►
From YouTube: KubeVirt Community Meeting 2019-03-27
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Welcome
everybody:
this
is
the
Qbert
community
meeting
March
27th
2019.
This
is
the
last
of
the
daylight
savings
confusion
episodes.
Next
we
go
back
back
to
our
normal
time
and
jumping
into
the
agenda
notes.
The
first
thing
to
point
out
is
live
migration
by
phonic.
The
link
is
in
the
document
and
just
wanted
to
say
this
PR
is
merged,
so
this
is
the
last
major
portion
of
the
live
migration
effort.
So
thank
you
for
that
Blahnik.
Is
there
anything
that
you'd
like
to
add
to
that?
A
Okay
and
next
up,
we
have
the
key
vert
operator
oil
marketplace:
integration
by
Merck.
B
D
B
A
D
E
E
Maybe
Malta's,
depending
on
whether
it's
already
deployed
then
plugins,
like
OBS,
SRV
bits
and
pieces,
create
network
attachment
definitions
like
stuff
like
that.
So
this
all
goes
in
this
operator.
The
idea
EQ
you
deploy
it.
Just
this
operator
single
operator,
you
create
a
config
that
lists
the
types
of
functions
you
want
and
a
configuration
for
your
networks
and
then
everything
in
deploy
this
as
later.
It
is
not
a
hyper-converged
operator
which
is
a
separate
operator,
but
hyper-converged
operator
tender
need
to
go.
E
It
will
use
this
network
and
owns
operator
to
deploy
networking
pieces,
so
it's
kind
of
like
lots
of
layers
of
operators
here.
So
it's
this
operator
sealed
up
in
active
development,
we're
still
sticking
like
even
basic
pieces
in
place,
so
lots
of
work
going
forward
now
for
the
covered
core.
So
we
discussed
the
previous
times
several
things
around
motifs
annotations
for
networks.
So
one
was
the
the
plan
to
switch
to
JSON
format
from
a
simple
list
that
we
did.
It
is
merged.
E
Now
the
the
second
beat
that
the
required
this
JSON
format
change
was
propagating
the
MAC
address
the
custom
MAC
address
if
it's
selected
by
couvert
down
to
CNI
plugin
layer
through
these
annotations,
and
so
that
CNI
plugin
has
a
chance
to
configure
the
backend
to
work
with
this
custom
MAC
address.
Sometimes
it
works
because
you
don't
need
specific
configuration
on
CNI
layer
for
the
MAC
address,
but
sometimes
it
doesn't
like
one
example
of
a
plug-in
that
doesn't
work
properly.
It's
OpenShift
as
the
end.
E
You
cannot
use
custom
MAC
addresses
there,
I'm,
not
saying
that
now
you
can
because
I
think
they
still
need
to
add
support
for
the
for
the
MAC
address
attribute
in
the
in
the
network
annotation.
But
you
know
at
least
now.
This
path
is
unblocked.
So
basically,
this
piece
is
also
meshed.
So
now
CNI
plugins
get
the
MAC
address.
They
have
a
chance
to
act
on
it
and
then
assuming
everything
all
the
pieces
are
working
together.
E
F
We
are
trying
to
find
a
different
way
to
implement
our
B
and
C
right
now
we
are
a
some
type
of
abusing
the
a
cube
API
to
create
a
WebSocket
tunnel
to
the
build
ipi
and
then
run
the
cube,
exact
command
and
get
the
console
and
the
PLC,
and
this
is
something
that
we
want
to
change,
because
some
guys
from
a
openshift
tell
us
that
they
may
be
going
to
block
the
ability
to
open
a
web
socket
API
of
the
OpenShift
itself.
So
this
can
break
our
implementation
right
now.
F
D
Hello
Sebastian
first
note:
let's
keep
that
I'm
strongly
voting
for
keeping
that
functionality,
but
I
do
understand
that
we
have
a
need
to
provide
an
alternative
solution.
