On Oct 5, 2011, at 11:03 AM, Florian Haas wrote:

> On 2011-10-05 16:44, Vadym Chepkov wrote:
>> 
>> On Oct 5, 2011, at 9:38 AM, Florian Haas wrote:
>> 
>>> On 2011-10-05 14:59, Vadym Chepkov wrote:
>>>> Hi,
>>>> 
>>>> It looks like iscsi multipath IO is not supported by iscsi RA.
>>>> What would be the proper way to configure iSCSI MPIO in a pacemaker 
>>>> cluster?
>>> 
>>> OK, so first of all are you absolutely positive you _must_ use MPIO?
>>> Most users find it much easier to just run their iSCSI initiator over a
>>> bonded network interface if all you want to protect against is network
>>> failure.
>>> 
>> 
>> This could start a holy war, but this is configuration which is recommended 
>> by Dell and, furthermore , not all storages have bonding capabilities, but, 
>> usually, they do have multiple ethernet interfaces.
> 
> Granted, if you want completely separate paths from your nodes to the
> storage boxes, then the bonding approach doesn't get you too far. Still,
> for most users protecting against failure of one uplink path is
> sufficient, and bonding does do nicely at that.
> 
>>> Secondly, you can always define two iscsi resources that Pacemaker
>>> manages via ocf:heartbeat:iscsi. Then as soon as those come online,
>>> provided the SCSI target sets consistent serial numbers and SCSI IDs,
>>> multipathd should happily put them together as one mpath device. In your
>>> configuration, does it not do that?
>>> 
>> 
>> This would create a very complicated configuration. You have to colocate 
>> both resources and you can't use inf: in colocation because it would break 
>> redundancy  (all or nothing)
> 
> Well, "inf colocation constraint breaks redundancy" is an exaggeration,
> but I think I see what you mean. I'm still thinking an advisory
> colocation constraint would help you achieve what you need, though.
> 
>> I guess lsb:iscsi RA is the only option.
> 
> I wonder how that makes the setup less complicated. So you've got one
> resource that manages all of your iSCSI connections. Then what if one of
> them fails? Do you ignore that? Do you reconnect all of them? What if
> not all of them fail, but two that make up the same mpath device?
> Correct, your Pacemaker configuration has a few lines less, but you're
> signing up for massive headaches if things go wrong. Things, that is,
> that Pacemaker could fix for you automatically.
> 


The problem is not with amount of lines, pacemaker has limitations which I 
don't see how to solve with your approach.
How would you configure dependencies on such "combined" resource? There is no 
way to express "if one of two is running", unless I miss something.

Regards,
Vadym

_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to