On Jan 31, 2008 5:02 PM, Sharpe, Sam J <
sam.sharpe@imperial.ac.uk> wrote:
On 31 Jan 2008, at 21:43, dbcooper wrote:
> Domenico, (and everyone else thanks)
>
> Thank you for your input. This seems to be a deeper rabbit hole than
> I expected.
>
> I tried (echo "1" > /sys/bus/scsi/drivers/sd/<SCSI-ID>/delete) and
> sure enough it cleanly wiped out the /dev/sdc and the /sys/block/
> device etc. Upon reboot, it was back and had the same old Buffer I/
> O errors and block errors on reboot and recreated the same /dev/sdc
> etc.
>
> So I started digging through the udev / udevinfo commands and
> comparing with lspci to find that yes the ghost /dev/sdc was hanging
> (supposedly) off my Qlogic Fiber Card, I then launched my Navisphere
> to make sure it only had one LUN being handed out (and it is). So I
> got even more confused. I've not got much SAN experince and I was
> handed this with my servers. I then took a look at the SANSurfer Cli
> tool (sfcli), I was able to see that the HBA does see (looks like
> it) 2 LUNS of exaclty the same size (I.E. it's really just the one I
> guess).
A barrage of questions:
Whose QLogic drivers are you using? Do you have multiple Fibre cards
in the system? Is your QLogic card dual-port? Do you have multiple SAN
switches?
This sounds to me very like our systems would look when we have two
SAN paths to the same LUN (i.e. redundant fibre cards and a redundant
fabric). Unless you are using QLogic drivers that natively understand
Multipathing (it's an option - Red Hat's ship without as the preferred
method is dm-multipath - HPs (just as an example) ship with it turned
on so you don't need dm-multipath.)
--
Sam
> The LUN that is being presented to the machine is being mounted as
> sdd. (the offender I'm hunting down is sdc).
> Do you think it's possible that when I delete the sdc with the above
> "echo 1" you suggested and reboot, that the kernel is trying to
> assign the presented LUN as the next device in line sdc? (since sda,
> sdb and sdd are taken).
>
> I don't know anything about the persistent binding (yet) but I'll be
> going through all the docs from the Qlogic and EMC products tomorrow
> (now that I know what to start looking at).
>
> Thanks for any other advice (if you can suggest any).
>
Thank you, I'll get back to this tomorrow.