So
no
question
on
that,
but
just
that
OpenShift
is
deciding
to
block
something.
I
mean
those
are
crazy
people
right
that
doesn't
mean
that
we
can't
do
that
in
plain,
vanilla
communities
or
others
can
do
it
so
I
think
for
admin.
Purpose.
D
I
would
like
to
keep
that
implementation,
but
provide
an
alternative
which
is
then
maybe
oh
yeah
I
had
a
quick
scan
over
a
proposal
which
is
then
using
a
property
like
solution.
I'll
float
that
but
I
think
there
are
two
different
pairs
of
shoes
to
broaden
out
relative
path
or
to
draw
one
of
the
solutions.
Just
my
take
but
I'll
take
that
to
the
document.
D
F
Yeah
I
also
have
a
question
here,
maybe
to
Roman
I'm,
not
sure
who
implement
right
now.
Dairy
Council,
but
I
would
like
to
ask
why
they
a
exit
command,
is
going
from
the
cube
API
and
not
directly
for
the
field
CTL.
So
it
will
make
make
easy
for
the
a
console
one,
the
VNC
one.
The
main
problem
is,
if
you
start,
let's
say:
I,
try
to
start
10,
V
and
C's,
so
the
cube
API
goes
to
100
CPU
and
start
to
not
working
so
good.
On
my
development,
environment,
yeah.
G
G
G
F
G
No,
we
should
instead
go
that
that's
not
really
an
option,
because
we
have
no
I
mean.
First
of
all,
we
can
achieve
the
same
throughput
and
load
on
the
API
server
without
using
versatile,
Derek
cube,
CTL
directly
and
second,
but
keeps
delwood,
but
we
would
have
to
miss
basically
a
hack
to
reach
our
VMs.
We
wouldn't
have
a
real
API,
so
we
should
keep
the
API
and
instead
of
and
just
change
how
weird
API
connects
to
the
qumu
process.
G
D
Just
an
addition,
I
think
the
busting
one
of
the
reasons
why
we
don't
want
to
do
exit
directly
from
verse.
Cdl
is
because,
with
the
current
approach,
we
hide,
you
know
the
implementation,
because
the
mandate
we
say
we
connect
to
the
web,
the
console
of
the
VM.
That
is
what
we
do
right.
We
have
an
AVM
based
endpoint
or
VMI
based
endpoint,
and
if
we
change
that
to
do
the
exit
calls
invert
CTL,
then
you
need
to
maintain
all
that
logic.
You
know
what
is
the
implementation?
G
It
also
means
we
don't
have
a
real
REST
API
for
it
that
as
well.
It
means
with
you,
you
would
have
to
know
what
happens
in
the
virtual
I
mean
you
have
a
REST
API
when
it
is
for
executing
arbitrary
commands
in
an
arbitrary
environment.
It's
not
really
something
which,
wherever
god,
that's
not
even
possible
from
a
security
perspective
which.
H
A
E
So
is
there
still
no
answers
there,
but
hopefully
there
will
be
some
and
they
will
give
us
a
better
name
for
their
standard
and
then
we
can
use
it
in
culture.
So
if
you
have
complaints
about
any
of
the
names
proposed,
perhaps
now
a
better
place
to
complain
about
that,
it's
there
or
like,
because
if
we,
if
I
said
I,
think
if
we
said
that
they
should
decide
on
the
bed
by
the
name,
then
we
should
take
part
in
their
discussions
now,
since
we
moved
it
there,
it's
all
I
had.
D
I
mean
if
it's
about
the
acronym
right
are
you
personally
looking
into
participating
in
the
calls
to
see
that
we
get
in
an
official
acronym
for
what
that
standard?
Well,.
E
D
E
E
I
E
Out
of
the
PR
when
it
it
started
to
look
like
this
PR
is
going
nowhere
like
it's
slow,
so
to
unblock
the
integration
with
the
tungsten
fabric,
which
was
like
the
original
goal.
