[dm-devel] multipath device question

[Josef Whiter]

On Sat, Feb 11, 2006 at 01:43:54AM +0100, Stanislav Polášek wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Hi! Let me please start with apology for using the development list to
> ask rather end-user question. I am looking for help with multipath
> setup. I have a few servers connected to a ibm storage. Two servers have
> two qlogic HBAs each. On the first server was created a multipath device
> out of all 4 scsi devices/paths. On the second server it is possible to
> assemble the same mp device and attach it without problems (there is the
> gfs).
> 
> However, the third server I would like to connect to the storage has
> just one HBA. It's only 1U format and I cannot add anything to it. And
> moreover, the visible storage scsi device order there is different.
> 
> As I know the mdadm that I used to create the multipath device writes to
> the superblock an information about the array, I would like to find out
> how much the multipath implementation in the kernel depends on this
> information. Would it be a problem to connect the third server to the
> storage by assembling the multipath device out of just two scsi
> devices/paths instead of four and in addition using scsi devices with
> diferrent minor/major numbers?
> 
> As I do really doubt that, is there any way how to connect the third
> server using the multipath setup?
> 
> Thank you very much for your help. I could get the answers by testing,
> but these are production servers with terrabytes of data and I cannot
> really stop them now for more then just a few minutes.
> 
> Stanislav Polasek

Hello,

DM multipathing is all automatic, it does not write any information to the disk,
it is completely nodal specific.  If you have different multipathing setup for
each node, or nodes that aren't multipathed, dm multipathing will figure
everything out by itself.  You will have to keep in mind that in order to add a
node to an existing two node GFS6.1 cluster you will have to bring the cluster
down because of the special two node option.  After you have more than two nodes
you can add nodes without rebooting the cluster.  Thank you,

Josef