[ClusterLabs] Fencing with a 3-node (1 for quorum only) cluster

[Andrei Borzenkov]

On Fri, Aug 5, 2016 at 7:08 AM, Digimer <lists at alteeve.ca> wrote:
> On 04/08/16 11:44 PM, Andrei Borzenkov wrote:
>> 05.08.2016 02:33, Digimer пишет:
>>> On 04/08/16 07:21 PM, Dan Swartzendruber wrote:
>>>> On 2016-08-04 19:03, Digimer wrote:
>>>>> On 04/08/16 06:56 PM, Dan Swartzendruber wrote:
>>>>>> I'm setting up an HA NFS server to serve up storage to a couple of
>>>>>> vsphere hosts.  I have a virtual IP, and it depends on a ZFS resource
>>>>>> agent which imports or exports a pool.
>>
>> ...
>>
>>>
>>> Note; If you lose power to the mainboard (which we've seen, failed
>>> mainboard voltage regulator did this once), you lose the IPMI (DRAC)
>>> BMC. This scenario will leave your cluster blocked without an external
>>> secondary fence method, like switched PDUs.
>>>
>>
>> As in this case there is shared storage (at least, so I understood),
>> using persistent SCSI reservations or SBD as secondary channel can be
>> considered.
>
> Yup. That would be fabric fencing though, or are you talking about using
> it under watchdog timers?

fabric is the third possibility :) No, I rather mean something like fence_scsi.

Although the practical problem of both fabric or scsi fencing is that
it only prevents concurrent access to shared storage; it does not
guarantee that other resources are also cleaned up, so may end up with
duplicated IP or similar.

> If fabric, then my worry is always a panic'ed
> admin clearing it without properly verifying the state of the lost node.
> With watchdog, it's fine, just slow.
>

As it is last resort better slow than never.