►
From YouTube: KubeVirt Community Meeting 2021-12-01
Description
Meeting Notes: https://docs.google.com/document/d/1kyhpWlEPzZtQJSjJlAqhPcn3t0Mt_o0amhpuNPGs1Ls/edit#heading=h.u74oyrl72es0
B
B
B
Yeah,
that's
pretty
much
yay!
That's
not
what
I
want.
Okay,
I
don't
want
to
waste
everyone's
time
here.
If
somebody
else
can
share
the
meeting,
I
will
try
to
take
notes,
or
at
least
commentate
over
this.
B
All
right,
so
I
guess
I'll
welcome
everyone
to
the
keyboard
community
meeting,
I'm
going
to
be
leading
it
today.
My
name
is
david
bossel
as
first
off,
we
would
like
to
identify
any
new
members.
So
if
anyone's
joining
for
the
first
time
and
wants
to
say
hi
we'd
like
to
meet
you
and
say
hi
as
well.
B
If
you
haven't
already
fill
in
the
attendees
at
the
top
of
our
dock,
I'll
post
it
in
the
chat
just
so
everyone's
aware,
this
document
is
viewable
for
and
editable
by
anyone
in
the
cube
vert,
the
bell
google
group.
So
if
you
haven't
joined
that
yet
you
might
have
a
access
denied.
But
after
you
join
that
you
should
be
able
to
view
it.
B
C
The
okay.
D
C
B
What
I
read
it's
way
more
exciting
than
leading.
A
A
Will
they
be
using
cute
verts?
I
am
going
to
push
hard
for
it,
they're
in
a
big
battle
between
ex-cat
cloud
and
openstack,
and
neither
are
filling
their
their
hybrid
cloud
roles,
so
they
don't
know
what
to
do
well.
That.
B
I
do
see
it
it's
clear:
okay,
okay,
so
fill
in
your
attendees
fill
in
the
agenda,
notes
open
floor
and
all
that
and
I'll
give
everyone
just
another
minute
or
so
to
do
that.
I
just
requested
edit
access
so
that
I
can
fill
out
it.
B
Oh
I'm
sorry,
I
can
barely
hear
you.
I
think
you
said
you
requested
edit
access.
Yes,
you
should
be
able
to
edit
if
you
are
part
of
the
cute
vert
develo
google
group,
so
maybe
double
check
that
you
are
either
on
the
accounts
that
you
registered
or
that
you
have
joined
that
google
group.
B
C
Yeah,
the
we're
down
to
kind
of
the
last
stages
of
filling
out
all
of
the
paperwork
for
the
incubation.
C
One
of
them
is:
we've
been
kind
of
back
and
forth
the
toc
members
on
which
of
our
users.
They
consider
end
users
so
that
we
can
have
them
listed
under
end
users
as
opposed
to
vendors,
the
so
and
that's
been
kind
of
a
chronic
problem
with
the
cncf,
because
they
keep
changing
the
definition
of
what's
an
end
user,
basically
from
one
toc
term
to
another.
C
So
once
we
can
actually
have
another
meeting
with
our
sponsor
we'll
get
that
straightened
out,
and
then
we
should
be
able
to
announce
it
and
start
collecting
votes.
B
Excellent,
and
do
you
feel,
like
you,
have
enough
support
during
that
upcoming
meeting
with
the
sponsor.
C
Yeah,
the
one
thing
that
we
might
actually
want
to
do
with
the
community
here
is,
if
they
come
back
to
us,
because
one
of
the
questions
has
been
I
you
know
in
terms
of
the
end
users,
we
might
actually
come
back
and
they
might
want
a
list
of
some
of
the
commercial
end
users,
that
is
people
using
productized
versions
of
kubevert
from
various
companies,
including
like
red
hat
and
susie,
et
cetera,
and
if
they
actually
want
those
people,
then
we
might
come
back
to
folks
here
and
ask
for
a
few
names.
B
C
But
yeah
I
mean
I
there's
something:
yeah,
it's
honestly
less
important
than
contributing
to
the
community
the
whole.
The
purpose
of
the
end
user
requirement
is
honestly
just
to
make
sure
that
the
software
is
useful.
