Mailing List Archive

Issue with VirtualDomain
Hi,

I tested migration with a recent version of libvirt. Essentially it was from
squeeze-backports. The transport qemu+ssh://<other_node>/system now needs the
additional parameters

--p2p --tunnelled

in the migration command line. So it should be:

virsh migrate <guest> --p2p --tunnelled qemu+ssh://<other_node>/system

I will try to write a patch and publish it here.

Greetings,

--
Dr. Michael Schwartzkopff
Guardinistr. 63
81375 München

Tel: (0163) 172 50 98
Re: Issue with VirtualDomain [ In reply to ]
Il giorno Lun 19 Set 2011 12:17:30 CEST, Michael Schwartzkopff ha scritto:
> Hi,
> I tested migration with a recent version of libvirt. Essentially it was from
> squeeze-backports. The transport qemu+ssh://<other_node>/system now needs the
> additional parameters
> --p2p --tunnelled
> in the migration command line. So it should be:
> virsh migrate<guest> --p2p --tunnelled qemu+ssh://<other_node>/system
> I will try to write a patch and publish it here.
> Greetings,

Hi Michael,
I'm using Debian Squeeze with Pacemaker and libvirt from backports, but
I didn't have to make any patch to make live migration work.

Have you got a specific setup?

--
RaSca
Mia Mamma Usa Linux: Niente è impossibile da capire, se lo spieghi bene!
rasca@miamammausalinux.org
http://www.miamammausalinux.org

_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
Re: Issue with VirtualDomain [ In reply to ]
> Il giorno Lun 19 Set 2011 12:17:30 CEST, Michael Schwartzkopff ha scritto:
> > Hi,
> > I tested migration with a recent version of libvirt. Essentially it was
> > from squeeze-backports. The transport qemu+ssh://<other_node>/system now
> > needs the additional parameters
> > --p2p --tunnelled
> > in the migration command line. So it should be:
> > virsh migrate<guest> --p2p --tunnelled qemu+ssh://<other_node>/system
> > I will try to write a patch and publish it here.
> > Greetings,
>
> Hi Michael,
> I'm using Debian Squeeze with Pacemaker and libvirt from backports, but
> I didn't have to make any patch to make live migration work.
>
> Have you got a specific setup?

What transport do you use? Which version of libvirt?

--
Dr. Michael Schwartzkopff
Guardinistr. 63
81375 München

Tel: (0163) 172 50 98
Re: Issue with VirtualDomain [ In reply to ]
Il giorno Lun 19 Set 2011 15:17:30 CEST, Michael Schwartzkopff ha scritto:
[...]
> What transport do you use? Which version of libvirt?

Transport ssh, libvirt version is 0.9.2-7 (the squeeze-backports version).

--
RaSca
Mia Mamma Usa Linux: Niente è impossibile da capire, se lo spieghi bene!
rasca@miamammausalinux.org
http://www.miamammausalinux.org

_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
Re: Issue with VirtualDomain [ In reply to ]
> Il giorno Lun 19 Set 2011 15:17:30 CEST, Michael Schwartzkopff ha scritto:
> [...]
>
> > What transport do you use? Which version of libvirt?
>
> Transport ssh, libvirt version is 0.9.2-7 (the squeeze-backports version).

hu? Have to try again.

--
Dr. Michael Schwartzkopff
Guardinistr. 63
81375 München

Tel: (0163) 172 50 98
Re: Issue with VirtualDomain [ In reply to ]
> Il giorno Lun 19 Set 2011 15:17:30 CEST, Michael Schwartzkopff ha scritto:
> [...]
>
> > What transport do you use? Which version of libvirt?
>
> Transport ssh, libvirt version is 0.9.2-7 (the squeeze-backports version).

I get an error:

migration job: unexpectedly failed

Only if I add the above mentioned options to the migration it works.

