Cloud image/ kernel/OVS followup

[Robyn Bergeron]

----- Forwarded Message -----
>  From: "Josh Boyer" <jwboyer(a)fedoraproject.org&gt;
>  To: "Fedora community advisory board"
> <advisory-board(a)lists.fedoraproject.org&gt;
>  Sent: Wednesday, January 22, 2014 5:03:09 AM
>  Subject: Re: Proposal: Revision of policy surrounding 3rd party and non-free	software
>  
>  On Wed, Jan 22, 2014 at 5:15 AM, Christian Schaller <cschalle(a)redhat.com&gt;
>  wrote:
>  > The thinking is simply that we can solve this and we can solve it in a way
>  > that isn't detrimental to the user experience of Fedora users.
>  >
>  > It is already likely there will be some changes in the Fedora kernel policy
>  > due to the Cloud WG wanting a different kernel, so as part of the process
>  > going forward there will be negotiations between the different WGs and
>  > with the Fedora kernel team to figure out what the different needs are,
>  > what is viable within the constraints of the resources we have, what are
>  > the possibilities and challenges each available solution bring.
>  
>  Just a note, the kernel efforts around the Cloud product thus far have
>  mostly be in terms of packaging, not a _different_ kernel.  Mostly
>  splitting it up into kernel-core and kernel-drivers, etc.  That
>  doesn't change policy.  It has also been discussed on the kernel and
>  cloud lists, as well as in the cloud meetings, just so that people
>  don't think we're off making changes without discussion. 
Along these lines (and a GIANT ADVERTISEMENT to those of you who don't follow the
advisory-board list that that is srs discussion going on, heh) - 

I know we talked last week about the inclusion of OVS-related kernel things in the Cloud
image and it was debated whether or not it was needed. (This was mostly me pondering aloud
and not being super-sure, and less of a debate, I suppose. :D) 

My investigation into this - which admittedly consisted of mentioning it to my significant
other (who is on this list, and does OpenStack things) over dinner a few nights ago -
resulted in the following:

* You guys were right, it's really only needed on the host
* Unless you are doing something along the lines of Triple-O (OpenStack-on-OpenStack,
https://wiki.openstack.org/wiki/TripleO) - where you have OpenStack running in the
traditional way, and then one or more OpenStack(s) running on *top* of that
infrastructure, in which case you'd be doing nested virtualization, and then the
additional OpenStacks running on top of the primary OpenStack would need OVS in the
kernel.

So I'm not really sure how in the heck we'd want to handle that, or what we'd
recommend for those "secondary" OpenStacks - the Cloud or Server product - I am
sure some folks on this list have opinions being more in tune with that particular
OpenStack project than I am. 

The third "investigatory point" - which I'm sure most of you guys know
anyway - is that including it in the kernel is not really something that is significant in
terms of size. But I know we are trying to get down to as small of a footprint as
possible. I also don't know if it presents an additional point of failure (could OVS
go haywire and break the kernel if it's not being utilized, but just present? more
technical things I am not sure about) - but if it does I am sure that might be something
to consider.

mestery, cdub, anyone else... if there are other thoughts (or, uh, corrections) you folks
have - they are (as always) welcomed :D

-robyn

ps. If you guys already figured all of this out and I'm just blind... ignore me :D


>  
>  josh
>  _______________________________________________
>  advisory-board mailing list
>  advisory-board(a)lists.fedoraproject.org
>  https://admin.fedoraproject.org/mailman/listinfo/advisory-board

[Steven Dake]

