[Linux-HA] HA of virtual machiens

[Andrew Beekhof]

On Feb 3, 2008, at 10:39 PM, Amos Shapira wrote:

> On Feb 4, 2008 7:13 AM, Lars Marowsky-Bree <lmb at suse.de> wrote:
>
>> On 2008-02-01T13:12:53, Michael Schwartzkopff <misch at multinet.de>  
>> wrote:
>>
>>> HA! Even better solution:
>>> 8 virtual machines on 2 physical servers form one cluster.  
>>> heartbeat in
>> that
>>> cluster can monitor application resources.
>>
>> Layered heartbeat clusters don't work well right now.
>>
>> Your best bet really is to run the 8 virtual machines on the two
>> physical nodes, and have heartbeat+PaceMaker/CRM run directly on the
>> physical machines (managing the guests) - service availability can be
>> checked by connecting to the guests via TCP and seeing whether their
>> hosted service is still available.
>>
>> That is the least complex solution to your problem.
>
>
> Actually, we are in the process of setting up something similar:
> 2 guest Xen hosts (each having separate function) on each physical  
> machine,
> with heartbeat directly between the Xen guests and their tweens on  
> another
> physical machine. The guests share data partitions via DRBD.
>
> I too was wondering whether I should do the HA at the host level but  
> it
> looks much simpler to do the heartbeat directly between the guests  
> because,
> for instance, it looks much simpler to manage the DRBD access
> (primary/secondary, mount/umount) directly from inside the Xen  
> guests than
> from the host (the Xen guest can't even access the DRBD'ed Logical  
> Volume).
>
> So far it works well in my tests. Still have to finish the setup and  
> see the
> performance, though.
>
> How should I go about if I wanted to do this at the host level? I  
> though
> maybe DRBD'ing the data partition while the stand-by guest is in "xm  
> save",
> then if the Xen hosts decide to take over the resources then it  
> should "xm
> restore" the Xen guest? I don't have enough experience with that  
> side of
> Xen, only reading the docs about this.
>
> Setup is Intel Xeon 5130, CentOS 5, Xen 3.0.3, heartbeat 2.1.2 using
> haresources (the new CRM-style config kept crashing on me, as I  
> asked here a
> few weeks ago).

Crashing?
What was the subject?  I don't recall this.