Revisit of U320 and U160 SCSI bus problems with FC3t2.91
Bill Cronk
ngc4013 at cox.net
Sun Oct 17 15:13:58 UTC 2004
Bill Cronk wrote:
> I have consistantly had problems with booting systems which have a
> JBOD RAID box and FC2 or FC3t2.91. The problem remains the same and
> the work around I figured out is only temporary.
>
> ----- CUT -----
> So now I ask the questions... Why?? Any further info needed? Can
> someone else setup a RAID box and recreate the problem to confirm that
> there are issues in FC2 and FC3 with RAID durring the boot cycle? So
> far it happens with a 144GB, 324GB, 657GB, and 2TB RAID boxes
> configured as either RAID0 or RAID5, on internal U160 or U320 SCSI
> ports and an addon U320 Adaptec card.
>
> I think there is an issue with the events durring the boot cycle in
> that it tries to mount the RAID before the RAID configuration sets the
> RAID up.
>
> Bill
>
The suggestion Tom made to change the hard drive ID to 'Linux Raid
Autodetect' seems to have worked just fine. I am amazed that I could not
find any reference to this need in any documentation I have or I just
completely missed it again!!
Also there is only a single device file such as /dev/md0 and no more
/dev/md? existing. So I tried to do 'mknod -m 660 /dev/md1 b 9 1' which
works fine, then chgrp disk /dev/md1' which works fine and then I can
get the second RAID online. However, if I reboot then any added in
/dev/md? all go away leaving the original /dev/md0. ???? I don't have a
clue and I can reproduce it on two different installs of FC3 and an FC2
install.
Now another question has come up, seems that any RAID box built on
another machine or on a different install instance, presents a problem
when hooking up two RAID boxes together on one server with a dual SCSI
port. The existing RAID box seems to have certain parameters burried
into the superblocks regarding whether it was setup previously as
/dev/md0, 1, 2... so on. This is a problem if two RAID boxes need to be
moved to a new server and they both were originally setup as /dev/md0
connections. How do you get around this without rebuilding the box and
blowing away the data?
I also seem to be experiencing a problem of SCSI RAID detection order
versus SCSI RAID mount order... seems to be flipped when they mount, but
this could be an artifact related to my previous question regarding
parameters burried in the superblocks.
Bill
More information about the fedora-test-list
mailing list