|
I CC'ed Wei and Ian, as I have an action to chase them on a xen-devel thread
Regards
Lars
== Attendees ==
Jim Fehlig
Joao Martins
Anthony Perard
Lars Kurth
Michael Glasgow
Cedric Bosdonnat
== Agenda ==
* Status of Bob Liu's tunneled migration patch for libvirt. I had some comments on V1 and don't recall seeing a followup.
- Bob Liu sadly left Oracle
- ACTION: Joao to follow up and report back to the group as he worked with Bob on this
* vif name too long when libvirt driver trying to create XEN HVM domain
Thus far no comments or ideas on a resolution
Lars: this was probably lost in the weeks prior to the release
Konrad: My comment was to go ahead on the first proposal
ACTION: Lars to ping Wei and Ian to have a look at it
ACTION: Lars to ping Konrad to chip in, if needed
*Unconditionally setting pae=1 for HVM domains
See https://www.redhat.com/archives/libvir-list/2017-January/msg00254.html
Issue was brought to attention via OpenStack community
This is a missing feature for Xen Domains that is needed for 64 bit HVM guests
Problem: in the old days of the xend stack, this would work because xend would always set pae=1.
For libvirt+libxl the default value is different, so it needs to be plumbed through and enabled.
Status: asked for how to handle this at libvirt level; waiting for feedback from libvirt community.
Two possible solutions
A) Preferred : always enable pae=1, but don't understand all implications
B) A tri-state implementation in libvirt. Unfortunately this would be VERY disruptive and become a huge patch affecting tests and openstack and is thus undesirable
Jim: Looking for input on whether always enabling this is fine and what the implications are
ACTION: Jim to call out question on xen-devel (as he has originally forgotten this)
* libvirt 3.0.0 release planned for Jan. 16
New Xen related features
- Per domain log files
- libxl driver will wed disk image format support (dependency with Xen 4.9)
- As of 1st of march new libvirt releases are beginning the 1st of each month again (Jan & Feb are exceptions)
Joao: libvirt 3 fixes detaching network interfaces
Lars to check whether this is mentioned in the docs and update if yes (Joao believes this is not the case)
* Ci Loop failures
The CI Loop started failing almost 100%
- Bob Ball believes this may have something to do with Neutron
- Antony believes this is not the case and may be a new race condition, but he does not yet understand what is going on
- More investigation needed
Antony will investigate and get back to the group
| 
