[Gluster-users] 3.7.12 disaster

[Kaushal M]

On Thu, Jun 30, 2016 at 5:47 PM, Kevin Lemonnier <lemonnierk at ulrar.net> wrote:
>>
>> Replicated the problem with 3.7.12 *and* 3.8.0 :(
>>
>
> Yeah, I tried 3.8 when it came out too and I had to use the fuse mount point
> to get the VMs to work. I just assumed proxmox wasn't compatible yet with 3.8 (since
> the menu were a bit wonky anyway) but I guess it was the same bug.
>

I was able to reproduce the hang as well against 3.7.12.

I tested by installing the pve-qemu-kvm package from the Proxmox
repositories in a Debain Jessie container, as the default Debian qemu
packages don't link with glusterfs.
I used the 3.7.11 and 3.7.12 gluster repos from download.gluster.org.

I tried to create an image on a simple 1 brick gluster volume using qemu-img.
The qemu-img command succeeded against a 3.7.11 volume, but hung
against 3.7.12 to finally timeout and fail after ping-timeout.

We can at-least be happy that this issue isn't due to any bugs in AFR.

I was testing this with Raghavendra, and we are wondering if this is
probably a result of changes to libglusterfs and libgfapi that have
been introduced in 3.7.12 and 3.8.
Any app linking with libgfapi also needs to link with libglusterfs.
While we have some sort of versioning for libgfapi, we don't have any
for libglusterfs.
This has caused problems before (I cannot find any links for this
right now though).

The pve-qemu-kvm package was last built or updated in January this
year[1]. And I think it was built against glusterfs-3.5.2, which is
the latest version of glusterfs in the proxmox sources [2].
Maybe the pve-qemu-kvm package needs a rebuild.

We'll continue to try to figure out what the actual issue is though.

~kaushal

> --
> Kevin Lemonnier
> PGP Fingerprint : 89A5 2283 04A0 E6E9 0111
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-users