ctdb relock file issues with glusterfs

[ronnie sahlberg]

At least for larger clusters, say 45 or more nodes
we could implement something like a
"as long as floor(n/2)+1  nodes are connected"
quorum

that would make it safe to use without a reclock file on clusters with
5 or more nodes.



On Tue, Oct 16, 2012 at 6:30 AM, Martin Schwenke <martin at meltin.net> wrote:
> Hi Morten,
>
> On Mon, 15 Oct 2012 15:59:41 +0000, Morten Bøhmer
> <Morten.Bohmer at pilaro.no> wrote:
>
>> For the heck of it I installed a couple of Centos virtual servers and configure ctdb+glusterfs+xfs+samba, got it working, but without relock.
>>
>> Not sure how important it is, but I guess time will show :)
>
> As others have pointed out, the reclock file is crucial unless you
> have some other way of avoiding a split brain.  You won't know you need
> it until you get a split brain and something gets badly corrupted.
>
> Additionally, as Ronnie mentioned, if the reclock file isn't working
> then there's something wrong with your locking, so you have more than a
> reclock problem...
>
> peace & happiness,
> martin