C
Oh,
my
goodness
the
I
I
mean,
that's
really
what
it
is,
the
just
so
people,
if
you've
been
around
open
source
for
a
long
time,
there's
been
a
few
organizational
mistakes
in
the
world
of
open
source.
C
C
B
I
understand
better
now,
that's
really
unfortunate.
The
nature
of
these
two
projects,
like
you
know,
c,
plus
plus
specifications
and
q
vert,
are
very,
very
different.
C
They
they
are
but
organizationally.
If
you're
looking
at
it,
you
know
you
know,
new
development
and
open
source
is
still
driven
by
vendors,
right
and,
and
so
the
cncf
and
the
lf
wants
to
make
sure
that,
as
things
progress
towards
hey,
this
is
something
that
not
just
the
contributors,
but
in
fact
the
linux
foundation
is
standing
behind,
which
is
is
what
happens
when
you
graduated
level
that
those
things
are
things
that
actually
have
broad
adoption.
C
If
you
follow
me,
yes,
the
which
we
know
that
which
cooper
does
right,
we
just
need
to
show
it
on
in
a
document.
So.
C
C
Yeah,
so
I
might
come
back
to
the
dev
list
if
it
turns
out
that
they
do
want
us
to
list,
because
we
had
made
the
assumption
that
they
did
not
want
us
to
list
productized
cooper
users,
but
but
based
on
some
things
that
our
sponsors
said
they
may
actually
in
which
case
will
come
back
and
say:
hey?
Could
you
all
drop
some
names
in
here
perfect.
B
C
Yeah
I
wanted
to
introduce
a
new
jewish
community
member
who's
volunteered
to
try
to
run
this
meeting
next
week.
Kat
the
cat,
the
with
chris
getting
busy
with
other
things.
That's
that's!
It's
both
gonna
be
some
extra
work
for
some
of
the
rest
of
us,
but
but
also
an
opportunity
for
other
community
members
to
step
up,
and
I'm
really
happy
that
that
one
of
them
already
has
so
meet
kat.
E
Oh,
but
it's
mostly
educational
and
talent
pipeline,
focused
for
people
who
don't
have
access
to
funding
for
east
high
aws
and
such
so
aimed
at
education
institutions
and
things
like
that.
That
containercraft.
B
B
B
Okay.
Currently,
the
signal
is
term
launcher
and
the
following
happens:
launcher
edits
its
domain
metadata.
D
B
D
Okay,
great,
so,
basically,
I
wanted
to
know
that
what
I'm
doing
is
to
the
right
direction,
because
I've
been
talking
to
vladik
and
we
try
to
understand
why
it
was
implemented
this
way
to
to
begin
with
so
yeah
just
would
like
to
hear
your
thoughts,
or
maybe
some
history
on
why
it
was
implemented
this
way
and
if
you
think
it's
the
right
direction,.
B
Yeah,
I'm
thinking
about
that.
I
understand
what
you've
got
that
so
just
to
summarize
really
quickly.
What
happens
today
is
when
vert
handler
or
excuse
me
burt
launcher,
which
is
the
pod,
that
the
the
virtual
machine
lives
in
and
receives
a
signal
like
six
term,
it's
going
to
set
a
file,
and
it's
like
directory
tree,
which
then
bert
handler
observes
to
see
the
graceful
shutdowns
occurring
and
vert
handler
is
actually
calling
the
acp
shutdown
kind
of
it's
the
thing
that's
triggering
it
rather
than
vert
launcher.
B
Knowledge
that
handler
has
launcher,
wouldn't
have
that
influences
how
the
shutdown
occurs.
That
would
be.
The
thing
that
I
would
be
looking
at
here
is
trying
to
understand
if
there's
a
reason
that
this
logic
was
split
because
there's
information
only
birth
handler
has
and
off
the
top
of
my
head,
I'm
not
aware
of
any.
B
D
Why
would
you
hear
that?
I
think
that
what
vladik
wrote
in
in
the
in
the
dark
makes
sense
that
simply
because
vert
handler
needs
to
treat
other
other
shutdown
situations,
so
it
was
implemented
this
way,
just
because
other
other
graceful
related
stuff
was
were
implemented
in
the
handler
okay.
So
if,
if
that's
the
only
reason
that
I
guess
it
is,
it
does
make
sense.
Okay
just
wanted
to
see.
If
somebody
has
something
interesting
to
say
about
it,.
