Upgrading Earlier Release of ClusterWare to ClusterWare 5.3.0

Penguin Computing recommends that if you wish to upgrade a RHEL4 or CentOS4 base distribution to RHEL5/CentOS5, that you accomplish this with a full install of Release 5, rather then attempt to update from Release 4 to Release 5. If you already have ClusterWare 4 installed on the physical hardware that you will convert to RHEL5/CentOS5, then we recommend that you backup your RHEL4/CentOS4 master node prior to the reinstallation of RHEL5/CentOS5, as some of the Release 4 configuration files may be a useful reference for Release 5, especially files in /etc/beowulf/.

When upgrading from an earlier ClusterWare 5 version to ClusterWare 5.3.0, you should perform the following steps:

  1. Stop the Beowulf cluster: /sbin/service beowulf stop

  2. Clean the yum cache to a known state: yum clean all

  3. Update the RHEL5/CentOS5 base distribution, taking care to not update or install ClusterWare:

        yum --disablerepo=cw* update
  4. If upgrading from ClusterWare 5.2.0, then edit file /etc/yum.repos.d/clusterware.repo to change the baseurl entries from version 5.2 to 5.3.

  5. If your cluster includes Infiniband hardware, then you can ensure that all the necessary Infiniband-related rpms from the base distribution are installed:

        yum groupinstall Infiniband
    where the Infiniband group is defined through the ClusterWare repo configuration file and refers to rpms that reside in the base distribution repository.

  6. Upgrade ClusterWare:

        yum groupupdate Scyld-ClusterWare
    If yum fails with a Transaction Check Error that complains that a base distribution rpm is newer than the Scyld ClusterWare rpm that is attempting to replace it, then you must manually install the downlevel Scyld ClusterWare rpm(s). For example, if the complaint is about the kernel rpms, then do:
        cd /var/cache/yum/cw-core/packages
        rpm -Uvh kernel-headers-*
        rpm -i --oldpackage kernel-2* kernel-dev*
    and then repeat the groupupdate:
        yum groupupdate Scyld-ClusterWare
    which should now complete! successfully.

  7. Compare /etc/beowulf/config, which remains untouched by the ClusterWare update, with the new CW5.3.0 config.rpmnew (if that file exists), examine the differences:

        cd /etc/beowulf
        diff config config.rpmnew
    and carefully merge the config.rpmnew differences into /etc/beowulf/config. Generally speaking, be reluctant to make changes to /etc/beowulf/config. However, you should merge the config.rpmnew changes to the nodename, libraries, and prestage keyword entries, and merge new config.rpmnew comments.

    Similarly, the pre-CW5.3.0 /etc/beowulf/fstab may have been saved as fstab.rpmsave if it was locally modified. If so, merge those local changes back into /etc/beowulf/fstab. See the Section called Resolve *.rpmnew and *.rpmsave configuration file differences for details.

  8. Examine /etc/grub.conf to confirm that the new CW5.3.0 kernel is the default, then reboot your master node.