--
Dr. Michael Schwartzkopff
Guardinistr. 63
81375 München

Tel: (0163) 172 50 98
Re: Issue with VirtualDomain [ In reply to ]
Il giorno Lun 19 Set 2011 21:20:12 CEST, Michael Schwartzkopff ha scritto:
>> Il giorno Lun 19 Set 2011 15:17:30 CEST, Michael Schwartzkopff ha scritto:
>> [...]
>>> What transport do you use? Which version of libvirt?
>> Transport ssh, libvirt version is 0.9.2-7 (the squeeze-backports version).
> I get an error:
> migration job: unexpectedly failed
> Only if I add the above mentioned options to the migration it works.

Hi Micheal, does this happens *every time* you try to migrate the
resource? I'm facing a strange problem with virtualdomain RA and node
fence (see my post on linux-ha ML "Understanding why a host fence (was:
Resource fail and node fence)") and I'm trying to understand where is
the problem.

Have you tried using another transport such ad tcp over sasl or tls?

--
RaSca
Mia Mamma Usa Linux: Niente è impossibile da capire, se lo spieghi bene!
rasca@miamammausalinux.org
http://www.miamammausalinux.org

_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
Re: Issue with VirtualDomain [ In reply to ]
> Il giorno Lun 19 Set 2011 21:20:12 CEST, Michael Schwartzkopff ha scritto:
> >> Il giorno Lun 19 Set 2011 15:17:30 CEST, Michael Schwartzkopff ha
> >> scritto: [...]
> >>
> >>> What transport do you use? Which version of libvirt?
> >>
> >> Transport ssh, libvirt version is 0.9.2-7 (the squeeze-backports
> >> version).
> >
> > I get an error:
> > migration job: unexpectedly failed
> > Only if I add the above mentioned options to the migration it works.
>
> Hi Micheal, does this happens *every time* you try to migrate the
> resource? I'm facing a strange problem with virtualdomain RA and node
> fence (see my post on linux-ha ML "Understanding why a host fence (was:
> Resource fail and node fence)") and I'm trying to understand where is
> the problem.

Yes. This happens every time. Also when I use the virsh command line to
migrate the virtual machine.

Your problem. Sorry, I have never seen this. I had several other issues with
VirtualDomain. Please see my patches to the RA:

http://www.gossamer-threads.com/lists/linuxha/dev/74103

> Have you tried using another transport such ad tcp over sasl or tls?

First tests with tcp show that it also needs the --p2p --tunnelled options. I
did not try sasl.

Greetings,

--
Dr. Michael Schwartzkopff
Guardinistr. 63
81375 München

Tel: (0163) 172 50 98
Re: Issue with VirtualDomain [ In reply to ]
Il giorno Mar 20 Set 2011 11:11:58 CEST, Michael Schwartzkopff ha scritto:
[...]
> Yes. This happens every time. Also when I use the virsh command line to
> migrate the virtual machine.

So the problem is inside libvirt. But are you using the same libvirt
version of mine (0.9.2-7)?

> Your problem. Sorry, I have never seen this. I had several other issues with
> VirtualDomain. Please see my patches to the RA:
> http://www.gossamer-threads.com/lists/linuxha/dev/74103

I saw your path and might be very useful. My problem is exactly on stop.
When the error come up, the vm is in state "paused" and it seems to be
this to make things broke up.

> First tests with tcp show that it also needs the --p2p --tunnelled options. I
> did not try sasl.
> Greetings,

Very strange. I think that the problem may be somewhere else (network
connection, collisions, physical staff)...

--
RaSca
Mia Mamma Usa Linux: Niente è impossibile da capire, se lo spieghi bene!
rasca@miamammausalinux.org
http://www.miamammausalinux.org

_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
Re: Issue with VirtualDomain [ In reply to ]
> Il giorno Mar 20 Set 2011 11:11:58 CEST, Michael Schwartzkopff ha scritto:
> [...]
>
> > Yes. This happens every time. Also when I use the virsh command line to
> > migrate the virtual machine.
>
> So the problem is inside libvirt. But are you using the same libvirt
> version of mine (0.9.2-7)?
>
> > Your problem. Sorry, I have never seen this. I had several other issues
> > with VirtualDomain. Please see my patches to the RA:
> > http://www.gossamer-threads.com/lists/linuxha/dev/74103
>
> I saw your path and might be very useful. My problem is exactly on stop.
> When the error come up, the vm is in state "paused" and it seems to be
> this to make things broke up.
>
> > First tests with tcp show that it also needs the --p2p --tunnelled
> > options. I did not try sasl.
> > Greetings,
>
> Very strange. I think that the problem may be somewhere else (network
> connection, collisions, physical staff)...

