Problems Creating/Overwriting partition tables on the slave nodes.

[Nate Summers]

Some believe that Robert Sand <rsand at d.umn.edu> on 12/08/00
04:21PM said:

>I've been having problems setting up the IDE drives on my slave
nodes.  First of
>all I am using the software installed from the Distribution CD
from Scyld.  The
>drives on the slaves are all the same type of drive and
currently have RedHat
>Linux installed on them.  What I want to do is instead of
booting from the
>floppy drive is to boot from the IDE drive.

>When I do a /usr/lib/beoboot/bin/beofdisk -q the partition table
in
>/etc/beowulf/fdisk looks good so I want to write that partion
table to the disks
>using /usr/lib/beoboot/bin/beofdisk -w.  This runs just fine and
it tells me I
>have to reboot the slave nodes for the new table to be saved.  I
reboot the
>slaves and they come up just fine but I can't do the
beoboot-install because it
>has problems writing to the disk.  Doing a bpsh 0 /sbin/sfdisk
-l /dev/hda it
>shows me that the partition table has not even changed: it is
still the old
>Redhat installation.

>Something is wrong here and the Help I've received from Richard
Walsh is great
>but it also brings up other problems.  He states that I should
copy everything
>in /lib to the slaves using bpcp.  Every single time I do this I
get error
>messages saying invalid option or file not found.  Then I do a
bpsh 0 ls -la
>/lib and it only shows me the one library that is there after
booting.  What is
?up here?  Does bpcp really work or does it only work when you
boot from the hard
>drive?

>These are really bad problems and I do need to get this thing
running again so
>that our faculity can use it.  If anyone has had this problem
I'd appreciate it
>if I could find out how you got around it.  If anyone at Scyld
can help me that
>would be great too.  I was told by someone at Scyld that I
should get the
>distribution CD when I was trying to install this from the RPMs.
 Now I have and
>I don't get answers to these questions from them.  I know this
is free software
>but I am perplexed at why they wouldn't like to know there may
be a problem
>here.

Looks very similar to a problem I had.  The node_up script would
crash, leaving the node in the "unavailable" state and the romfs
mounted as root. So of course you can't copy anything over.  The
solution is to fix the "extra free" bug in bpcp that cause it, and
thus node_up, to crash.  Obtain the source from the bproc srpm,
and then change clients/bpcp.c line 314 (I believe) to say
something like /* free(bp); */

Hope this helps.

Rockwalrus