This
PR
we
moved
to
this
piece
out
immersed
it.
So
now
the
fabric
can
work
with
who
here
apparently
I
haven't
checked.
That
should,
and
we
we
are
left
with
just
this
anonymity,
which
is
arguably
not
related
to
at
all.
Okay.
A
J
J
J
K
That's
me,
so
let
me
summarize
the
question
we
have
a
hyper
be
supporting
qubit.
We
want
to
have
support
more
flags
and
to
make
windows
guest
works
nicely.
This
is
the
wall
purpose
of
fiber
tea
in
the
last
couple
of
weeks,
Nishan,
fortunately
arised,
that
is,
we
have
a
dependency
also
on
the
host
cardinal
on
the
machine,
which
is
running
all
our
containers
or
VMs.
We
need
to
be
sure
that
the
oscar
results
up-to-date
and
capable
to
actually
support
all
the
hyper-v
flags.
We
require
either
implicitly
using
templates
or
explicitly
from
the
user.
K
So
we
discussed
a
couple
of
approaches
in
the
last
days
and
we
reached
agreement
just
today
about
how
to
to
fix
this
in
manageable
and
containable
way.
Long
story
short.
We
are
going
to
expose
the
hyper-v
flags
like
like
we
do
for
CPU
features,
and
after
that
we
are
going
to
do
a
nonce
cube
root
core
to
make
scheduling
decision
based
on
the
hyper-v
feature
question
and
what
is
supported
by
the
node.
K
We,
this
is
actually
supposed
to
be
a
rare,
but
still
we
have
to
deal
with
that
and
to
do
to
deliver
a
good
experience
so
and
in
the
coming
days,
I
am
going
to
update
the
NFD
plugin,
which
might
determine
working
into
the
SSP
team,
is
maintaining
to
expose
those
flags
and,
after
that,
I
am
going
to
resume
tour
help
shape
another
Cubert
core
patch
to
make
use
of
those
levels.
This
is
what
I
have.
K
Yes,
this
is
actually
well.
They
did
at
the
QM
developer
or
Sayed
also
record
on,
for
example,
on
DEFCON
Brno
conference.
You
should
you
expected
to
encourage
to
turn
on
all
the
feature,
because
in
the
common
case,
they
help
worst
case
they
do
nothing
actually.
If
what
what
could
happen
if
the
host
kernel
doesn't
support
a
hyper-v
feature,
the
VM
won't
start,
so
it
is.
It
is
easy
to
detect.
K
K
A
D
I
thought
it
was
time
because
of
those
requests
in
the
meaningless.
So
just
a
word,
we
here
at
redhead
we're
working
on
something
called
a
hyper-converged
cluster
operator
that
is
actually
the
component
to
set
up
or
to
bring
out
Qbert
in
an
environment
which
we
think
is
the
or
to
us
right,
the
the
environment.
We
were
optimizing
it
for
so,
if
I
become
go
to
class
or
operator
we
want
to
unify,
you
know
to
ring
up
Qbert
to
bring
up
everything
that
all
the
infrastructure
components
for
Cuba
to
operate
nicely.
D
A
D
D
We
know
that
Qbert
has
a
thorough,
IV
support,
but
today
you
know,
if
you
deploy
Cooper
to
a
cluster
doesn't
mean
you
can
use
s
auravie
out
of
the
box
because
you
don't
know
of
any
SRB
capable
device,
plugins
or
supporting
infrastructure
was
deployed,
so
the
hyper-converged
operated
will
contain
all
the
components
you
need
in
order
to
use
all
all
all
relevant
features.
From
our
perspective
from
our
opinionated
view
to
use
Qbert-
and
that
includes
you
know,
you
want
to
use
storage,
so
we
we
may
decision
on
what
storage
we
provide.
D
You
need
to
use
networking,
so
we
make
a
decision.
