Views
NewNewBioBrewClusterSetup
I've upgraded puya to ROCKS 3.3 (3 required cd, base, hpc+kernel, area51+sge+java) and BioBrew? 3.3. The upgrade preserved the user home directories (in /export/home), user accounts (/etc/shadow, password, auto.home, ...) and the extend-compute.xml customizations (since they are stored on /export/home/install). Pretty slick.
Trigger an upgrade by booting the frontend with frontend upgrade, we'll actually do a clean install on the frontend, but preserve /home and the /etc files. When prompted, do a manual partition with Disk Druid, format the / and swap partitions, and change the filesystem options to migrate the /export partition. If you format the /export partition, or autoformat, we'll loose all user data and customizations.
After the frontend comes back up, you need to run insert-ethers to capture the mac addresses of all nodes, switches, and laptops again. The upgrade instructions have instructions to run on the nodes to make them upgrade cleanly, but since we're going to be installing more nodes we need to boot them from the CD.
DANGER Will Robinson, DANGER
After the upgrade, the extend-compute.xml was not integrated into the node builds. It was in the wrong (old) directory. We moved the file to the current path as described in the ROCKS 3.3 documentation and rebuilt the dist.
We also had a really hard time getting the biobrew roll installed on the compute nodes, apparently ROCKS was being confused by the presence of 2 biobrew rolls, 3.1.0 and 3.3.0. I deleted 3.1.0 and now it works.
Stupid node tricks: qlogin -pe fork 2 gives you an interactive session into a dedicated node. If you do that as root you can trigger a reinstall of an otherwise unused node by executing /boot/kickstart/cluster-kickstart.
Carlos wrote a script to reinstall all the nodes that are missing biobrew after the current jobs end.
See notes on whiteboard for more info.
Limiting number of jobs per user --humberto, Wed, 21 Nov 2007 17:35:37 -0400 reply
Set 25 jobs per user limit:
http://gridengine.sunsource.net/servlets/ReadMsg?list=users&msgNo=3942
# qconf -ssconf |grep maxu maxujobs 25
# qconf -msconf
Brings up the file in an interactive editor.