High-availability virtualization with Xen and DRBD

Please note: if you are interested in a full-fledged, complete HA virtualization solution delivered as one fixed-bid service package, check out the product information for DRBD Virtualization Enterprise Cluster on our corporate web site.

For the German speakers among you, Hans-Joachim Baader has written an interesting howto on leveraging DRBD, Xen, and Heartbeat to make a full HA virtualization platform.

While this isn’t exactly quite a brand new approach (we at LINBIT have deployed this in projects), Hans-Joachim makes insightful observations when comparing Xen paravirtualization to the OpenVZ OS-level virtualization solution. Do take a look.

What I’m delighted to hear is that he has found DRBD/Heartbeat to be a painless and hassle-free add-on to both Xen and OpenVZ. My additions to his howto are:

  • Use Heartbeat V2 in CRM configuration mode, and the OCF-compliant Xen Resource Agent. Saves you all the hassle of hacking your own Heartbeat V1-compatible resource script.
  • Live Migration for Xen is doable with DRBD, but it’s a little tricky. Drop me an email if you are planning to deploy such a setup.
  • The Debian xen-tools package is great, but always requires a little tweaking. Especially the kernel and initrd options.

3 Responses to “High-availability virtualization with Xen and DRBD”

  1. Vic Says:

    Hi, I’m trying to set up two systems here with Xen, DRBD and Heartbeat to achieve high availability and live migration but am unable to find any english documentation to do so (it’s quite possible that the documentation is out there and I’m just blind). The physical machines currently run Fedora Core 8 and are currently running the Xen kernel as dom0. Can you assist me with this?

    Thank you.

  2. Florian Haas Says:

    Vic,
    this ought to be helpful:
    http://www.drbd.org/users-guide/ch-xen.html
    Cheers,
    Florian

  3. Charles Sibbald Says:

    Hi Florian,

    I have tried to get DRBD to work, and this seems possible on a pair of servers, but is it possible to have it functional in an N+1 cluster.

    I have 16 Cluster nodes, and a 17th node for failover.

    Please advise

    Regards

    charles

Leave a Reply