On 01/22/2014 06:54 AM, Robyn Bergeron wrote:
>  ----- Forwarded Message -----
> > From: "Josh Boyer" <jwboyer(a)fedoraproject.org&gt;
> > To: "Fedora community advisory board"
> <advisory-board(a)lists.fedoraproject.org&gt;
> > Sent: Wednesday, January 22, 2014 5:03:09 AM
> > Subject: Re: Proposal: Revision of policy surrounding 3rd party and
> non-free	software
> >
> > On Wed, Jan 22, 2014 at 5:15 AM, Christian Schaller <cschalle(a)redhat.com&gt;
> > wrote:
> >> The thinking is simply that we can solve this and we can solve it in a way
> >> that isn't detrimental to the user experience of Fedora users.
> >>
> >> It is already likely there will be some changes in the Fedora kernel policy
> >> due to the Cloud WG wanting a different kernel, so as part of the process
> >> going forward there will be negotiations between the different WGs and
> >> with the Fedora kernel team to figure out what the different needs are,
> >> what is viable within the constraints of the resources we have, what are
> >> the possibilities and challenges each available solution bring.
> > Just a note, the kernel efforts around the Cloud product thus far have
> > mostly be in terms of packaging, not a _different_ kernel.  Mostly
> > splitting it up into kernel-core and kernel-drivers, etc.  That
> > doesn't change policy.  It has also been discussed on the kernel and
> > cloud lists, as well as in the cloud meetings, just so that people
> > don't think we're off making changes without discussion.
>  Along these lines (and a GIANT ADVERTISEMENT to those of you who don't follow the
> advisory-board list that that is srs discussion going on, heh) -
>
>  I know we talked last week about the inclusion of OVS-related kernel things in the Cloud
> image and it was debated whether or not it was needed. (This was mostly me pondering aloud
> and not being super-sure, and less of a debate, I suppose. :D)
>
>  My investigation into this - which admittedly consisted of mentioning it to my
> significant other (who is on this list, and does OpenStack things) over dinner a few
> nights ago - resulted in the following:
>
>  * You guys were right, it's really only needed on the host
>  * Unless you are doing something along the lines of Triple-O (OpenStack-on-OpenStack,
> https://wiki.openstack.org/wiki/TripleO) - where you have OpenStack running in the
> traditional way, and then one or more OpenStack(s) running on *top* of that
> infrastructure, in which case you'd be doing nested virtualization, and then the
> additional OpenStacks running on top of the primary OpenStack would need OVS in the
> kernel.
>
>  So I'm not really sure how in the heck we'd want to handle that, or what we'd
> recommend for those "secondary" OpenStacks - the Cloud or Server product - I am
> sure some folks on this list have opinions being more in tune with that particular
> OpenStack project than I am.
>
>  The third "investigatory point" - which I'm sure most of you guys know
> anyway - is that including it in the kernel is not really something that is significant in
> terms of size. But I know we are trying to get down to as small of a footprint as
> possible. I also don't know if it presents an additional point of failure (could OVS
> go haywire and break the kernel if it's not being utilized, but just present? more
> technical things I am not sure about) - but if it does I am sure that might be something
> to consider.
>
>  mestery, cdub, anyone else... if there are other thoughts (or, uh, corrections) you folks
> have - they are (as always) welcomed :D 
I'm not quite a kernel expert, but I think in addition to OVS support 
you would want netns enabled to support OpenStack on OpenStack for the 
Neutron project.

Regards
-steve

>  -robyn
>
>  ps. If you guys already figured all of this out and I'm just blind... ignore me :D
>
>
> > josh
> > _______________________________________________
> > advisory-board mailing list
> > advisory-board(a)lists.fedoraproject.org
> > https://admin.fedoraproject.org/mailman/listinfo/advisory-board
>  _______________________________________________
>  cloud mailing list
>  cloud(a)lists.fedoraproject.org
>  https://admin.fedoraproject.org/mailman/listinfo/cloud
>  Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct

[Josh Boyer]

On Wed, Jan 22, 2014 at 10:00 AM, Steven Dake <sdake(a)redhat.com&gt; wrote:
>  I'm not quite a kernel expert, but I think in addition to OVS
> support you
>  would want netns enabled to support OpenStack on OpenStack for the Neutron
>  project. 
We have NET_NS enabled.

josh

[Matthew Miller]

On Wed, Jan 22, 2014 at 10:08:15AM -0500, Josh Boyer wrote:
>  > I'm not quite a kernel expert, but I think in addition to
> OVS support you
>  > would want netns enabled to support OpenStack on OpenStack for the Neutron
>  > project.
>  We have NET_NS enabled. 
And this is something that we need for Docker too.