B
B
The
evolution
of
kufert
was
that
we
started
looking
at
virtual
machines
as
pods
and
looking
at
the
vmi
as
being
kind
of
an
ephemeral
virtual
machine
similar
to
a
pod
which
is
ephemeral
and
over
time
we
split
the
vmi
into
a
vm
and
vmi.
So
the
vm
is
a
stateful
thing.
The
vmi
is
the
ephemeral
one
shot
thing
similar
to
the
pod,
so
the
vm
represents
a
stateful
virtual
machine.
B
In
the
early
days
of
cubert,
we
created
a
vmi
replica
set
which
allows
us
to
scale
out
replicas
of
the
ephemeral
virtual
machines,
the
bmi,
but
we
haven't
had
a
great
solution
for
scaling
out
the
stateful
virtual
machines,
which
is
really
what
people
are
probably
more
more
familiar
with
coming
from,
like
aws
or
gcp,
or
something
like
that.
B
So
the
virtual
machine
pool
essentially
works
like
that.
It's
going
to
allow
us
to
scale
out
safely
virtual
machines
and
create
update
strategies,
and
things
like
that
for
kind
of
a
a
pool
of
virtual
machines
that
we
want
to
treat
a
little
bit
differently
than
you
would,
for
example,
pods
in
deployment
or
pods
in
a
replica
set
or
dmis
in
a
virtual
machine
recommends
that
so
that's
coming
down
the
line.
I
think
that
that
will
be
hopefully
adopted.
B
B
B
B
G
A
He's
basically
running
the
virtual
machine
within
the
vmware
environment,
vmware
vsphere.
B
D
A
I
think
the
user
had
multiple
things
going
on.
He
was
wondering
why
he
it
seemed
at
first.
He
couldn't
start
the
virtual
machine
because
his
his
host
was
a
virtual
machine
itself.
B
B
All
right,
so,
if
you
are
interested
in
contributing
a
talk,
bostom
2022,
it
looks
like
it's
gonna,
be
hosted
virtually
on
february
5th,.
C
C
If,
if
you
do
get
a
talk,
accepted
I'll,
happily
send
you
a
t-shirt,
the
but
yeah.
This
is
a
good
audience
for
this,
because
it's
a
virtualization,
a
dev
room
and
they
already
explicitly
include
kubrick
in
that
which
not
all
virtualization
events
do
so.
It
would
be
nice
to
have
a
couple
of
talks
in
there.
C
C
I
don't
know
how
attendance
at
virtual
fosdem
has
been
the
compared
to
the
real
life
event
compared
to
the
in-person
event.
That
is,
but
on
the
other
hand,
you
don't
have
to
travel
anywhere,
it's
the
right
audience
to
hear
about
cooper
and
if
you're,
if
for
people
unfamiliar
with
fosdem,
they
tend
to
lean
towards
the
new
tech,
experimental
et
cetera.
C
So
it's
a
better
audience
for
talking
about
new
features.
Next
release
of
cooper,
you
know
weird
use
cases
than
necessarily
just
what
is
cooper.
C
I
don't
know
is
fabian
on
the
call.
C
The
given
the
inclusion
of
cooper
in
the
list
of
technologies,
I
suspect
that
we
have
for
the
red
hat
contingent
here.
Fosdem
is
one
of
the
events,
that's
officially
sponsored
by
red
hat,
so
for
hatters
that
have
wanted
to
attend
in
person
it's
pretty
much
automatically
approved.
C
So
I'm
I
work
for
anybody
who
doesn't
know
me.
I
work
for
red,
hat's,
open
source
practice
office,
so
it
has
not
involved
me
if
somebody
has
gone,
but
I
suspect
that
we
have
given,
given
that.
Why
else
would
we
be
listed
in
their
list
of
technologies.
B
C
So
I'll
put
a
link
in
the
in
the
minutes
for
that
excellent.
B
And
the
last
mailing
list
post
is
about
extending
cloud.net.
This
is
something
I'm
giving
attention
to
right
now.
So
ryan
from
nvidia
wants
to
add
some
information
to
cloud
and
net
about
the
instance
type,
which
would
be
the
name
of
the
flavor.