What
networking
infrastructure
you
should
be
using,
which
is
motors
in
our
case-
and
we
probably
will
deploy
also
the
SRV
related
stuff
and
we'll
deploy
an
FD
for
you,
so
that
you
can
schedule
according
to
you
windows
needs
that
you
can
pass
through
an
S
or
a
V
device
if
you
wanted
and
want
to
get
raw
or
high
performance.
And
but
this
is,
you
know,
this
is
one
specific
configuration
of
cuber.
D
D
Yeah
I
think
that's
I
would
underline
underline
that
you
know
it's
a
use.
You
know
it's
an
it's
an
autumn.
It's
for
our
use
case.
It's
an
optimized
configuration
and
deployment
of
Qbert
with
additional
components.
Sorry
I
was
speaking
a
lot
and
fast
to
you.
I
think
was
hard
to
keep
up
with
that.
I
think.
A
All
right
is
there
anything
else
we
need
to
bring
up
for
the
agenda
and
notes
today.
J
Yeah,
just
two
quick
questions
just
to
conclude
for
myself.
If
we
know
that
we
want
to
enable
those
features
regardless
all
the
time
and
it's
the
best
thing
to
do,
then
why
just
not
to
enable
them
and
then
try
if
we
know
that
this
is
Windows,
we're
just
not
to
enable
it
and
then,
if
we
couldn't
just
retry
without
all
the
scheduling
overhead.
A
K
My
take
is
that
I,
don't
have
any
opinion
I'm
fine
with
you,
cubed
cubed
core
guys
want
to
do
I'm,
fine,
another.
Well.
The
approach
I
I,
followed
in
my
past
PR,
was
that
since
all
the
individual
low-level
flags
were
exposed,
I
just
followed
suit.
In
my
opinion,
according
to
the
Oviatt
experience,
what's
best
is
that
okay,
this
is
a
Windows
VM.
You
want
it
fast,
so
we
enable
all
that
early
features,
maybe
using
the
labels
and
not
trying
to
do
so.
This
is
my
take,
but
here
I'm
here
to
help.
J
A
Thanks:
okay,
moving
on
I
guess
to
the
open
floor
section
unless
there's
any
last
call
for
agenda
notes
all
right,
open
floor,
just
a
reminder
that
we're
once
again
changing
daylight
savings.
This
time
Europe
is
moving
to
daylight
savings.
So
the
apparent
time
of
this
meeting
in
the
United
States
time
zones
will
change
back
to
10:00
a.m.
so
it'll
be
4
p.m.
c
est
and
10
a.m.
Eastern,
Daylight,
Time
I,
believe
in
the
last
24
hours
the
European
Parliament
voted
to
abolish
daylight
savings,
starting
in
2020.
That's
good
news!
If
we
can
get
the
u.s.
A
C
A
I
Way:
yeah
great
so
yeah
we
have
some
left
from
the
community
that
country
works
hard
to
compile
to
run
Cooper
from
PowerPC
architecture,
and
you
can
check
the
issue
and
you
can
change
the
PR
it
just
some
earlier
stage.
Pr
is
some
hardcore
:
Bibles,
because
if
I
create
to
change
a
lot
of
things,
not
a
lot
but
something's
under
our
code
that
we
also
have
call
it
like
emulated,
type
type,
machines
and
architecture
that
currently
we
don't
have
possibility
to
change
architecture
malware
guy.
I
D
I
He
currently
has
some
problems
with
which
capabilities
I
hope
it
mostly
connect
to
the
machine
type.
So
he
counter
trying
to
change
the
machine
type
under
cubed
config
and
just
avouch.
It
there's
no
problems
at
changing.
After
changing
the
cubed
config,
you
need
to
restart
all
relevant
components,
because
we
currently
don't
accept
runtime
change
of
the
configuration
also
some
problem
that
I
believe
we
need
to
seem
to
change
in
the
future.
So
yeah
like
I,
said
I
hope
today
or
tomorrow
we
will
have
some
real
results:
cool
thanks.