Le 20/01/2011 11:57, Bart Coninckx a écrit :
> On Thursday 20 January 2011 11:37:58 Bart Coninckx wrote:
>> On Thursday 20 January 2011 09:44:47 Jean Baptiste Favre wrote:
>>> Hello Bart,
>>> Answers inline.
>>>
>>> Le 19/01/2011 21:49, Bart Coninckx a écrit :
>>>> On Wednesday 19 January 2011 21:15:46 Jean Baptiste FAVRE wrote:
>>>>> Hello Bart,
>>>>> I wrote such an howto some times ago. It's available here in french
>>>>> (http://publications.jbfavre.org/virtualisation/cluster-xen-corosync-p
>>>>> ac ema ker-drbd-ocfs2.fr) and here in english
>>>>> (http://publications.jbfavre.org/virtualisation/cluster-xen-corosync-p
>>>>> ac ema ker-drbd-ocfs2.en).
>>>>>
>>>>> It's LVM based but could be easily adapted for img files.
>>>>>
>>>>> Regards,
>>>>> JB
>>>>>
>>>>> Le 19/01/2011 17:54, Bart Coninckx a écrit :
>>>>>> Hi all,
>>>>>>
>>>>>> could somebody point me to what is considered a sound way to offer
>>>>>> Xen guests on a two node DRBD cluster in combination with Pacemaker?
>>>>>> I prefer block devices over images for the DomU's. I understand that
>>>>>> for live migration DRBD 8.3 is needed, but I'm not sure as to what
>>>>>> kind of resource
>>>>>> agents/technologies are advised (LVM,cLVM, ...) and what kind of DRBD
>>>>>> config (seperate devices for each DomU I guess?)
>>>>>> Thank you!
>>>>>> Bart
>>>>
>>>> Hi Jean,
>>>>
>>>> thank you for this document, it seems highly educational. Could you
>>>> please verify if I understand correctly:
>>>>
>>>> - you use LVM to build your DRBD resources on (not the other way
>>>> around)
>>>
>>> At dom0 level, I have 2 VG:
>>> - system, for... system :)
>>> - XenHosting, for domUs and common Xen related FS
>>>
>>> Inside XenHosting, I create LVs:
>>> - One for common stuff (config, kernels, iso, ...).
>>> - One for each domU
>>>
>>> Each LV are defined as DRBD resources.
>>>
>>> Having one DRBD ressource per domU allow you to migrate them
>>> independently. More, if your cluster grows and get a third server, you
>>> can balance DRBD ressources between them. It's just easier to manage.
>>>
>>> Each domU boots on its LV. That means that at domU level, your DRBD
>>> ressource (or LV) is saw as a disk. Then you install your domU system.
>>> The way I install domUs makes me use LV as well
>>>
>>> As a summary, you have LV inside DRBD ressource onto LV.
>>> It seems complicated but in fact it's not so tricky :)
>>>
>>> If you want to use img files, either you store them on one LV only, and
>>> you'll have to create cluster FS on it, or you use separate LV. In case
>>> of separate LV, I prefer installing domU in LV directly instead of using
>>> img file because it's easier to access LV based FS from dom0 in case
>>> domU crashes.
>>>
>>>> - you use a DRBD resource with a OCFS2 filesystem on to offer the ISO's
>>>> and config files on both nodes for every DomU
>>>
>>> Yes, using cluster FS allow you to mount it on each node and provides
>>> concurrent access for each dom0. I choose OCFS2 at the time of writing,
>>> but I had to do it now, I would give a try to GlusterFS.
>>>
>>> You can choose another way (like rsync), but I prefer cluster FS as I
>>> don't have to think about synchronisation.
>>>
>>>> - for each DomU you create a seperate DRBD resource that is allowed to
>>>> be dual primary so you can do live migrations.
>>>
>>> You MUST allow dual primary if you want to use live migration. So it's
>>> not cluster specific.
>>>
>>>> Am I doing all right? ;-)
>>>
>>> Yep, seems ;)
>>>
>>> Regards,
>>> JB
>>>
>>> _______________________________________________
>>> Xen-users mailing list
>>> Xen-users@xxxxxxxxxxxxxxxxxxx
>>> http://lists.xensource.com/xen-users
>>
>> Jean,
>>
>> all clear, except for the LVM on top of the DRBD resource: is that LV
>> created on Dom0 or during the installation of Dom0.
>>
>> It seems a sound way of implementing stuff. I roughly did the same in the
>> past but with an external Pacemaker for an iSCSI cluster. Since this
>> project is without it, I was a bit unsure a to how doing the live
>> migration and such.
>>
>> You don't use any stonith?
>
>
> Oops, I mean:
>
>> all clear, except for the LVM on top of the DRBD resource: is that LV
>> created on Dom0 or during the installation of DomU?
Basically, it depend on how you install your domU.
For example, I'm working on Debian, if you use debootstrap from dom0 to
install your domU, you'll have to create first LV.
If you install your domU with PXE, then the OS installer will deal with
LV creation.
About stonith, I did not configure it because it was only a proof of
concept. The only stonith I could use was SSH, not very powerfull (I've
no IPMI capable card on my servers).
But if you plan to deploy such an architecture in production, you're
very strongly encouraged to setup stonith, indeed :)
Regards,
JB
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
Home