So
we
have
this
new
flavor
api
for
virtual
machines
and
we
will
be
passing
in
the
flavors
name
as
cloudinet
metadata,
so
it
would
tell
us
if
you're
running
a
virtual
machine,
you
would
have
some
information
cloud
and
network
data
about
what
flavor
was
used
to
start
your
your
guest.
B
B
All
right,
so
I
will
look
through
this
link,
so
vms
cannot
reboot
when
emulation
is
used.
That
seems
to
be
a
problem.
B
For
some
context
here,
emulation
mode
is
not
something
that
we
seem
to
test
very
well.
So
we
have
seen
problems
with
this
in
the
past.
So
when
used
emulation
is
set
to
true
vms
are
able
to
soft
reboot.
So
that
means
rebooting
within.
B
H
I
think
they
just
do.
We
just
do
a
augusta
reset
or
reboot,
but
the
point
was
is,
I
think
the
point
is
that
the
memory
it
keeps
consuming
memory
and
we
found-
I
think
he
found
out
that
if
you
try
to,
if
you
try
to
start
another
one,
then
it
will
just
block
the
boot.
So
it's
like
something
with
memory.
I
think
it's
that's
the
this.
It's.
B
That's
really
suspect
so
we're
saying
that
rebooting,
the
gas
internally,
the
cumu
process,
begins
to
consume
twice
the
memory
resources
from
the
pre.
So
it's
it's
not
freeing
or
reusing
the
memory
resources
from
the
previous.
B
That's
I'm
not
sure.
I
completely
understand
I.
B
H
I
don't
know,
I
know
it
happens
on
a
conformance
test
that
one
of
the
one
of
the
users
they
want
to
to
pass.
It.
H
Okay,
oh
and
they
run
it
on.
I
think
they
run
it
on
aws,
but
it's
not.
I
don't
think
it's
it
matters
because
he
said
it.
It
was
able
to
reproduce.
B
If
it
doesn't,
it
might
be
a
chemo
bug
if
the
resources
aren't
being
used
efficiently,
but
we
need
to
know
if
it's
us
all
right
next,
one
how
to
attach
nvme
devices.
I
Yeah,
but
in
general
we
are
not
going
to
support
multiple
devices
in
the
same
in
the
same
imu
group
just
going
to
support
one.
That's
one,
one
map
to
another,
like
one
mmu
group
to
a
device.
J
B
I
So
sometimes
you
can
plug
something
into
into
a
bridge
some
kind
of
device
into
a
bridge,
and
then
that
bridge
will
be
represented
and,
like
all
of
these
devices
in
that
bridge,
will
be
represented
under
one
immu
group.
And
then,
if
you
pass
that
all
of
these
devices
together
to
a
virtual
machine,
then
they
will
all
have
access
to
their
to
their
own
to
the
memory
of
each
other
and
yeah
kind
of
problematic.
B
So
it
looks
like
with
fedora
35.
The
ip
address
is
no
longer
working.
Excuse
me,
the
ipv
address
is
no
longer
being
reported
via
the
guest
agent
and
the
way
that
we
expect.
B
B
All
right,
triage
accepted
this
would
be
a
great
bug
for
somebody
to
take
a
look
at.
Is
anyone
interested
in
looking
at
this.
B
So
what
is
happening
just
top
level
glance
at
this
is
that
with
fedora
35,
when
the
guest
agent
is
installed,
we
are
not
getting
the
ip
address
reported
back
as
we
expect.
So
we
have
this
logic
that
uses
the
guest
agent.
We
request
some
information
from
the
guest
agent
and
that
gets
reported
back
to
vert
handler,
which
then
fills
that
ip
address
information
in
onto
the
vmi
spec
and
the
status
section.
Sorry
that
inspect
the
status
section,
so
that's
being
left
blank,
probably
because
we
know
that
we've
connected
to
the
guest
agent.
B
We
know
we've
requested
this
information
and
we
didn't
get
anything
usable
in
response,
so
either
the
data
structure
that's
been
reported
back
from
the
community,
gas
station
has
changed
or
something
else
has
occurred
where
we
don't
get
that
ip
address
from
the
agent
anymore.
H
So
so
now
we
are
now
working
on
on
something
very
odd
there
that
happens
and-
and
we
have
another
bug
about
it
like
if
someone
changes
the
mac
address
inside
the
guest
and
it
caused
a
lot
of
mess,
but
so
you
can
assign
it
to
me.