-- 
Matthew Miller    --   Fedora Project    --    <mattdm(a)fedoraproject.org&gt;

[Josh Boyer]

On Wed, Jan 22, 2014 at 8:54 AM, Robyn Bergeron <rbergero(a)redhat.com&gt; wrote:
>  The third "investigatory point" - which I'm sure most
> of you guys know anyway - is that including it in the kernel is not really something that
> is significant in terms of size. But I know we are trying to get down to as small of a
> footprint as possible. I also don't know if it presents an additional point of failure
> (could OVS go haywire and break the kernel if it's not being utilized, but just
> present? more technical things I am not sure about) - but if it does I am sure that might
> be something to consider. 
If it's modular (which it currently is) and nothing is using it, it
won't be loaded.  Nothing will break in that case.  If it is loaded
for some reason, but nothing is using it still then it should still be
relatively risk free.

josh

[Collins, Robert (Converged Cloud)]

TripleO isn't nested virt: The lower layer cloud is nova-baremetal or Ironic, so your
hypervisor hosts are running on bare metal, and they need ovs etc indeed.

Note that for Fedora based deploys we *start* with a fedora cloud image, so if it's
not possible to install ovs and physical hardware drivers into those images, it won't
be possible to deploy OpenStack using TripleO w/Fedora.

Cheers,
Rob

Robert Collins <rbtcollins(a)hp.com&gt;
Distinguished Technologist
HP Converged Cloud

[Robyn Bergeron]

----- Original Message -----
>  From: "Robert Collins (Converged Cloud)"
> <rbtcollins(a)hp.com&gt;
>  To: "Fedora Cloud SIG" <cloud(a)lists.fedoraproject.org&gt;
>  Sent: Wednesday, January 22, 2014 11:50:58 AM
>  Subject: RE: Cloud image/ kernel/OVS followup
>  
>  TripleO isn't nested virt: The lower layer cloud is nova-baremetal or Ironic,
>  so your hypervisor hosts are running on bare metal, and they need ovs etc
>  indeed. 
Okay! I stand corrected. Perhaps my nested virt comment was more like "a
similarity".. or not.. anyway. :)

>  
>  Note that for Fedora based deploys we *start* with a fedora cloud image, so
>  if it's not possible to install ovs and physical hardware drivers into those
>  images, it won't be possible to deploy OpenStack using TripleO w/Fedora. 
When you say "for Fedora based deploys we start with a fedora cloud image" -
this is *just* for the TripleO case, or is this for a "regular" (SingleO, hah)
deploy as well? 

Sorry if I'm being confusing (or if I'm just confused) :) I just know there are
plenty of projects out there who are developing new tech rather rapidly and not everyone
in the cloud SIG is intimately involved in all of them. Until we get clones. Working on
that. :) But "additional feedback" never hurts.

Thanks!

-robyn

>  
>  Cheers,
>  Rob
>  
>  Robert Collins <rbtcollins(a)hp.com&gt;
>  Distinguished Technologist
>  HP Converged Cloud
>  
>  _______________________________________________
>  cloud mailing list
>  cloud(a)lists.fedoraproject.org
>  https://admin.fedoraproject.org/mailman/listinfo/cloud
>  Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
>

[Collins, Robert (Converged Cloud)]

>  > Note that for Fedora based deploys we *start* with a fedora
> cloud image, so
>  > if it's not possible to install ovs and physical hardware drivers into those
>  > images, it won't be possible to deploy OpenStack using TripleO w/Fedora. 
>  When you say "for Fedora based deploys we start with a fedora
> cloud image" - this is *just* for the TripleO case, or is this for a
> "regular" (SingleO, hah) deploy as well? 
>  Sorry if I'm being confusing (or if I'm just confused) :) I
> just know there are plenty of projects out there who are developing new tech rather
> rapidly and not everyone in the cloud SIG is intimately involved in all of them. Until we
> get clones. Working on that. :) But "additional feedback" never hurts. 
>  Thanks! 
So, there are lots of ways to deploy OpenStack :). Folk that build golden images of their
hypervisor/control plane/network nodes etc with diskimage-builder will be starting from
$DISTRO cloud images. Folk that use packstack or RDO or crowbar or MAAS or whatever will
have whatever that deployment toolchain uses to get the OS on the metal.

