Backup + suspend can lookup vm

[Thalhammer]

Hi,

I think I hit a bug today.
I tried to do a backup (in suspend mode) of my FreeNAS VM to my NFS Backup share.
I know now that in retrospect this obviously was doomed to fail, because the NFS Share lives (you might guess it) on the FreeNAS VM.
TLDR: Once I hit backup Proxmox suspended the VM and tried to do a backup to the (now defunct) NFS share which obviously failed.
However instead of just erroring out and resuming the VM again the VM keep suspended and I could not find any way whatsoever to get it to resume.
Trying to shut it down just hangs and the only way to get back to a working state was issueing "qm stop" via ssh which ended up killing it via SIGKILL.

I know this is a verymuch unsupported setup, but I don't think this is limited to this specific setup. I.e. if your NFS share goes down during a suspend backup you will have to KILL the VM. I dont know if this is a proxmox issue or a kvm issue but it definitly feels like a bug waiting to bite someone.

Sincerely,
Thalhammer

 

[Stefan_R]

Might be related to this issue: https://bugzilla.proxmox.com/show_bug.cgi?id=3225

In general, NFS is sometimes a bit tricky with lockups and the like, but you're right, this shouldn't happen - though in your case it can't work in general of course, backing up to itself

 

[Thalhammer]

In general, NFS is sometimes a bit tricky with lockups and the like, but you're right, this shouldn't happen - though in your case it can't work in general of course, backing up to itself

I mean it kind of would work if it keept the vm running (the backup only affects the 32Gb boot virtual disk, the actuall storage lies on a raid array with the controller passed through). But yeah I figured out why it wasn't my brightest Idea pretty quick. And I am not at all complaining about this special case.

What I mean is that there should be any way to "unpause" the vm in case a backup in suspend mode fails.

 

[Stefan_R]

Well, there is, 'qm stop' does the trick, the fact that it stopped in suspend mode is the actual bug that needs fixing (or potentially has been fixed with the patch from the bug entry I mentioned).