Windows Server - [IMPORTANT!!!] Windows erasing SAN disk

Asked By Gaspar

05-Feb-10 08:14 AM

I think I discovered a BIG problem with Windows 2003 x64 R2 SP2.
Suppose I set a setup a server and connecting it to a SAN disk and then
format this disk. When I setup a secondary server to the same disk the disk
is seen as unformatted and Windows has to initialize it again, so I lost all
data in it.

This is my configuration:
- Two HP ML 370 servers with Emulex HBA (2 x FC) connected to a HP MSA
StorageWorks SAN.
- HBAs and HP have latest firmwares
- There is only one RAID-1 5 Gb volume (for testing purposes)

This are the steps to reproduce the problem:
1) Disconnect server from SANs (unplug FC cords from HBAs)
2) Install Windows 2003 in Server A, and setup up all drivers
3) Shut down server A and plug FC cords.
4) Start server. Go to disk management. Windows shows the "Initialize Disk"
Wizard.
5) Format the drive. Assign drive letter. Copy some test files.
6) Shut down server A.
7) Install Windows 2003 in Server B, and setup up all drivers
8) Shut down server B and plug FC cords.
9) Start server. Go to disk management. Windows AGAIN shows the "Initialize
Disk" Wizard (I click cancel).
10) The already initialized disk by Server A is now shown as "Unformatted"
in Server B. All data is lost (I test shutting down and starting Server A).

However, after Server B initializes disk (i.e. the second disk
initialization), and then I shut it down, Server A can see all files. So,
this problem is related to the first time Windows sees the disk.

Final conclusions:
- I can confirm that this is not related to hardware, because I tried
different configurations / HBAs / changing LUNs, etc. and the problem still
happens
- I can reproduce the problem over and over. I also tried with virtual
machines (using VSphere + ESXi) using snapshots so as to rollback and test
again and again.

Have you ever seen something like? What do you suggest?
Thanks a lot for your time!
Gaspar.

07-Feb-10 08:53 AM

Hi Gaspar,

sounds for me like a normal behaviour in a non-clustered environment.
Generally a Disk can become corrupted if no "instance" (like a cluster disk
service) is taken the control of r/w.

If you just want to migrate / copy the date via FC, use clone or shutdown
the server first, change zoning, and present the disk again to the other
server. But NEVER present disks in a non-clustered environment to more
servers. Windows 2008 can handle that but 2003 can under several
circumstances corrupt the disks.

Regards
Ramazan

08-Feb-10 12:25 PM

I agree with Ramazan, but from what you have described, something is not right
in your configuration or your description of the issue.

At step 10, when you plug in the cables, Server B should not be asking to
initialize the disks...unless you have other non-initialized disks presented
to this host. Initializing a disk simply writes a disk signature and the
disk should already have a signature from Server A.

You say the disk is shown as "Unformatted"...I would expect to see that the
disk is initialized and you should be able to see a partition on the disk
(should be blue in disk manager) but does not show a disk label or file
system (RAW). If this is the case, this is the expected behavior of Windows
2003 Enterprise as the "automount" feature is disabled by default in Windows
2003 enterprise. When you present a partitioned/formatted disk to a Windows
2003 EE server for the first time, by default, it does not mount the disk.
You need to manually assign a driver letter/mount point to the disk in order
for it to show up mounted after rebooting. You can adjust this default
behavior, but I would not recommend doing so.

Anyway, at this point, you simply need to add a drive letter to this disk,
and then close and re-open disk manager. Once this is done, you should be
able to see the filesystem and all files that were created on Server A.

Hope this helps.

Regards,
John

15-Feb-10 05:44 AM

Step 10 is the key: it is supposed to do what you describe, but it does not
see a formatted partition (so as to assign a drive letter).
I understand the comments from Ramazan but the fact is that there is no
simultaneous access from the two servers. What I have done is exactly the same
as removing an IDE drive from a computer, plug-it in another, and then this
second computers must see all data in the disk.

Thanks again.
Gaspar