Get confirmation from the user
Get user's confirmation that the machine can actually be reinstalled (via mail).
Put machine into maintenance
- Login to
lxadm
-
sms set maintenance 'os upgrade' lxbuild023
The difference between
standby and
maintenance in sms, is that
standby machines are still monitored and alarms are still sent to the operators.
Update machine's CDB profile
$ cdbop
cdbop> get profiles/profile_lxbuild023.tpl
cdbop> !vi profiles/profile_lxbuild023.tpl
- Replace
pro_type_lxbuild_slc3
with pro_type_lxbuild_x86_64_slc4
- And commit:
cdbop> update profiles/profile_lxbuild023.tpl
cdbop> commit
The
commit
can take up to 30 minutes... (!)
On lxadm, run
PrepareInstall lxbuild023
You might get the following message:
[lxadm01] ~ > PrepareInstall lxbuild023
Conflicting values for OS found in CDB ("slc4") and CDBSQL ("slc3")
Note that it can take up to 30 minutes to synchronize CDBSQL with CDB.
[ERR] Not all necessary information could be found, exiting...
In this case, the only thing to do is to wait for the synchronization to happen...
Reboot the machine
On the machine itself:
shutdown -r now
Takes ~1 hour (2 reboots are actually done).
Check the ongoing reinstallation
Two different ways to do that:
- Try to ssh to the machine
$ ping lxbuild023
$ ssh lxbuild023
$ tail -f ksxxx.log
On lxadm:
$ /afs/cern.ch/group/c3/bin/connect2console.sh lxbuild023
$ tail -f /var/log/ncm/ncd.log
To exit the console:
Ctrl-e c .
To reboot a machine:
Ctrl-e c l 1 b
Check that the reinstallation went ok
On the machine:
$ lemon-host-check
$ df -k
The "df -k" should show a
/build
partition on the build machines...
Clear the machine from maintenance state
Warning: do not set the machine to 'production' state, but clear it from the 'maintenance' state.
Like that, the machine:
- goes back to the previous state it was in (ex:
standby
),
- or goes back to its default state (ex:
production
)
$ sms clear maintenance "os upgrade" lxbuild023
Troubleshooting
If the machine doesn't get reinstalled after reboot, it can be because:
- the MAC address given by
ifconfig
and registered in LANDB are different.
To solve this:
- Add the MAC address in LANDB (but keep the other one registered).
If this didn't help, create a Remedy Ticket in IT-CM:
File > Open > Object List > ITCM:CallManagement
Double click on Search (the upper part banner), it should change into New.
Enter the Category, Host, etc.
And Save the ticket, it will actually be created like that...
In the end, the BIOS settings needed to be corrected to boot on the network.
--
SophieLemaitre - 13 Dec 2007