zfs code - Oct 2007 - zfs + iscsi target + vmware esx server

I''m posting here as this seems to be a zfs issue.  We also have an open
ticket with Sun support and I''ve heard another large sun customer also
is reporting this as an issue.

Basic Problem:  Create a zfs file system and set shareiscsi to on.  On a vmware
esx server discover that iscsi target.  It shows up as 249 luns.  When
attempting to then add the storage esx server eventually times out, if you view
it from command line you see it checking each lun and crashing out before it
gets to 249.

Test Environment: 
Sun ISCSI Target Host:  compaq PC with 2 80GB SATA drives, 1GB RAM, 3.2GHz CPU
running solaris 10 x86 update 4 (08/07).  The second drive is setup as a storage
pool with 1 filesystem created with shareiscsi on.

ESX Server: compaq PC with SCSI HD, 1GB RAM, 3.2GHz CPU running vmware esx
server 3.0.2

Other test scenarios:

We also created an iscsi target via iscsitadm using a spare slice on the primary
disk.  esx server sees this target just fine with a single lun as expected.

Trying to get more creative I did an iscsitadm modify admin and set the path to
my zfs filesystem.  I then used iscsitadm to create a new target that was a file
(which would get created on the zfs filesystem).  However in esx server I see
the same results with 249 luns.  The one difference this time is the first lun
is the size I created the target as and the other 248 are the size of the zfs
filesystem.

So if zfs is involved it screws up, if it''s not it''s fine.

I do not yet have another solaris 10 update 4 system to test as an initiator but
my update 3 system sees it just fine in all test scenarios.

It seems to be an issue with how the iscsi target is being broadcast when zfs is
involved that the solaris initiator doesn''t seem to mind but esx server
sure does.  Since it works fine when I use a regular disk slice I think
it''s something to do with zfs but I wouldn''t rule out an issue
with esx completely yet.

Any help is greatly appreciated.  We are looking to role out multiple thumpers
and vmware servers using the thumpers as their backend storage via iscsi (or nfs
if we have to, but would rather go iscsi)

Thanks!
Adam
--
This messages posted from opensolaris.org

Heya,

I''ve not got any useful information for you I''m afraid, just
posting to ask if you could update this thread if you get a resolution for this.

We''re a small company just looking at the thumper systems. 
We''re planning to use them instead of a SAN, so we''ll be using
iSCSI targets for VMware as well as Samba shares for our windows clients.

While the VMware rollout wouldn''t be for a while yet, any problems with
iSCSI would be a big concern.
--
This messages posted from opensolaris.org

Hmm... doing a quick google, it looks like you''re not alone with this. 
I''m finding people reporting what looks like the same problem from
March.  This thread sounds like a similar bug, but with a workaround:
http://communities.vmware.com/message/598035

Forgive me if this isn''t relevant, or if it''s old news. 
I''m a windows admin by trade so Solaris is sill very new to me :)

It also looks like there''s a iSCSI / ZFS freeze bug, which could also
be a concern:
http://bugs.opensolaris.org/view_bug.do;jsessionid=d60126b5a9b7c364a1fa8b83aeee?bug_id=6563222
--
This messages posted from opensolaris.org