win2008r2 amd64 crash when using same backingfile about kvm-guest-drivers-windows

[YanVugenfirer]

Hi,

Thank you for the report.

First of all please send provide minidump file. I am on vacation now, but
will look at it on Monday when I will be back.

Also what do you mean by "backing file" - one images for multiple VMs? One
base image for multiple VMs?

If possible please also provide the exact way you deploy the VMs.

Thanks,
Yan.

On Fri, Dec 14, 2012 at 2:23 PM, paymentslab [email protected]:

Environment:

labuser@os-cloud-2:~$ uname -a
Linux os-cloud-2 3.2.0-29-generic #46-Ubuntu SMP Fri Jul 27 17:03:23 UTC
2012 x86_64 x86_64 x86_64 GNU/Linux

labuser@os-cloud-2:~$ kvm --version
QEMU emulator version 1.0 (qemu-kvm-1.0), Copyright (c) 2003-2008 Fabrice
Bellard

Multiple version drivers (up to build-48) from:
http://people.redhat.com/vrozenfe/

Openstack Folsom is used for deployment

libvirt:
cache=none
image=qcow2

I've been tracking down an issue that when I deploy multiple win2k8r2
instances that end up using the same backingfile (at it seems this is the
issue) I end up with a BSOD for every instance except the first. This
happens often directly when applying the new settings to the generalized
instance and sometimes to a running instance.

If I make the backingfile unique per instance the problem seems to go
away, but needless to say that is not a situation I would like to maintain.

I can provide a minidump and even provide access to the environment if
that helps tracking down the issue.

—
Reply to this email directly or view it on GitHubhttps://github.com/YanVugenfirer/kvm-guest-drivers-windows/issues/6.

---

Daynix Computing LTD
Yan Vugenfirer, Technology expert and consultant
Email: [email protected]
Phone: [phone]
Web: www.daynix.com

[paymentslab]

Hi,

Thanks for the reply.

You can find the minidump here:
https://docs.google.com/open?id=0B_Y7S4YFVWvYLU5WdDlFdmdQS0E

I'm not entirely sure what drivers I'm using here, I have been experimenting with them. Most of the time I'm not able to pass the "installing device drivers" stage. This one is by (coincidence) just after this so I was able to recover and get the minidump.

Backing file is used for thin provisioning. Ie: multiple instances use the same base image but write their own changes to a separate file.

It seems pretty reproducible, as said I can give access the the environment if required.

Cheers
Bolke

[YanVugenfirer]

Hi Vadim,

Please take a look, the crash is in viostor. If needed we can get an access
to the setup.

Thanks,
Yan.

On Sun, Dec 16, 2012 at 9:44 PM, paymentslab [email protected]:

Hi,

Thanks for the reply.

You can find the minidump here:
https://docs.google.com/open?id=0B_Y7S4YFVWvYLU5WdDlFdmdQS0E

I'm not entirely sure what drivers I'm using here, I have been
experimenting with them. Most of the time I'm not able to pass the
"installing device drivers" stage. This one is by (coincidence) just after
this so I was able to recover and get the minidump.

Backing file is used for thin provisioning. Ie: multiple instances use the
same base image but write their own changes to a separate file.

It seems pretty reproducible, as said I can give access the the
environment if required.

Cheers
Bolke

—
Reply to this email directly or view it on GitHubhttps://github.com/YanVugenfirer/kvm-guest-drivers-windows/issues/6#issuecomment-11421863.

---

Daynix Computing LTD
Yan Vugenfirer, Technology expert and consultant
Email: [email protected]
Phone: [phone]
Web: www.daynix.com

[vrozenfe]

Using the same backend file is a quite unusual and dangerous scenario. Modern qemu fails with " Failed to get "write" lock error" to prevent such kind of situation. Using virtio-scsi (+LIO Target) is the right way to establish multiply connection to the same backend storage from single or multiply VM(s).

Closing this issue as it will not be fixed.
Vadim.

[bolkedebruin]

That’s a bit strange, cause this is often the default for cloud systems in order to preserve storage. But I’m quite sure you are misinterpreting the issue? One backing file works in a copy on write fashion. It works as the base image and changes are written to a different file. I don’t consider this “dangerous”.

This issue was opened 5 years ago not sure if it still persists.

[vrozenfe]

I'm probably wrong, but in my understanding sharing the same backend storage between several VMs supposed to imply some sort of IO synchronisation like SCSI-3 PR to keep it running safely. viostor driver was not designed to work in this mode.