[Pkg-libvirt-maintainers] Bug#863266: Bug#863266: libvirt-daemon: spice port conflict - multiple VMs want Port 5900

[Guido Günther]

On Wed, May 24, 2017 at 05:24:51PM +0200, Marc Haber wrote:
> Package: libvirt-daemon
> Version: 3.0.0-4
> Severity: normal
> 
> Dear Maintainer,
> 
> somewhere in mid-may, libvirt broke in regard to autoport and SPICE in
> sid. On my systems, I cannot start more than a single VM any more;
> libvirt complains that Port 5900 is already in use:
> 
> [3/6034]mh at swivel:~ $ virsh start test01
> error: Failed to start domain test01
> error: internal error: process exited while connecting to monitor: ((null):9786): Spice-Warning **: reds.c:2499:reds_init_socket: listen: Address already in use
> 2017-05-24T14:26:01.824277Z qemu-system-x86_64: failed to initialize spice server
> 
> This affects both newly defined domains and older domains as well,
> including domains that used to happily run concurrently before may.
> There is a number of reports regarding this out on the net, mostly from
> Red Hat users, but Red Hat is on a fairly more recent version of libvirt
> than we are. Also, those bug reports all do suggest that this is a race
> condition when domains get started quickly after each other, but I am
> also seeing this when starting the second domain minutes after the first
> one was started, the system has settled down and a qemu process with
> "-spice port=5900,addr=127.0.0.1,disable-ticketing,image-compression=off,seamless-migration=on"
> has become visible in the process list.
> 
> I doubt that this issue is actually caused by libvirt itself. The last
> libvirt update was back in March, and I upgrade my sid about twice a
> month, and the breakage is fairly recent. I do create my domains with
> virt-manager's help, but virt-manager also hasn't been updated recently.

I'm not seeing this problem here but I'm lagging on some sid
updates (though I'm running libvirt 3.0.0-4). Once I have updated I can
try to reproduce but it's unlikely s.th. outside of libvirt.

Can you check in the domain log that qemu gets broken command line
already (one with colliding ports). Check for -spice port= in

    /var/log/libvirt/qemu/<domain>.log

If so you can try to increase debuggig esp. for qemuProcess and
qemuProcessSPICEAllocatePorts. See log_filters in
/etc/libvirt/libvirtd.log.

This doesn't seem to affect too many people otherwise I'd expect more
reports (and we then should bump the severity).
Cheers,
 -- Guido