No. See man virsh.

I also have 0.9.2-7~bpo60+1.

--
Dr. Michael Schwartzkopff
Guardinistr. 63
81375 München

Tel: (0163) 172 50 98
Re: Issue with VirtualDomain [ In reply to ]
Il giorno Mar 20 Set 2011 12:27:20 CEST, Michael Schwartzkopff ha scritto:
[...]
>>> First tests with tcp show that it also needs the --p2p --tunnelled
>>> options. I did not try sasl.
>>> Greetings,
>> Very strange. I think that the problem may be somewhere else (network
>> connection, collisions, physical staff)...
> No. See man virsh.
> I also have 0.9.2-7~bpo60+1.

Man page says only the p2p is peer to peer migration and tunnelled is
tunnelled migration. And I agree with this :-)

For what I've found:

- tunnelled migration sets a migration in progress in the background, so
that libvirt is able to cancel a migration when a problem has happened.

- peer2peer make the libvirtd server connect to the destination libvirtd
server directly (peer-to-peer) setting up a secure channel.

In the first case I may agree that the parameter should help (even only
on failures), but for p2p... I can't say it.
Anyway, it doesn't explain why for me this version works correctly. If
you want I can send you my crm and libvirt configurations so we can do
some comparisons.

--
RaSca
Mia Mamma Usa Linux: Niente è impossibile da capire, se lo spieghi bene!
rasca@miamammausalinux.org
http://www.miamammausalinux.org

_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
Re: Issue with VirtualDomain [ In reply to ]
> Il giorno Mar 20 Set 2011 12:27:20 CEST, Michael Schwartzkopff ha scritto:
> [...]
>
> >>> First tests with tcp show that it also needs the --p2p --tunnelled
> >>> options. I did not try sasl.
> >>> Greetings,
> >>
> >> Very strange. I think that the problem may be somewhere else (network
> >> connection, collisions, physical staff)...
> >
> > No. See man virsh.
> > I also have 0.9.2-7~bpo60+1.
>
> Man page says only the p2p is peer to peer migration and tunnelled is
> tunnelled migration. And I agree with this :-)
>
> For what I've found:
>
> - tunnelled migration sets a migration in progress in the background, so
> that libvirt is able to cancel a migration when a problem has happened.
>
> - peer2peer make the libvirtd server connect to the destination libvirtd
> server directly (peer-to-peer) setting up a secure channel.
>
> In the first case I may agree that the parameter should help (even only
> on failures), but for p2p... I can't say it.
> Anyway, it doesn't explain why for me this version works correctly. If
> you want I can send you my crm and libvirt configurations so we can do
> some comparisons.

the cluster is completely irrelevant here. the plain command

virsh migrate <guest> qemu+ssh://<other_node>/system

doesn't work here. I need the options --p2p --tunnelled.

So this is a libvirt issue, no cluster issue.

But if there parameters are really needed, the VirtualDomain has to be
patched.

Greetings,

--
Dr. Michael Schwartzkopff
Guardinistr. 63
81375 München

Tel: (0163) 172 50 98
Re: Issue with VirtualDomain [ In reply to ]
Il giorno Mar 20 Set 2011 13:45:30 CEST, Michael Schwartzkopff ha scritto:
[...]
> the cluster is completely irrelevant here. the plain command
> virsh migrate<guest> qemu+ssh://<other_node>/system
> doesn't work here. I need the options --p2p --tunnelled.
> So this is a libvirt issue, no cluster issue.
> But if there parameters are really needed, the VirtualDomain has to be
> patched.
> Greetings,

This is for sure. Anyway this is my libvirt.conf:

listen_tls = 0
unix_sock_group = "libvirt"
unix_sock_rw_perms = "0770"
auth_unix_ro = "none"
auth_unix_rw = "none"

I haven't changed anything else.

