[ovirt-users] oVirt + gluster + selfhosted + bonding

[Simone Tiraboschi]

On Tue, Sep 15, 2015 at 9:58 PM, Joachim Tingvold <joachim at tingvold.com>
wrote:

> Hi,
>
> First-time user of oVirt, so bear with me.
>
> Trying to get redundant oVirt + gluster set up. Have four hosts;
>
>   gluster1 (CentOS7)
>   gluster2 (CentOS7)
>   ovirt1 (CentOS7)
>   ovirt2 (CentOS7)
>
> Using replica 3 volume with arbiter node (new in 3.7.0). Got that part up
> and running (using ovirt1 as the arbiter node), and it works fine.
>
> Initial goal (before reading up on both gluster and oVirt) was to have
> everything v6-only, but found out quickly enough that we had to scratch
> that plan for now (I see that there are some activity on both gluster and
> oVirt on this, which is nice).
>
> Anyways. We wanted to use the "self hosted engine gluster"-feature (which,
> by the looks of it, is only present in 3.6). We installed 3.6b4
> (3.6.0.1-0.1.20150821.gitc8ddcd8.el7.centos).
>
> I already had the network set up (couldn't find any specifics on this in
> the somewhat lacking oVirt-documentation?), something along these lines;
>
>  * eth0 + eth1, bonded in bond0 (LACP)
>  * vlan110 on top of bond0: v6-only for mgmt of host
>  * vlan111 on top of bond0: v4 for gluster + ovirt
>
> We then ran the 'hosted-engine --deploy' command, filling out the
> information as best as we could (some of these options seemed to lack
> documentation, or at least we had trouble finding it). The end-result was
> like this[1].
>
> Accepting this, we suddenly found ourselves without connectivity to the
> host. Logged in via KVM, and this[2] was the last part of the log.
>
> All of the interfaces we had before (bond0, vlan110, vlan111) was "wiped
> clean" for it's configuration, and VDSM seems to have taken control on that
> part (however, since the script failed, we seem to have ended up in some
> kind of "limbo mode"). Rebooting didn't help bring things up again, and
> we're currently looking into manually configuring things via VDSM.
>


Hi Joachim,
unfortunately you hit this one:
https://bugzilla.redhat.com/1263311
The latest VDSM build sometimes fails setting up the management network and
so the issue you described.
It's not really systematic, we are seeing it on our CI env: some execution
correctly work while others, with the same code and the same parameters,
fail.
So, till we solve it, just clean up VDSM network conf and retry deploying
hosted-engine.
Feel free to add any relevant findings on that bug.


> Thought I'd post here meanwhile, seeing if we've missed something obvious,
> or if oVirt should've handled this any different?
>
> If relevant, the content of the answer-file referenced in [2] can be found
> here[3].
>
> [1] <http://files.jocke.no/b/dump_2015-09-08_22.14.21.png>
> [2] <http://files.jocke.no/b/dump_2015-09-08_22.06.06.png>
> [3] <http://files.jocke.no/b/answers-20150908215613.conf>
>
> --
> Joachim
> _______________________________________________
> Users mailing list
> Users at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20150916/b77f0d45/attachment-0001.html>