It's
idi
dev.
I
think,
and
then
I'll
I'll
check
what
to
do
maybe
to
pass
it
to
the
person
or
fix
the
current
pr.
B
Perfect,
thank
you
thanks
yeah,
that's
gonna!
I
mean
fedora
is
kind
of
our
canary
here
when
things
happen
in
the
latest
fedora
that
are
odd.
It's
just
a
matter
of
time
before
that
trickles
out
to
a
larger,
larger
audience.
So
it's
great
for
us
to
get
ahead.
K
H
Guest
is
changing
the
market,
so
the
virtual
nigda
it
will
mess
up
all
our
logic
pretty
badly,
and
so
we
asked
the
guest
agent
developer.
I
think
it's
the
under
chemo
project.
We
asked
them
to
to
help
us
and
report
back,
not
the
not
the
soft
soft
mac
by
the
hardware,
mac
and
prep.
Maybe
even
the
pci
addresses
so
we'll
be
able
to
to
have
a
better
matching
capability
here
and
we'll.
B
All
right
signed
to
you
thanks
for
looking
into
that,
the
the
last
one
I
guess
just
going
a
week
back-
is
for
control
ssh
the
fedora
35
guest,
it's
not
working,
but
we've
we've
kind
of
figured
out.
We
know
why
at
least
that's
occurring
and
there's
some
workarounds,
and
then
we
have
fixes
and
coming
from
golang
itself
and
the
crypto
libraries
to
make
this
work
moving
forward.
B
K
Hey
david,
I
wanted
to
hey
it's
ryan.
I
want
to
bring
one
other
thing
up
is
this
is
so
if
you,
I
think
a
few
meetings
ago,
I
brought
up
the
the
issue
6756,
which
is
the
the
res
control
qos
performance,
sensitive,
workflows
issue.
K
I
wrote
a
comment
in
there
before
thanksgiving,
but
I
I
wanted
to
just
bring
it
up.
Did
you
post
the.
K
Yeah
so
the
we
there's
been
some
discussion
on
this
and
you
know
I've
talked
a
lot.
I
can
others
I
so
I
just
wrote
some
kind
of
ideas
that
I
had
in
mind
for
how
this
could
look
and
I
just
want
to
get
some
thoughts
on
on
you
know.
Maybe
you
know
what
if
this
makes
sense
to
people
or
whether
it's
the
right
direction
or
whatever?
Maybe
some
young
people,
people
think
of
the
options.
B
I
Yeah,
sorry,
actually,
no,
but
I
just
didn't
get
to
it.
I'm
sorry
I'll
get
to
it.
B
Can
you
summarize
quickly,
ryan
what
yeah
this
was?
I
looked
at
it
last
week,
thanksgiving
happened
and
my
mind
does
not
remember.
K
Because
I
I
so
basically
we're
where
we
were
on
this
is
like
that
there,
there
really
isn't
a
way
like
kubernetes
doesn't
support
the
ability
to
to
control
the
l3
cache
on
on
on,
like
a
pneuma
node,
we
don't
have
like
the
ability
to
like
control
it
at
the
pod
level
and
this
this
work
you
like,
we
need
to
do
it
in
cpu
manager,
and
I
mean
probably
a
lot
of
other
things
that
need
to
go
into
this,
and
so
a
lot.
A
lot
of
things
would
need
to
happen.
K
K
But
but
I
I
ultimately
expect
kubernetes
to
be
the
real
vehicle,
but
to
at
least
enable
it
in
some
way,
and
so
I
because
ultimately
like
there
could
be
custom
solutions
that
that
could
look
at
that
are
out
there
that,
instead
of
using
like
cpu
manager
or
whatever
to
solve
this,
so
so
I
I
made
three
suggestions
in
terms
of
ways
to
solve
this
and
so
primarily
focused
on
like
enabling
sort
of
exposing
those
res
control
groups
to
through
the
kubernetes
through
the
covert
api.
K
So
the
one
of
them,
the
simplest
one.
The
first
one
is,
that
is
to
make
the
vert
launcher
privileged
and
basically
we
could
mount
the
rest
control
pseudo
file
system
and
to
be
vert
launcher.
