Fedora virt status

[Mark McLoughlin]

F-12
====

F-12 is still coming at us fast:

http://fedoraproject.org/wiki/Releases/12/Schedule

2009-07-28 Feature Freeze (26 days)
  2009-09-22 Beta (Final Development) Free (82 days)
  2009-10-20 Final Release Compose (110 days)

libguestfs
==========

Much libguestfs activity has continued on [e-mail]
lately and Rich Jones has posted two related F-12 feature pages:

https://fedoraproject.org/wiki/Features/libguestfs
  https://fedoraproject.org/wiki/Features/Xen_to_KVM_migration

FWN
===

Yet another weekly installment from Dale Bewley:

https://fedoraproject.org/wiki/FWN/Issue182#Virtualization

libvirt-0.6.5
=============

The libvirt folks have released another version of libvirt:

http://www.redhat.com/archives/libvir-list/2009-July/msg00060.html

This is mostly a bug fix release, though it includes some serious
  improvements for storage/NPIV and on the OpenNebula driver:

* New features
  - create storage columes on disk backend (Henrik Persson)
  - drop of capabilities based on libcap-ng when possible (Daniel Berrange)

qemu-fedora.git
===============

git is now being used to manage Fedora's qemu patches:

http://www.redhat.com/archives/fedora-virt/2009-June/msg00157.html

Glauber and I are going to start using git to manage the patches for
  the qemu package. The idea is that we'll use git's magical powers to
  make it easier for us to fix problems in patches, cherry-pick patches
  from upstream, re-base our patches to a newer upstream, etc.

http://git.et.redhat.com/?p=qemu-fedora.git

Already, this has proved useful re-basing to the latest
qemu-kvm-0.10.50 snapshot (kvm-87).

F-11 virtio-blk Performance
===========================

Charles Duffy, Michael Tokarev, Rich Jones and others recently
stumbled across the apparent cause of a dramatic slowdown in
virtio-blk performance in F-11 guests:

https://bugzilla.redhat.com/509383

Thanks for Michael Tokarev and Charles Duffy for the
  following suggestion which has mostly fixed this problem:

for f in /sys/block/vd*/queue/rotational; do echo 1 > $f; done

It seems the kernel folks deemed it prudent to turn KVM and Xen
paravirt disks into "non-rotational" disks in 2.6.29, thereby
preventing guests from attempting to optimize block I/O and, instead,
leaving it up to the host:

http://lkml.org/lkml/2008/10/27/84

No doubt there'll be more on this as people look deeper into the
issue.

KVM USB Passthrough
===================

A discussion at FUDCon lead to some tips on USB passthrough being
posted to the fedora-virt list:

http://www.redhat.com/archives/fedora-virt/2009-June/msg00182.html

"KVM, enabling iPhone-touting geeks everywhere to use iTunes" :-)

Wiki
====

Dan Berrange has pulled together a couple of new wiki pages which
people might find interesting.

The first outlines the history of virt in Fedora, with details of what
changes were made in each release:

https://fedoraproject.org/wiki/Virtualization/History

The second pulls together links to all the virt status reports and FWN
virt sections:

https://fedoraproject.org/wiki/Virtualization/News

Xen
===

Michael Young posted more Dom0 kernel builds here:

http://www.redhat.com/archives/fedora-virt/2009-June/msg00205.html

Michael also pointed out Jeremy Fitzhardinge's latest prognosis of
getting the dom0 patches merged upstream:

http://lists.xensource.com/archives/html/xen-devel/2009-06/msg01193.html

For now I've backed off putting anything much into the current merge
  window, or even the next, in favour of getting the tree into good
  shape for Xen users and working out what needs to happen to get a
  solid set of upstreamable patches.

Bugs
====

DOOM-O-METER: 242 open bugs a couple of weeks ago, 250 now

= New =

== kvm ==

https://bugzilla.redhat.com/508801
    kvm "emulation failed (pagetable)" during OpenBSD 4.6-beta guest
    boot (OOS shadow bug?)

It looks like this OpenBSD guest boot failure on F-11 might be due
    to an OOS shadow bug but, then again, that should have been fixed
    in 2.6.29.3.

https://bugzilla.redhat.com/509383
    mkfs.ext2 much slower with virtio-backed disks than with
    IDE-backed disks (NOT sparseness issue)

The problem here seems to be that virtio_blk is now defaulting to
    non-rotational mode since 2.6.29. This may also be the cause of
    bug #505695 and bug #504376.

== qemu ==

https://bugzilla.redhat.com/507899
    qemu's console-via-VNC generates invalid rect position(s)

qemu's VNC server sends bogus rectangle sizes causing vncviewer to
    abort.

https://bugzilla.redhat.com/507834
    Clock drift in RHEL5.3 KVM Guests on Heavily Loaded Servers

A report of x86_64 RHEL5 guests time drifting under KVM with the
    jiffies clocksource. Suggestion to boot with "notsc
    divider=10". Macelo also suggests "clock=tsc" or "clock=jiffies"
    with -no-kvm-pit-reinjection.

https://bugzilla.redhat.com/508258
    qemu init script does nothing useful by default

The qemu initscript for registering binary handlers for the
    usermode emulators does nothing unless you obtain the runtime
    environment for the emulator from somewhere else. This is
    lame.

https://bugzilla.redhat.com/509532
    Move /usr/bin/qemu-kvm into the qemu-kvm package

A minor packaging change we should make for F-12.

