unwanted SCSI target reassignment during fibre channel disk replacement

From: Sean Walmsley (sean.p.walmsley@opg.com)
Date: Fri Jul 05 2002 - 19:53:44 EDT


machine: Sunfire V880
disk array: IBM ProFibre DF4000J (JBOD)
HBA card: Qlogic QLA2300F 2Gb/s
OS: Solaris 8 with latest patches

I have a fibre channel array of disks numbered c3t0d0..c3tNd0 where the
target addresses conveniently match the arbitrated loop physical
addresses (i.e. slot 3 in the array is scsi target 3). I didn't do
anything special to accomplish this but it is very nice since it makes
finding disks easy.

My problem is that I can't seem to replace a disk and maintain the slot
to target correspondence without rebooting the machine.

If I remove the disk in slot 3 (c3t2d0) and replace it with a new unit then
the target 3 disk shows up as "drive not available: formatting" from within
the format command and the new disk doesn't appear at all. If I issue the
devfsadm command (with or without -C) to get around this:

  - the new disk in slot 3 gets assigned target N+1 (i.e. the next available
    target) rather than the expected target 3
  - target 3 still shows up as being "not available"

What's worse, if I later reboot the machine I get a second surprise as the
new disk goes back to being target 3 again as originally expected.

If I shut the machine down during the drive replacement, the new disk is seen
as target 3 right away and everything is ok.

My hypothesis is that as long as the OS is running, it is somehow remembering
the original disk's world wide name (WWN) and linking this to target 3.
When I put a new disk with a different WWN in the same physical slot the OS
does not see it as an identical replacement and therefore assigns it a new
target.

Rebooting must clean up this "memory" somehow and allow the new disk to be
seen as an identical replacment.

This behaviour is completely different from SCSI drives (which don't have a
WWN) and where hot swapping a drive does not cause target reassignment.

I'd appreciate any suggestions to ensure that target always matches physical
slot and is completely independent of disk WWN occupying the slot. Is the
system somehow remembering the WWN of the original disk until a reboot. If
so, where is this information stored and how can I zap it?

It seems completely ridiculous that I should need to reboot just to avoid
having a a disk replacement result in target reassignment.

Thanks,

Sean Walmsley

===============================================================
Sean Walmsley, Nuclear Analysis Dept., Ontario Power Generation
sean.p.walmsley@opg.com 416-592-4608 (V) 416-592-5528 (F)
700 University Ave (H11 G26), Toronto, Ontario, M5G 1X6, CANADA
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers



This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 23:24:33 EDT