And
then,
let's
you
know
their
launcher,
do
all
the
work
and
settle
the
parameters.
This
would
probably
be
needed
behind
a
flag.
Considering
that
we
need
to
change
the
privilege
model
for
convert
launcher,
that's
one.
The
second
one
was,
we
could
add
some
tunables
into
the
bmi's
spec,
the
virtual
machine
api
vert
handler.
K
Could
let
the
cpu
manager
handle
you
know
all
the
work
it
does
now
assigning
the
cpus
to
the
pod,
and
then
we
could
look
at
the
tunables
that
were
set
on
the
on
the
bmi
and
then
apply
some
rest
control
settings.
So
that
would
need
to
be
word
handler.
K
We
need
to
do
some
accessing
the
host
and
do
some
work
to
set
that
up,
and
then
the
third
option
was
that
we
have
some
tunables
on
the
vmi
on
the
virtual
machine
and
then
we
assume
that
either
cpu
manager
or
some
other
solution
handles
everything,
and
we
just
all
we
do
is
we
just
take
the
the
tunables
and
just
bring
propagate
them
all
the
way
down
to
the
to
the
domain
and
then
on
the
guest
on
the
in
the
launcher.
K
B
So
if
we,
if
we
created
a
feature
gate
that
allowed
privileged
launcher
pods
and
you
had
a
a
hook,
a
pre-start
hook,
would
that
technically
allow
you
to
express
the
configuration
that
you're
looking
for
yeah.
K
B
Path
for
you,
ryan.
This
is
a
really
complex
topic
because
it
involves
you
know
multiple
projects
and
we've
seen
that
it's
really
difficult
to
have
these
conversations.
Take
they
don't
take
weeks.
They
don't
take
months.
They
actually
take
years
to
to
really
invoke
change.
We've
seen
the
same
thing
with
like
topology
manager
and
schedule
with
pneumonia,
aware
scheduling
and
things
like
that.
Making
making
changes
in
these
areas
are
things
that
takes
quite
a
bit
of
time.
B
So
all
that
said,
I
think
that
we
would
be
in
favor
of
enabling,
whatever
you
need
to
make
this
work
as
optional
feature
gates
and
hooks
for
your
own,
like
third-party
code.
K
So
it
sounds
like
so
option,
one
which
was
so
make
it
privileged
and
then
and
then
hooks
and
so
there's
so
is
it
and
from
your
perspective,
would
you
say
that
like
is
it
option
one?
You
know
the
most
favorable
in
terms
of
like
well
like
from
your
perspective
that
because
it's
like
we're
not
changing
the
api
like
we're,
not
you
know
we're
not
looking
at
like.
B
Because
it's
achievable
all
the
other
ones,
don't
look
achievable
in
the
near
or
medium
term.
It's
something
that
we
can
start
discussions
about
with
the
because
it
involves
cpu
manager
and
other
other
things
are
outside
of
our
control
today.
So
essentially,.
K
B
K
B
That
option
two:
while
it's
technically
totally
possible,
it's
not
something
that
we
can
maintain
long
term.
So
that's
my
hesitance
with
option.
Two.
Okay.
K
H
K
K
B
Three,
I
don't
have
a
problem
with
three
if
we're
just
saying
an
annotation
the
bmi
and
you
have
a
node
level,
daemon
set.
That's
interpreting
that
I
mean
that's
that
doesn't
matter
to
me
from
a
maintenance
perspective
of
cubert.
I
don't
care
at
all.
That's
totally
fine.
I
Sorry,
I'm
just
trying
to
understand,
even
if
you
have
annotation
or
anything
like
this,
I
mean
will
libra
be
able
to
set
this
this.
My
understanding
essentially
right.
You
just
want
to
send
this
to
propagate
these
configurations
to
libert
and
then
let
liberton
to
set
this
is
that
what.
K
Well,
so
the
well,
so
we
would
like
well,
so
I
guess
so
here's
my
expectation
is
that,
like
we
would
do,
we
would
do
everything
like
in
terms
of
rest
control.
It
would
already
be
done
for
the
pod.
The
only
thing
that
we
would
do
is
is
we
would
take
the
any
of
the
the
expectations
and
expose
it
delivered.
So
it's
actually
no
different
than
not
an
option
one.
It's
just
we're.
