I guess you misunderstood me
I'm going to try this scheme:
|STORAGE|
FC / \
|SERV1/tgtd| |SERV2/tgtd|
iSCSI \ /
|ethernet switches|
iSCSI ||||||||
|blades|blades|blades|
serv1/serv2 - connectivity isnt a problem, multipathed FC scheme, all
good. Same lun accessible for both servers and than exported via tgtd to
iSCSI: with different target names ("iqn.2013-03.serv1:store",
"iqn.2013-03.serv2:store"), but same vendor_id, product_id, scsi_sn,
scsi_id. That way client can login into both targets and see lun as
multipathed device.
And multipath failover scheme (via custom config with
path_grouping_policy=failover for corresponding vendor_id/product_id) is
on blades-clients - so they use only one target at time (no round-robin
or similar stuff), but with ability to switch to another target in case
one of serv1/serv2 is down.
However, in my case "serv2" would not be available during oVirt setup
(need to setup ovirt and virtual servers to move stuff first), so i cant
enter both targets on storage domain initialization - that's why I'm
asking if there's any way to edit storage domain details after
initialization without destroying it (maybe directly via sql or something).
Yuriy Demchenko
On 04/02/2013 06:26 PM, Shu Ming wrote:
I am not sure if the multipathd can recognize the FC path to the
storage when the second server is available and regards it as the same
as the iSCSI path used before. If it is not, I think the device under
/dev/mapper may change when you cut the iSCSI path off and then enable
FC path. That will definitely corrupt the meta data of the volume
group which the storage domain is sitting on and the storage domain
will be corrupted finally.