Easy is Hard


One Line Summary

Why can't we have an HA appliance that you just plug in, turn on and forget about?


This discussion explores some of the issues encountered when trying to completely automate setup of an HA cluster, including:

  • Someone still has to configure networking and shared storage.
  • Synchronization of config files during bootstrap, or when adding another node, is remarkably fiddly (even with csync2).
  • Difficulty/inability to automatically detect corosync clusters on a network.
  • Chicken-and-egg problems configuring cluster and LVM, OCFS2 etc. before the cluster is up.

And after all that, you still have to configure your actual workload (e.g.: network bridging, MAC address and VM image deployment in a cluster of VMs).


ha, clustering, appliance


  • Tim Serong



    Tim is currently employed by Novell as Senior Clustering Engineer, working on the SUSE Linux Enterprise High Availability Extension. He has been building, breaking and fixing Pacemaker clusters for the past several years, through which experience he has discovered that you can never ever have too many log files.

    Interests include motorcycling, beer, and making technology simpler and more reliable (although not necessarily in that order, and usually not all at the same time).

Leave a private comment to organizers about this proposal