== libvirt ==

https://bugzilla.redhat.com/508645
    libvirt should use qemu's autoconnect mode for USB device
    passthrough

qemu has a nice autoconnect mode for USB passthrough which libvirt
    should probably use by default. Dan suggests that it should only
    be used with managed='yes'.

https://bugzilla.redhat.com/508317
    Allow sound devices to be used with svirt - tunnel sound over VNC

Covers the fact that with svirt we now disable sound cards until
    we can tunnel sound over VNC.

https://bugzilla.redhat.com/509458
    libvirtd default virtual network should not be enabled when
    libvirt installed in a guest

This issue has been known for a while now - if you install
    libvirtd in a guest, the default network is enabled and since it's
    using the same IP range as the default network in the host, things
    get screwed up. Suggestions on how to avoid having libvirtd or the
    default network enabled by default in guest installs welcome :-)

== virt-manager=

https://bugzilla.redhat.com/508665
    virt-manager (running as unprivileged user) doesn't let me install
    from NFS locations

Because virt-manager no longer runs as root, NFS installs don't
    work. Suggestion is that we'll use PolicyKit to allow virt-manager
    to do NFS mounts.

== virt-top ==

https://bugzilla.redhat.com/508197
    virt-top segfaults if run in Japanese

Now there's a bug that's right up Rich's alley!

= Resolved =

== qemu ==

https://bugzilla.redhat.com/508326
    Force use of /sys/bus/usb for all USB host devices

qemu is currently prefering to use usbfs instead of sysfs for USB
    passthrough. Fix sent upstream and built in rawhide.

https://bugzilla.redhat.com/506170
    removing virtualization packages leaves qemu-system-x86

This is NOTABUG - if 'yum groupinstall' pulls in packages via
    dependencies, then 'yum groupremove' doesn't remove them again.

== libvirt ==

https://bugzilla.redhat.com/506590
    libvirt should ignore NUMA cells with missing topology

The libvirt fix for this was committed upstream but a recent F-11
    numactl update is enough to fix the original issue for the
    reported, so this has been marked as fixed upstream.

https://bugzilla.redhat.com/507555
    libvirt starting a guest with ISO on NFS mount fails when
    unnecessarily setting SELinux file context

Patch from Tim Waugh sent upstream and included in
    libvirt-0.6.2-13.fc11 F-11 update.

https://bugzilla.redhat.com/505625
    libvirtd segfault during virsh dominfo on security model

This libvirtd segfault which occurs when you do "virsh
    capabilities" followed by "virsh dominfo" is fix in the
    libvirt-0.6.2-13.fc11 F-11 update.

https://bugzilla.redhat.com/493692
    Creating new VM; SELinux prevents opening iso image of install
    media
  https://bugzilla.redhat.com/499569
    Guest with source-less cdrom fails to start :: Failed to set
    security label

Two patches which we included in F-11, but neglected to send
    upstream and were then dropped accidentally when we rebased to
    0.6.4. All sorted out again now for F-12.

== virt-viewer ==

https://bugzilla.redhat.com/508825
    virt-viewer doesnt have "full screen" hot key

It seems sensible that we would hook up F11 for fullscreen and
    have virt-manager handle it, rather than passing F11 through to
    the guest. Dan thinks this is a dumb idea, so the bug is closed as
    WONTFIX.

[Farkas Levente]

Mark McLoughlin wrote:
> qemu-fedora.git
>
> git is now being used to manage Fedora's qemu patches:
>
> http://www.redhat.com/archives/fedora-virt/2009-June/msg00157.html
>
> Glauber and I are going to start using git to manage the patches for
>   the qemu package. The idea is that we'll use git's magical powers to
>   make it easier for us to fix problems in patches, cherry-pick patches
>   from upstream, re-base our patches to a newer upstream, etc.
>
> http://git.et.redhat.com/?p=qemu-fedora.git
>
> Already, this has proved useful re-basing to the latest
> qemu-kvm-0.10.50 snapshot (kvm-87).

wouldn't be it too much? there is an upstream qemu git, there is a
qemu-kvm git (which is also a redhat project) now fedora fork a fork!?

[Eduardo Habkost]

On Fri, Jul 03, 2009 at 07:03:51PM +0200, Farkas Levente wrote:
> Mark McLoughlin wrote:
> > qemu-fedora.git
> >
> > git is now being used to manage Fedora's qemu patches:
> >
> > http://www.redhat.com/archives/fedora-virt/2009-June/msg00157.html
> >
> > Glauber and I are going to start using git to manage the patches for
> >   the qemu package. The idea is that we'll use git's magical powers to
> >   make it easier for us to fix problems in patches, cherry-pick patches
> >   from upstream, re-base our patches to a newer upstream, etc.
> >
> > http://git.et.redhat.com/?p=qemu-fedora.git
> >
> > Already, this has proved useful re-basing to the latest
> > qemu-kvm-0.10.50 snapshot (kvm-87).
>
> wouldn't be it too much? there is an upstream qemu git, there is a
> qemu-kvm git (which is also a redhat project) now fedora fork a fork!?

This is just storing the (five) patches that already exist on the Fedora
package (that are all supposed to be submitted upstream) in a git
repository, to help manage them, not the creation of a fork.

[Richard W.M. Jones]

On Fri, Jul 03, 2009 at 05:30:07PM +0100, Mark McLoughlin wrote:
> libguestfs
>
> Much libguestfs activity has continued on [e-mail]

I also just requested a new mailing list for libguestfs, so that
development discussions and patches and so on won't disturb people
asking Fedora virt questions.

Rich.