[PVE-User] OpenBSD 5.4 under ProxmoxVE 3.1 / KVM 1.4: problems so far

[Adam Thompson]

Just an FYI at this time for anyone else searching on this problem. On 
the other hand, feel free to share ideas if you have 'em.

OpenBSD 5.4 (RELEASE) does not appear to reliably receive ACPI signals 
delivered by KVM.  Or, the version of kvm/qemu (1.4) that ships with 
ProxmoxVE 3.1 (pve 3.1) fails to deliver ACPI shutdown signals to 
OpenBSD reliably.  I'm not sure which.  Sometimes it works, sometimes it 
doesn't.  Limited testing shows that ACPI events fail after the VM has 
been up and running for a while - not sure how long, yet.

I'm using virtio drivers for both network and disk, but limited testing 
so far does not show that this makes any difference.

I do note that vio(4) networking in this setup occasionally stops 
transmitting or receiving; switching back to em(4) resolves that 
particular issue (so far).  When the vio(4) driver goes awry, the only 
immediate symptom is that the VM stops sending and receiving packets.  
Later, I discover that afflicted VMs can no longer shut down cleanly, 
either... presumably a KVM/OpenBSD interaction of some sort, I'm not 
pointing fingers in *any* direction right now. (Especially since it 
could be something I've done, too.)

So far everything appears stable enough to run in production with the 
exception of vio(4).  I have had to virtually yank the plug on a few VMs 
in order to shut them down, however... back to the good 'ol days of 
SunOS 3: "shutdown() { 'sync;sync;sync;halt -npq' }" ;-).

-- 
-Adam Thompson
  athompso at athompso.net