[Pacemaker] Two virtual machines start when they fail in "Live migration".

Andrew Beekhof andrew at beekhof.net
Mon Feb 14 08:38:44 UTC 2011


On Mon, Feb 14, 2011 at 4:04 AM, Yuusuke IIDA
<iidayuus at intellilink.co.jp> wrote:
> Hi, Andrew
>
> (2011/02/09 23:55), Andrew Beekhof wrote:
>>
>> 2011/2/7 Yuusuke IIDA<iidayuus at intellilink.co.jp>:
>>>
>>> Hi,
>>>
>>> I am going to manage the virtual machine in environment of
>>> Pacemaker-1.0.10.
>>>
>>> However, the problem that two virtual machines are started occurs when a
>>> transition graph of "Live Migration" was canceled on the way.
>>>
>>> I show a procedure to generate below a problem.
>>>
>>> 1. I was going to let you do "Live Migration" of a virtual machine from
>>> host 1
>>> to host 2.
>>>
>>> 2. In the middle of "Live Migration", monitor by pingd broke down with
>>> host 2,
>>> and an attribute changed.
>>>
>>> 3. A virtual machine is done start of with host 3 by the management of
>>> the cluster.
>>> Because the virtual machine which moved in migrate_to with host 1 then
>>> works, I
>>> am in a condition that two virtual machines start.
>>>
>>> This phenomenon thinks that it is revised in Pacemaker-1.1.
>>
>> I'm not at all surprised that the 1.1 behavior was better, we put a
>> lot of work into it recently.
>>
>>> If this phenomenon is a bug, I want you to revise it in Pacemaker-1.0.
>>
>> The necessary patch is quite large and invasive, I don't feel its
>> appropriate to backport into 1.0.
>
> Will it be not to be able to recommend it in pacemaker-1.0 to move a
> resource by setting of "allow_migrate=true"?

Well you can set that and if the RA doesn't fail everything will be happy.
But as you saw, the recovery characteristics if/when part of the
migration fails are not ideal.




More information about the Pacemaker mailing list