[libvirt] libvirt-guests.sh without or with failing ACPI support

[Henning Schild]

Am Tue, 10 Dec 2019 12:01:24 +0100
schrieb Christian Ehrhardt <christian.ehrhardt at canonical.com>:

> On Tue, Dec 10, 2019 at 10:46 AM Henning Schild
> <henning.schild at siemens.com> wrote:
> 
> > Hi all,
> >
> > the systemd shutdown scripts work sequentially with a 300s timeout
> > (seen on Debian). If a VM does not have ACPI support, or the ACPI
> > support failed for some reason, you are looking at a 300s timeout
> > per instance for a host shutdown/reboot.
> > i.e. 10 instances without working ACPI = 3000s to shut down
> >
> > I think the systemd scripting should be parallel instead of
> > sequentially. So if you have many VMs without working ACPI you just
> > have to wait 300s in total for the host to shut down.
> >  
> 
> Hi Henning,
> this is configurable in /etc/default/libvirt-guests
> For example Ubuntu (otherwise using the same bits) changes that to run
> PARALLEL_SHUTDOWN=10
> SHUTDOWN_TIMEOUT=120

Sweet. I went for the PARALLEL_SHUTDOWN=10 and left the 300. Maybe the
default PARALLEL_SHUTDOWN value should not be 0 ?

> I never got bugs about that config being too aggressive.
> The change is old and as easy as:
> https://git.launchpad.net/ubuntu/+source/libvirt/tree/debian/patches/ubuntu/parallel-shutdown.patch?h=ubuntu/focal-devel
> Maybe you just want to open a bug with Debian to change the default
> config there as well?

No it is a bug in libvirt having the "wrong" defaults. And a bug in
ubuntu not fixing it upstream ;).

Thanks,
Henning

> Steps to reproduce:
> >  - star a VM that does not support ACPI
> >  - reboot the host and wait 300s for the VM to be shut down
> >  - now start it multiple times
> >  - wait multiples of 300s for the shutdown
> >
> > Expected behaviour:
> >  - no matter how many instances do not support ACPI, make it 300s
> > max because we shut them down in parallel
> >
> >
> > regards,
> > Henning
> >
> >
> > --
> > libvir-list mailing list
> > libvir-list at redhat.com
> > https://www.redhat.com/mailman/listinfo/libvir-list
> >
> >  
>