Over the past couple of months I have experienced 4 dead 2008 R2 SP1 domain controllers after power failure on Hyper-V 2012 hosts. The domain controller VMs will start after power failure and have varying degrees of file system corruption. In each instance the corruption has rendered the domain controller unusable. The problem has not occurred with every power failure, but in testing the failure rate has been over 10%.
The Hyper-v 2012 hosts are as follows:
- Dell PowerEdge r720 with flash backed write cache on Raid controller
- Dell PowerEdge T710 with battery backed write cache on Raid controller
- Dell PowerEdge T310 with a single SATA hard drive and write cache disabled
- Generic system with a single SATA hard drive and write cache disabled
The VM configuration experiencing corruption is as follows:
- Each VM was created from a base 2008 R2 SP1 syspreped VHDx image template file (40 Gigs)
- The image template was originally created as a VHD and was then converted to a VHDx
- The VHDx file has 512k sectors instead of the native 512e of VHDx files (a result of VHD - VHDx conversion)
- Each VM was assigned 1024 Meg RAM and 1 virtual processor
- The domain controllers were created by promoting the base 2008 r2 install to a DC after base image deployment
- Only one corrupt VM was not running the 2012 integration components. The rest were running current 2012 integration components
I have done extensive testing on this issue and the problem for me seems to revolve around the VHDx file format. I have managed numerous Hyper-v installations since the original 2008 server version was released and I have never seen corruption like this
until 2012 and VHDx.
For the past few days I have been testing fixed sized VHD VMs on a 2012 host and I have not been able to reproduce the data corruption issue. I seem to only be able to reproduce the problem when using dynamic VHDx files. I have not done any testing on 2012
hosts with fixed size VHDx files or dynamic VHD files.
It would be great to hear from anybody else experiencing similar issues so that we can compare notes and hopefully get to the bottom of this problem.