The only unique thing about TripleO is that rather than use a different toolchain to
manage the baremetal, we went all-in on using the OpenStack native facilities for that.
(Thus OpenStack on OpenStack).

-Rob

Robert Collins <rbtcollins(a)hp.com&gt;
Distinguished Technologist
HP Converged Cloud

[Kashyap Chamarthy]

On 01/23/2014 12:42 AM, Robyn Bergeron wrote:
>   
[. . .]

>> TripleO isn't nested virt: The lower layer cloud is nova-baremetal or
Ironic,
>> so your hypervisor hosts are running on bare metal, and they need ovs etc
>> indeed.
>   
> Okay! I stand corrected. Perhaps my nested virt comment
was more like "a similarity".. or not.. anyway. :)

Was wondering the same. As, they were s

Side note: /me volunteers to do nested virt testing (KVM on KVM, Intel)
with rawhide/upstream bits of Kernel/KVM/QEMU. If you need any help on
that front, give me a yell.

(Shameless self-plug: Last update on that at LinuxCon EDI, 2013 [1])


>> Note that for Fedora based deploys we *start* with a fedora cloud image, so
>> if it's not possible to install ovs and physical hardware drivers into those
>> images, it won't be possible to deploy OpenStack using TripleO w/Fedora.
>   
> When you say "for Fedora based deploys we start with
a fedora cloud image" - this is *just* for the TripleO case, or is this for a
"regular" (SingleO, hah) deploy as well? 
>   
> Sorry if I'm being confusing (or if I'm just
confused) :) I just know there are plenty of projects out there who are developing new
tech rather rapidly and not everyone in the cloud SIG is intimately involved in all of
them. Until we get clones. Working on that. :) But "additional feedback" never
hurts.

I hear you, especially so in the Cloud land!


PS: A small side suggestion, Robyn. Your mails appear as long lines
making them rather difficult to read. (You might want to wrap them for
better readability.)

If you're using thunderbird as email client, here's how:
=====
1] Edit -> Preferences -> Advanced -> Click on "Config Editor..."
2] Click on "I'll be careful, I promise!"
3] Search for the string -- mailnews.wraplength -- and edit that value
to 72 (or 80 or whatever you prefer)
=====
(I do not remember if you have to restart thunderbird or not)



[1]
http://kashyapc.fedorapeople.org/virt/lc-2013/nested-virt-kvm-on-kvm-Clou...

-- 
/kashyap

[Kashyap Chamarthy]

On 01/23/2014 05:40 PM, Kashyap Chamarthy wrote:
>  On 01/23/2014 12:42 AM, Robyn Bergeron wrote:
> >
>  
>  [. . .]
>  
> >> TripleO isn't nested virt: The lower layer cloud is nova-baremetal or
> Ironic,
> >> so your hypervisor hosts are running on bare metal, and they need ovs etc
> >> indeed.
> >
> > Okay! I stand corrected. Perhaps my nested virt comment was more like "a
> similarity".. or not.. anyway. :)
>  
>  Was wondering the same. As, they were s 
Ugh, truncated sentence, I meant: "As they're slightly different from
each other, although nested virt can be a part of it, etc."

(I was notified by a family member about my truncated sentence ailment,
and I'm /still/ repeating my mistake. /me closes a few tabs to reduce
multi-tasking.)

>  
>  Side note: /me volunteers to do nested virt testing (KVM on KVM, Intel)
>  with rawhide/upstream bits of Kernel/KVM/QEMU. If you need any help on
>  that front, give me a yell.
>  
>  (Shameless self-plug: Last update on that at LinuxCon EDI, 2013 [1])
>  
>   


-- 
/kashyap