From: sysop@cybergrants.com
Date: Thu Mar 25 2004 - 16:24:17 EST
Hello managers,
my apologies for the late summary. I planned to summarize upon a problem
resolution (whether positive or negative), but it seems that we won't be
able to obtain downtime for a while to try things out and see the results.
Once the problem resolves, I will send another summary.
I would like to thank these folks for their kind replies: Darren Dunham,
Octave Orgeron, John Bossert, Joon Martin Hansen, and Alex Madden
**** Original problem:
Each time we attempted to import a large table into an Oracle 9.0.1.4
database running on a V480R, the server froze, without logging
messages or any traceable readouts whatsoever. All we could do was power it
OFF and then back ON via RSC. The box wouldn't respond to a console 'break'
or XIR.
*****
Joon Martin recommended to decrease the size of shared memory segment to
~75-80% or physical RAM. He had an almost
identical situation; after changing the shmmax his issue got resolved. I
changed shmmax to 3GB (75% size of RAM),
rebooted the box, and will re-try the import as soon as we obtain a
maintenance window (management won't let us do it
anymore without at least a 2-hour downtime request, because they are afraid
it may crash and we'll be in the infamous "get the
tapes" situation again) A strange thing is that we were able to import a
similar dump into the same database a week earlier, although that doesn't
prove nor disprove anything...
*****
Darren Dunham suggested that a kernel can be booted into kadb (OK> boot
kadb) to load OS into the debugger and
from that point the RSC break may be able to get through the debugger. Once
in debugger, using the $q sequence can
drop out of the debugger to the OBP to generate a core dump.
*****
Alex Madden informed that when using both ce interfaces and both 66Mhz PCI
slots on v480 may cause fatal errors and hang
the server. In our case, we are using one 66Mhz slot (the only card) and
both ce interfaces. I would assume if we
encountered this condition it would have happened whether or not we were
doing imports.
*****
John Bossert and Octave Orgeron suggested this is most likely an oracle
problem (or a bug) and should be escalated.
This is true, oracle's TAR support seems to be quite vague with the
answers. I haven't escalated yet, but definitely
will if we have another failure next time.
Thank you!
Max.
CyberGrants, Inc.
_______________________________________________
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 - 23:28:21 EDT