Upgrading the Red Hat Cluster suite to version 2.0

This document describes how to upgrade a Red Hat Cluster from version 1.03 (as shipped in etch) to version 2.0.

The upgrade procedure consists of stopping the cluster, removing the old cluster suite, adding node IDs to the cluster.conf, installing and booting a new kernel, and finally installing the new cluster suite packages.

The upgrade procedure is as follows:

* Stop client access to the clustered services.

* On each cluster node:

  1. Stop all applications using the cluster infrastructure, like the shared storage
  2. Stop the rgmanager, if installed:
     killall clurgmgr 
  3. Umount all gfs shares:
     umount -a -t gfs 
    if a gfs share is still blocked by another process, run
     lsof -n | grep <mountpoint> 
    to identify the process in order to terminate it.
  4. Stop the cluster lvm daemon:
     /etc/init.d/clvm stop 
  5. Stop fencing:
     /etc/init.d/fence stop 
  6. Stop the cluster manager:
     /etc/init.d/cman stop 
  7. Stop the cluster configuration service:
     /etc/init.d/ccs stop 
  8. Remove the old Red Hat Cluster software:
     dpkg --purge ccs libccs-dev cman libcman1 libcman-dev libdlm1 libdlm-dev fence gfs-tools gnbd-client gnbd-server gulm libgulm1 libgulm-dev libiddev-dev libmagma1 libmagma-dev magma-plugin-gulm magma-plugin-sm redhat-cluster-source 
    Note: not all packages might be installed, just ignore the warnings dpkg will issue about it.
  9. Update the cluster.conf file to contain nodeids:

    Open the file /etc/cluster/cluster.conf in your favorite editor. In each <clusternode> element, insert nodeid="number" after name="name". This should look this way:

     [...]
     <clusternode name="node1" nodeid="1">
     [...] 
    Note: the nodeid parameter was optional in the old Red Had Clustersuite, but is now mandatory. If you have already nodeid parameters in your cluster.conf, skip this step. After saving your changes to the cluster.conf file, copy it to all the other nodes in the cluster.
  10. Update the kernel, and reboot the Node in order to activate the new kernel. You might of course upgrade the other parts of the system before rebooting.
  11. Install the new version of the Red Hat Cluster suite:
     apt-get install redhat-cluster-suite redhat-cluster-modules 
    Note: existing GFS version 1 filesystems are fully supported. The new GFS2 filesystem format is not production-ready and usage is recommended only for experimental setups.

TODO

- custom kernels: patch required

- GULM:

- double-check with real upgrade