Greetings. We recently had 2 panics of a GS140 running Tru64v5.1a and
TruCluster, PK6. The panics were of the type:
panic (cpu 0): kernel memory fault
HP backline UNIX support spent quite a while analyzing the kernels,
dumps, etc. They say there is a CSP (Customer Specific Patch) which
seems to address this particular panic, which is supposedly very rare,
and that the patch is for T64 v5.1a at PK6.
We have installed PK6 recently (actually 8 days prior to the first
panic), but the support person's analysis of our system makes him
apprehensive of recommending the patch without our first installing
every subset on the T64 standard distribution kit. He says that because
we omit various subsets (non-mandatory), when PK6 installed on top of
PK2, a lot of patches were not installed because of missing
dependencies. He feels that since our patched kernel has more
differences to a PK6 system than it does to a PK2 system, he can't
insure that the CSP will actually work correctly.
I understand his hesitation, but it seems incredible to me that we
should have to install every unnecessary package that is on the
distribution and re-do PK6 again just to safely install a CSP which, as
far as he's told me, contains no specific dependencies to uninstalled
packages.
I am suspicious of PK6 generally, due to the proximity of the install
and the 2 panics of this previously very stable cluster node. However,
there was some odd behavior (hangs, error msgs) during state changes in
the no-roll patch install, and the session.log file that recorded the
patch install says that it installed the kernel of the other
(non-panicking) node onto this server, which seems very odd:
" New Kernel (/sys/[OTHER NON-PANICKING NODE]/vmunix) copied to
cluster/members/{memb}/boot_partition/vmunix."
I am hoping someone out there may have some experience or insights on
this issue.
TIA...
Judith Reed
Service delivery manager
Navisite, Inc.
Received on Thu Oct 20 2005 - 20:38:22 NZDT