

[athompso]

I've been migrating VMs from NFS to Sheepdog, and I'm suddenly noticing that some of the virtual disks are consuming 100% of their stated capacity, i.e. thick-provisioned, not thin.

Sheepdog is supposed to be able to thin-provision... and in fact does for some other disks - how do I "thin-ify" these images? (Naturally, they're the biggest ones.)

 

[spirit]

hi,
do you have migrate vm from nfs to sheepdog with proxmox move disk feature ?
if yes, online or offline ?

Online, it's not possible for qemu to detect zeroes on nfs, so the destination volume on sheepdog will not be sparse.
Offline, it should work.

If you want to reclaim space,you can configure your vm with virtio-scsi + discard, than use "fstrim" command in your vm.

 

[athompso]

Oh, shoot, yes online-vs-offline would explain the results I'm seeing. I wasn't aware of that limitation.
Some of the disks I can trim - and now have, to good effect - but others predate OSes with TRIM support (notably one particular Win2008R2 domain controller with a needlessly large disk). I haven't had any luck using any of the free tools that supposedly let me do TRIMs from a Win2k8 server, yet.
Thanks,
-Adam

 

[Angelov72]

Hi Athonso,

can you share your impressions on Sheepdog? I am really interested in this project, so need some advise.

Thanks

Angelo

 

[spirit]

Oh, shoot, yes online-vs-offline would explain the results I'm seeing. I wasn't aware of that limitation.
Some of the disks I can trim - and now have, to good effect - but others predate OSes with TRIM support (notably one particular Win2008R2 domain controller with a needlessly large disk). I haven't had any luck using any of the free tools that supposedly let me do TRIMs from a Win2k8 server, yet.
Thanks,
-Adam

yes for windows2008R2, trim is not working.
you can try a software like sdelete for example, it's filling a big file with zeros the delete it, for some storage, it's freeing space.

you can also try to move disk again offline if you can shut the vm.

 

[athompso]

Hi Athonso,

can you share your impressions on Sheepdog? I am really interested in this project, so need some advise.

Thanks

Angelo

Still not as stable as CEPH. I just tried rolling back on a snapshot, only to have sheepdog delete the base disk when the rollback failed. Not exactly a graceful failure mode...

Sheepdog also fails miserably at online cluster operations like altering the # of copies kept - the entire cluster *claims* to go into recovery mode, but all the QEMU guests start dropping/failing *all* I/O requests immediately.