Re: [RFC] quorum module configuration bits

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, Jan 12, 2012 at 5:58 AM, Dejan Muhamedagic <dejan@xxxxxxx> wrote:
> Hi,
>
> On Wed, Jan 11, 2012 at 01:08:56PM -0500, David Teigland wrote:
>> On Wed, Jan 11, 2012 at 11:21:59AM -0500, David Teigland wrote:
>> > > I much prefer the expected_votes field to any enumeration of the nodes.
>> >
>> > Expect admins to keep track of what expected_votes should be?
>> > That sounds nearly impossible to do correctly.
>> >
>> > > Does anyone need the list of nodes to be preconfigured?
>> >
>> > The alternative would be trying to remember what they all are?
>> > Defining the set of nodes that compose the cluster seems like
>> > a very good thing just for its own sake.
>>
>> I guess that's just one view.  There really are two legitimate styles
>> of assembling a cluster that we're trying to merge.
>>
>> In the first, you have a conceptual view of just a big pool nodes, more
>> or less anonymous.  They are just "workers", which higher level software
>> (e.g. pacemaker) can use as needed to run things.
>>
>> The second is a more fixed, static approach.  You start with with some
>> specific nodes and you want to access some shared resource from those
>> nodes.  Because the resource is shared, there needs to be clustering
>> involved to access it.  e.g. this node needs to mount that gfs2 file
>> system.
>>
>> The first view has historically been taken by corosync and pacemaker,
>> and the second by cman/gfs2.  Folks with the first view would think
>> of a static node list as unnecessary, whereas it's the starting point
>> in the second view.
>
> Not sure if I can recall exactly, but HPC clusters belong to the
> first category. At least I'd expect them to.
>
> Even though corosync doesn't require (for multicast) a fixed list
> of nodes, I'd put it also in the second category. I doubt that
> there are many corosync or openais clusters which have node sets
> frequently changing. As for pacemaker, it just gets the
> membership events and creates nodes as necessary. Heartbeat
> requires that all nodes are listed in static configuration files.

No it doesn't.  Thats just one mode of operation, it also supports autojoin.

>
> As for whether admins can err counting nodes I guess it's not
> likely for small clusters. Though one never knows ;-)
>
> Thanks,
>
> Dejan
>
>> Anyway, I think that explains the disconnect.  Getting back to the real
>> question of what needs a node list, quorum is the only thing I can think of.
>>
>> _______________________________________________
>> discuss mailing list
>> discuss@xxxxxxxxxxxx
>> http://lists.corosync.org/mailman/listinfo/discuss
> _______________________________________________
> discuss mailing list
> discuss@xxxxxxxxxxxx
> http://lists.corosync.org/mailman/listinfo/discuss
_______________________________________________
discuss mailing list
discuss@xxxxxxxxxxxx
http://lists.corosync.org/mailman/listinfo/discuss



[Index of Archives]     [Linux Clusters]     [Corosync Project]     [Linux USB Devel]     [Linux Audio Users]     [Photo]     [Yosemite News]    [Yosemite Photos]    [Linux Kernel]     [Linux SCSI]     [X.Org]

  Powered by Linux