On 08/12/2013 06:32 PM, Andrew Cathrow wrote:
----- Original Message -----
From: "Deepak C Shetty" deepakcs@linux.vnet.ibm.com To: vdsm-devel@lists.fedorahosted.org Sent: Monday, August 12, 2013 8:59:37 AM Subject: Re: [vdsm] How to handle qemu 1.3 dep for Gluster Storage Domain
On 08/12/2013 04:51 PM, Andrew Cathrow wrote:
----- Forwarded Message -----
From: "Itamar Heim" iheim@redhat.com To: "Sahina Bose" sabose@redhat.com Cc: "engine-devel" engine-devel@ovirt.org, "VDSM Project Development" vdsm-devel@lists.fedorahosted.org Sent: Wednesday, August 7, 2013 1:30:54 PM Subject: Re: [vdsm] How to handle qemu 1.3 dep for Gluster Storage Domain
On 08/07/2013 08:21 AM, Sahina Bose wrote:
[Adding engine-devel]
On 08/06/2013 10:48 AM, Deepak C Shetty wrote: > Hi All, > There were 2 learnings from BZ > https://bugzilla.redhat.com/show_bug.cgi?id=988299 > > 1) Gluster RPM deps were not proper in VDSM when using Gluster > Storage > Domain. This has been partly addressed > by the gluster-devel thread @ > http://lists.gnu.org/archive/html/gluster-devel/2013-08/msg00008.html > and will be fully addressed once Gluster folks ensure their > packaging > is friendly enuf for VDSM to consume > just the needed bits. Once that happens, i will be sending a > patch to > vdsm.spec.in to update the gluster > deps correctly. So this issue gets addressed in near term. > > 2) Gluster storage domain needs minimum libvirt 1.0.1 and qemu > 1.3. > > libvirt 1.0.1 has the support for representing gluster as a > network > block device and qemu 1.3 has the > native support for gluster block backend which supports > gluster://... > URI way of representing a gluster > based file (aka volume/vmdisk in VDSM case). Many distros > (incl. > centos 6.4 in the BZ) won't have qemu > 1.3 in their distro repos! How do we handle this dep in VDSM ? > > Do we disable gluster storage domain in oVirt engine if VDSM > reports > qemu < 1.3 as part of getCapabilities ? > or > Do we ensure qemu 1.3 is present in ovirt.repo assuming > ovirt.repo is > always present on VDSM hosts in which > case when VDSM gets installed, qemu 1.3 dep in vdsm.spec.in > will > install qemu 1.3 from the ovirt.repo > instead of the distro repo. This means vdsm.spec.in will have > qemu >= > 1.3 under Requires. > Is this possible to make this a conditional install? That is, only if Storage Domain = GlusterFS in the Data center, the bootstrapping of host will install the qemu 1.3 and dependencies.
(The question still remains as to where the qemu 1.3 rpms will be available)
RHEL6.5 (and so CentOS 6.5) will get backported libgfapi support so we shouldn't need to require qemu 1.3 just the appropriate qemu-kvm version from 6.5
So IIUC this means we don't do anything special in vdsm.spec.in to handle qemu 1.3 dep ? If so... what happens when User uses F17/F18 ( as an example) on the VDSM host.. their repos probably won't have qemu-kvm which has libgfapi support... how do we handle it. Do we just release-note it ?
For Fedora SPEC we'd need to handle use a >=1.3 dependency but for *EL6 it'd need to be >0.12-whaterver-6.5-has
I would love to hear how. I am waiting on some resolution for this, so that I can close the 3.3 blocker BZ
For Fedora if I put qemu-kvm >= 1.3 in vdsm.spec.in, then F17/F18 can't be used as a VDSM host, that may not be acceptable.
thanx, deepak