[Users] VM migrations failing

Dead Horse deadhorseconsulting at gmail.com
Mon Feb 4 16:38:16 UTC 2013


VDSM built from commit: c343e1833f7b6e5428dd90f14f7807dca1baa0b4 works
Current VDSM built from master does not work.

I could try spending some time trying to bisect and find out where the
breakage occurred I suppose.
- DHC


On Sun, Feb 3, 2013 at 10:13 AM, Martin Kletzander <mkletzan at redhat.com>wrote:

> On 02/03/2013 08:40 AM, Dan Kenigsberg wrote:
> > On Fri, Feb 01, 2013 at 11:44:08PM +0100, Martin Kletzander wrote:
> >> On 02/01/2013 09:29 PM, Dead Horse wrote:
> >>> To test further I loaded up two more identical servers with EL 6.3 and
> the
> >>> same package versions originally indicated. The difference here is
> that I
> >>> did not turn these into ovirt nodes. EG: installing VDSM.
> >>>
> >>> - All configurations were left at defaults on both servers
> >>> - iptables and selinux disabled on both servers
> >>> - verified full connectivty between both servers
> >>> - setup ssh (/root/authorized keys) between the servers --> this
> turned out
> >>> to be the key!
> >>>
> >>> Then using syntax found here:
> >>> http://libvirt.org/migration.html#flowpeer2peer
> >>> EG: From the source server I issued the following:
> >>>
> >>
> >> So your client equals to the source server, that makes us sure that the
> >> connection is made on the same network for p2p and non-p2p migration.
> >>
> >>> virsh migrate --p2p sl63 qemu+ssh://192.168.1.2/system
> >>>
> >>
> >> You're using ssh transport here, but isn't vdsm using tcp or tls?
> >
> > It is!
> >
>
> So then testing it with '+ssh' does not help much.  But at least we know
> the addresses are reachable.
>
> >> According to the config file tcp transport is enabled with no
> >> authentication whatsoever...
> >>
> >>> It fails in exactly the same way as previously indicated when the
> >>> destination server does not have an ssh rsa pub ID from the source
> system
> >>> in it's /root/.ssh/authorized_keys file.
> >>> However once the ssh rsa pub ID is in place on the destination system
> all
> >>> is well and migrations work as expected.
> >>>
> >>
> >> ..., which would mean you need no ssh keys when migrating using tcp
> >> transport instead.
> >>
> >> Also during p2p migration the source libvirt daemon can't ask you for
> >> the password, but when not using p2p the client is connecting to the
> >> destination, thus being able to ask for the password and/or use
> >> different ssh keys.
> >>
> >> But it looks like none of this has anything to do with the problem as:
> >>
> >>  1) as you found out, changing vdsm versions makes the problem go
> >> away/appear and
> >
> > I've missed this point. Which version of vdsm makes it go away?
> >
>
> Sorry, I've got it stuck in my head that part of the thread was about
> it, but when going through the mail now it makes less sense than before.
>  I probably understood that from [1] and maybe some other sentence that
> mixed in my head, but was related to the ssh migration.
>
> Sorry for that,
> Martin
>
> [1] http://www.mail-archive.com/users@ovirt.org/msg06105.html
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20130204/97088592/attachment-0001.html>


More information about the Users mailing list