--
RaSca
Mia Mamma Usa Linux: Niente è impossibile da capire, se lo spieghi bene!
rasca@miamammausalinux.org
http://www.miamammausalinux.org

_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
Re: Issue with VirtualDomain [ In reply to ]
> Hi,
>
> I tested migration with a recent version of libvirt. Essentially it was
> from squeeze-backports. The transport qemu+ssh://<other_node>/system now
> needs the additional parameters
>
> --p2p --tunnelled
>
> in the migration command line. So it should be:
>
> virsh migrate <guest> --p2p --tunnelled qemu+ssh://<other_node>/system
>
> I will try to write a patch and publish it here.
>
> Greetings,

No patch needed. It just works inside the cluster.

--
Dr. Michael Schwartzkopff
Guardinistr. 63
81375 München

Tel: (0163) 172 50 98
Re: Issue with VirtualDomain [ In reply to ]
Hi,

19.09.2011 13:17, Michael Schwartzkopff wrote:
> Hi,
>
> I tested migration with a recent version of libvirt. Essentially it was from
> squeeze-backports. The transport qemu+ssh://<other_node>/system now needs the
> additional parameters
>
> --p2p --tunnelled
>
> in the migration command line. So it should be:
>
> virsh migrate <guest> --p2p --tunnelled qemu+ssh://<other_node>/system

Do you have iptables enabled?
If yes, then do you allow incoming TCP traffic on ports 49152:49216 in
addition to 16514
(http://wiki.libvirt.org/page/FAQ#What_are_the_different_migration_methods.3F)?

Best,
Vladislav
_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
Re: Issue with VirtualDomain [ In reply to ]
> Hi,
>
> 19.09.2011 13:17, Michael Schwartzkopff wrote:
> > Hi,
> >
> > I tested migration with a recent version of libvirt. Essentially it was
> > from squeeze-backports. The transport qemu+ssh://<other_node>/system now
> > needs the additional parameters
> >
> > --p2p --tunnelled
> >
> > in the migration command line. So it should be:
> >
> > virsh migrate <guest> --p2p --tunnelled qemu+ssh://<other_node>/system
>
> Do you have iptables enabled?
> If yes, then do you allow incoming TCP traffic on ports 49152:49216 in
> addition to 16514
> (http://wiki.libvirt.org/page/FAQ#What_are_the_different_migration_methods.
> 3F)?

No firewall involved. nothing listing on port 49152ff.

If I use qemu+ssh for the transport I do not need libvirt to listen on
tcp/ports.

--
Dr. Michael Schwartzkopff
Guardinistr. 63
81375 München

Tel: (0163) 172 50 98
Re: Issue with VirtualDomain [ In reply to ]
Hi,

On Wed, Sep 21, 2011 at 11:28:07AM +0200, Michael Schwartzkopff wrote:
> > Hi,
> >
> > 19.09.2011 13:17, Michael Schwartzkopff wrote:
> > > Hi,
> > >
> > > I tested migration with a recent version of libvirt. Essentially it was
> > > from squeeze-backports. The transport qemu+ssh://<other_node>/system now
> > > needs the additional parameters
> > >
> > > --p2p --tunnelled
> > >
> > > in the migration command line. So it should be:
> > >
> > > virsh migrate <guest> --p2p --tunnelled qemu+ssh://<other_node>/system
> >
> > Do you have iptables enabled?
> > If yes, then do you allow incoming TCP traffic on ports 49152:49216 in
> > addition to 16514
> > (http://wiki.libvirt.org/page/FAQ#What_are_the_different_migration_methods.
> > 3F)?
>
> No firewall involved. nothing listing on port 49152ff.
>
> If I use qemu+ssh for the transport I do not need libvirt to listen on
> tcp/ports.