Not
we
don't
need
to
do
anything
on
the
host.
That's.
B
It
sounds
to
me
like
option
three,
it's
just
taking
the.
I
might
be
totally
wrong
about
this.
Taking
the
privileged
stuff
out
of
the
tripod,
putting
somewhere
else.
B
And
then
option
three
just
involves
setting
the
bits,
so
you
really
option
one
and
three
are
kind
of
the
same.
You
can
use
a
hook
for
the
live
vert
part,
so
you
get
your
custom
live
vert
logic
and
you
can
send
annotation
to
bmi
perhaps
to
to
do
the
special,
privileged
risk
control
stuff
externally.
That.
K
Yeah
yeah
they're
similar,
I
mean
the
only
I
guess
the
difference
is
this:
three,
like
I
don't
expect
for
launcher,
have
to
be
privileged
because
we're
not
going
to
be
doing
anything
with
res
control.
We
would
just
it
would
already
be
done
for
us
on
the
pod
we'd
assume
it's
there.
We
just
we're
gonna
pass
any
of
the
information
to
deliver
like
you
know,
whatever
the
cash
settings
are,
and
they
should
already
be
assigned
to
the
pod,
we're
just
letting
the
guests
know
about
them.
I
Right
but
libert
will
okay,
so
that's
my
question.
Liberty
will
run
invert
launcher
that
doesn't
have
any
privileges,
even
though
you've
set
up
something
on
the
on
the
pod.
I
K
So
that
was
okay.
Maybe
I
had
the
wrong
expectation.
The
wrong
assumption
here
because,
like
why
would
liver,
if,
if
it
was
already
done
for
liver,
so
liver
is
assuming
then
that
it
when
it
you
execute,
when
you
provide
this
to
the
domain
like
to
the
dom
xml
that
it's
going
to
have
to
do
all
the
rest,
control
work
itself.
I
I
guess
this
is
not
really
100
related
cpu
manager.
The
only
thing
is
that
related
to
cpu
manager
is
the
knowledge
on
which
cpus
do
you
need
to
configure
that
option?
I
But
if
you
take
this
out
of
the
equation,
then
then-
and
you
know
on
which
cpus
they
need
to
do
it,
then
you
just
tell
libert:
where
do
you
want
that
and
it
will
be
done,
but
it
will
you
need
to
interact
with
the
rest
controllers.
It's
fine.
My
understanding.
K
I
What
david
suggests
is,
I
think
it's
the
best
way
forward,
just
to
enable
to
have
some
kind
of
control
to
enable
privileged
libert
and
then
and
then
either
hook
or
annotation,
to
enable
that
okay.
B
Okay,
I
would
suggest
just
doing
a
rapid
prototype
of
whatever
it
takes
to
get
this
to
work
for
you
and
then
gaining
understanding.
This
end-to-end-
maybe
maybe
you've,
already
done-
that
understanding
end-to-end
of
what
it
takes
to
get
this
work
without
changing
a
lot
of
qford
or
anything
else,
and
then
we
can
fill
in
any
gaps.
So
the
gap
will
probably
be
that
you
need
us
to
enable
privileged
mode
like
as
a
feature
gate.
We
can
certainly
talk
about
that.
If
there's
anything
else.
B
Actually
modifying
deliver
domain
xml.
I
think
we
would
prefer
that
to
be
in
a
hook
like
a
sidecar
hook
of
some
sort,
but
that's
something
we
can
look
at
as
well,
but
that
seems
less
likely
that
we'll
be
able
to
help
much
there
do
you
feel,
like
you,
have
a
at
least
a
path
forward.
Now
that
makes
sense.
K
Yeah,
okay,
so
I
think,
like
the
most
appealing
at
least
so
I've
I've
done
a
little
bit
of
work
on
this,
and
that's
what
led
me
to
one.
But
let
me
try.
Let
me
do
this
again
with
just
to
make
sure
that,
like,
like
my
assumptions,
are
criteria
are
yeah
just
to
make
sure
that,
because
I
think
one
in
yeah,
I
think
I
have
a
password
to
answer
your
question.
K
B
All
right
sounds
great.
I
think.
Okay,
we
are
at
the
top
of
the
hour
here.
Unless
there's
anything
else,
I
think
we
can
close
out
the
meeting.