Hyper-V snapshots: not for production

One of the greatest things about virtual machines is the ability to take snapshots, which can be used to quickly roll a VM back to a known state.

Previously, I have written about some of the shortcomings with Microsoft’s Hyper-V snapshots.  Or were they checkpoints?  I forget… but that’s not important.  The main problem with the Hyper-V snapshot implementation — that does not affect VMware ESX — is that the snapshots are not actually removed until the VM is powered off.  All the way off — reboots don’t count.  And the VM must remain powered off until the AVHD file is completely merged, otherwise the process is aborted.

By the way, if you are using System Center Virtual Machine Manager 2008, you will have no way of knowing when the snapshot merge is complete.  SCVMM just simply does not report this information — switch to the other single pane of glass (Hyper-V Manager) if you need to be sure.  Hyper-V administrators are used to this by now.

So, if you find your VM in “Paused-Critical” state, you might want to check for a hidden snapshot file filling your LUN.  If you’ve landed on VCritical in search of such a solution, you are not alone — quite a few Google searchers end up here due to this issue.  Before Ben Armstrong’s recent — and very informative — series on snapshots, VCritical was among the first to elucidate this issue.

If a hypervisor’s snapshot functionality is not recommended for production, is that hypervisor itself ready for production?

Tags: , ,

5 comments

  1. Shawn’s avatar

    If you create a Hyper-V snapshot, a snapshot file is created and grows as changes are made to the VM. If I then decide to delete that snapshot, (merge not revert) the changes are not merged until the VM is powered off.
    Thats easy to understand, but where are the writes happening between the time I ‘delete’ the snapshot and the VM being powered off and the merge happens. Do they continue against the snapshot file, causing it to continue to still grow? That is just retarded if that is the case. What if the reason I ‘deleted’ the snaphot was because of space constraints? Will I have to suffer downtime to poweroff the VM because of this?

    On another note – I would do this but I’m lazy 🙂
    You should compile a list with brief description and links, for each of these little issues with Hyper-V and SCVMM youve blogged about over the past months. This allow those of us in pre-sales roles to give customers looking at a Hyper-V solution complete facts about what they may be getting themselves into. Not everything is in the feature/price comparison, day to day operational issue matter as well.

    1. Eric Gray’s avatar

      Shawn, thanks for the comment. Yes, it is as you describe — the existing snapshot file just keeps growing until you power the VM off. When Microsoft says not recommended for production, they are not kidding!

      I’m glad you find the VCritical articles useful, good idea to write up a summary page on the topic. I’ll consider it.

    2. Etienne Pouliot’s avatar

      I totally agree with you. Furthermore, that’s exactly the same reason I do not recommend Xenserver. You don’t get snapshots with Xenserver unless you are connected to a Dell or a NetApp SAN (Details in blog (in french)). In my mind, VMware is the only “production ready” virtualization product.

    3. Eric Gray’s avatar

      Etienne, great point about XenServer, I’m sure that comes as a surprise to many people. VMware ESX snapshots are storage-independent.

Comments are now closed.