I think that the right way is to add a parameter with which
users can define extra options for virsh migrate. I suppose that
not everybody needs the very same options.

Thanks,

Dejan

> --
> Dr. Michael Schwartzkopff
> Guardinistr. 63
> 81375 München
>
> Tel: (0163) 172 50 98



> _______________________________________________________
> Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
> Home Page: http://linux-ha.org/

_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
Re: Issue with VirtualDomain [ In reply to ]
21.09.2011 13:03, Dejan Muhamedagic wrote:
> Hi,
>
> On Wed, Sep 21, 2011 at 11:28:07AM +0200, Michael Schwartzkopff wrote:
>>> Hi,
>>>
>>> 19.09.2011 13:17, Michael Schwartzkopff wrote:
>>>> Hi,
>>>>
>>>> I tested migration with a recent version of libvirt. Essentially it was
>>>> from squeeze-backports. The transport qemu+ssh://<other_node>/system now
>>>> needs the additional parameters
>>>>
>>>> --p2p --tunnelled
>>>>
>>>> in the migration command line. So it should be:
>>>>
>>>> virsh migrate <guest> --p2p --tunnelled qemu+ssh://<other_node>/system
>>>
>>> Do you have iptables enabled?
>>> If yes, then do you allow incoming TCP traffic on ports 49152:49216 in
>>> addition to 16514
>>> (http://wiki.libvirt.org/page/FAQ#What_are_the_different_migration_methods.
>>> 3F)?
>>
>> No firewall involved. nothing listing on port 49152ff.
>>
>> If I use qemu+ssh for the transport I do not need libvirt to listen on
>> tcp/ports.

Ah, missed +ssh.

>
> I think that the right way is to add a parameter with which
> users can define extra options for virsh migrate. I suppose that
> not everybody needs the very same options.

With sane (different) defaults for each transport?

Best,
Vladislav
_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
Re: Issue with VirtualDomain [ In reply to ]
Hi,

On Wed, Sep 21, 2011 at 09:34:57PM +0300, Vladislav Bogdanov wrote:
> 21.09.2011 13:03, Dejan Muhamedagic wrote:
> > Hi,
> >
> > On Wed, Sep 21, 2011 at 11:28:07AM +0200, Michael Schwartzkopff wrote:
> >>> Hi,
> >>>
> >>> 19.09.2011 13:17, Michael Schwartzkopff wrote:
> >>>> Hi,
> >>>>
> >>>> I tested migration with a recent version of libvirt. Essentially it was
> >>>> from squeeze-backports. The transport qemu+ssh://<other_node>/system now
> >>>> needs the additional parameters
> >>>>
> >>>> --p2p --tunnelled
> >>>>
> >>>> in the migration command line. So it should be:
> >>>>
> >>>> virsh migrate <guest> --p2p --tunnelled qemu+ssh://<other_node>/system
> >>>
> >>> Do you have iptables enabled?
> >>> If yes, then do you allow incoming TCP traffic on ports 49152:49216 in
> >>> addition to 16514
> >>> (http://wiki.libvirt.org/page/FAQ#What_are_the_different_migration_methods.
> >>> 3F)?
> >>
> >> No firewall involved. nothing listing on port 49152ff.
> >>
> >> If I use qemu+ssh for the transport I do not need libvirt to listen on
> >> tcp/ports.
>
> Ah, missed +ssh.
>
> >
> > I think that the right way is to add a parameter with which
> > users can define extra options for virsh migrate. I suppose that
> > not everybody needs the very same options.
>
> With sane (different) defaults for each transport?

Well, whatever makes sense :) I'm no expert on libvirt and
migrations. Hopefully, some day Florian will get involved again
as, IIRC, he's the author of VirtualDomain.

Cheers,

Dejan

> Best,
> Vladislav
> _______________________________________________________
> Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
> Home Page: http://linux-ha.org/
_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/