On Mon, Mar 8, 2021 at 8:15 AM Alex McWhirter <alex@triadic.us> wrote:

I apologize for the spam, this seems to be a long standing gluster issue. https://github.com/gluster/glusterfs/issues/597

sharding does not support SEEK_DATA/SEEK_HOLE, with a preallocated image you likely never see this issue as there are no holes. However with a sparse image, that's very much not the case. I'm not sure when qemu-img changed to use these syscalls, as this is not something i experience on 4.3 / CentOS 7.

I'd be interested if anyone else can replicate this image copy behavior using raw sparse (thin provision) disks as the source on a gluster volume with sharding enabled, on oVirt 4.4.4+ (possibly earlier is also affected)

If this is something current qemu-img cannot handle, i don't think supporting sparse disks on sharded gluster volumes is wise.




I only remember for sure that with 4.4.0 (and perhaps up to 4.4.2) with single host HCI with Gluster I couldn't use sharding in a lab where I had to create, destroy and recreate many times an Openshift environment composed by many VMs and having thin disks was a must to improve times.
I sent several messages regarding this to the list but no active resolution and was forced to disable sharding.

I also wrote to Gluster guys.

I can try to search and send the links.

Gianluca