Hi,
this is my upgrade experience I'd like to make a FAQ out it, so comments
from more knowleadgeable people are more than welcome - required.
cheers
alessandra
#########################################################################
The main problem in the procedure has been the kernel/grub upgrade when I
tried on one node it automatically updated /boot/grub.conf but not the
rpms, so when I rebooted without either running by hand updaterpms or
without reinstalling it couldn't find the kernel.
Also there is a mismatch between the list of rpms directories and what is
really downloaded when you update the repository. A directory apps_common
is not there where it should be in /opt/local/linux/7.3/RPMS/ you should
create it (I actually added it to /etc/updaterep.conf so the updaterep
command downloaded it).
So to be safe:
a) Announce you are going to upgrade and an outage
b) Drain the queues switching off the mds on your CE
c) Wait for the remaining jobs to finish
(it will take some time but you can starts with the nodes that frees
themself before)
d) Download the new lcg-2_1_1 and check the differences. I made a cvs diff
on the example directory to see what has been added or removed and
applied the differences to my files, in particular site-cfg.h.
e) Replace the /etc/updaterep.conf with the new one:
to add apps_common in it
MIRROR 73 7.3/RPMS/apps_common http://grid-deployment.web.cern.ch/grid-deploy
ment/download/RpmDir/apps_common/index.html
f) Do the following
cd $YOUR-LCG2-DIR/tools
/usr/sbin/updaterep
./lcfgng_server_update.pl $lcg2Conf/rpmlist/lcfgng-common-rpm.h
/tmp/lcfgng_server_update_script.sh
./lcfgng_server_update.pl $lcg2Conf/rpmlist/lcfgng-server-rpm.h
/tmp/lcfgng_server_update_script.sh
/tmp/lcfgng_server_update_script.sh
The last I run it twice in case some dependencies failed
There is no need to reinstall nsinstallroot
Recompile your profiles and try on one node
/etc/obj/nfs restart <== I had problems with nfs so I restarted the object
/etc/obj/autofs restart <== same reason
/etc/obj/updaterpms <== to make sure the kernel is in place in the
when you reboot
reboot
I know you can use also rdxprof node_name on the node called node_name
which should reload the profile but updaterpms doesn't seem to pick it up
unless you run it by hand. I've honestly reinstalled all of the nodes and
the CE to have a fresh installation (but that's my choice). I didn't
reinstall the SE though not to delete anything experiments had on it, that
I did by hand.
|