Chris Lalancette wrote:
Daniel Veillard wrote:
> On Mon, Aug 24, 2009 at 09:51:05PM +0100, Daniel P. Berrange wrote:
>> If a file descriptor with events=0 was added to the libvirtd
>> event loop, it would still be added to the poll() fds' array.
>> While it wouldn't see any POLLIN/OUT events, it'd still get
>> triggered for HANGUP/ERROR events which was not in compliance
>> with the libvirt events API contract.
> Independantly of the rest of the stream patches this sounds fine to
> push,
Hm, this patch seems to be causing a regression. If I start up a KVM guest, let
it completely boot, then run "shutdown -h now" inside the guest, the guest
properly shuts down and qemu-kvm exits. However, it is never removed from the
list of "active" guests (i.e. it shows up in "virsh list"), and the
state is set
to "no state". Reverting just this patch fixes the issue. I'll look at it
some
more to try to figure out what is happening, but we might want to revert it for
the time being (since we don't technically need it at the moment).
OK, danpb pointed out the problem. Don't revert this patch, I'll post a follow
up to fix the problem in a couple of minutes.
--
Chris Lalancette