►
From YouTube: KubeVirt Community Meeting 2021-10-20
Description
Meeting Notes: https://docs.google.com/document/d/1kyhpWlEPzZtQJSjJlAqhPcn3t0Mt_o0amhpuNPGs1Ls/edit#heading=h.2dzf25e4q2bm
A
Hello,
everyone.
This
is
the
weekly
kubert
meeting,
I'm
your
host
chris
caligari.
It
is
wednesday
october
20th
2021
and
looks
like
we
have
a
few
folks
that
I
don't
recognize
on
the
participant
list.
If
you
want
to
take
a
minute
and
introduce
yourself
you're
more
than
welcome
to.
B
May
I
start
hi
everybody.
This
is
michael.
This
is
the
first
time
I
attend
this
meeting.
I
am
working
for
for
arm.
D
A
F
Hi
everyone,
my
name,
is
greatness.
I
am
currently
an
outreach
intern.
This
is
my
first
time
attending
a
meeting
as
well.
Actually
it's
my
second
time,
the
first
time
I
was
actually
in
it.
I
think
I
met
you.
Chris.
G
Okay,
hi
guys,
I'm
federico,
I
joined
the
red
dot
and
in
the
previous
week
it's
my
second
time.
I
joined
this
team
and
this
is
a
meeting.
Sorry
and
yes,
I'm
currently
working
on
cube
beard
and
hello.
Everybody.
A
A
We
definitely
need
a
lot
of
help
from
all
across
the
board
and
just
in
case
you
don't
know,
the
meeting
notes
are
post.
I
just
posted
the
chat.
So
if
you
can
fill
out
your
attendance,
we
do
track
attendance
week
to
week
and
we
feel
free
to
add
anything.
You
would
like
to
talk
about
to
the
agenda
or
open
floor.
A
And
michael,
if
you're
there's
nothing
on
the
agenda,
so
if
you
want
to
get
started
with
talking
about
level,
one
hypervisors
you're
more
than
welcome!
Oh
thank
you.
B
A
B
E
B
As
I
mentioned,
that
we
are
working
for
and
are
interested.
B
And
the
the
proposal
here
is
to
extend
another
kind
of
hypervisor
in
kubernetes.
It's
a
type
one
hypervisor.
So
the
first
ask
maybe
what
is
type
hyperion,
the
type
2?
What's
the
difference
to
be
simple
type,
1
hypervisor
runs
directly
on
the
host
hard
wire
and
the
chart.
The
on
the
left
is
the
type
1
hypervisor
and
the
type
2
hypervisor
runs
on
an
operating
system.
B
B
B
Of
everything
for
a
coupe
word,
the
exchange.
It
would
be
like
that
on
the
left
side
of
the
chart,
it
is
the
current
deployment
instances
and
a
vm
is
contained
in
the
port
where
the
word
launcher
is
running,
but
for
a
type
1
hybridizer
I
mean,
then
it
will
be
different
because
of
the
difference
in
the
architecture.
H
Hi
everyone
here,
first
thanks
mikey
for
preparing
this,
this
preparation,
what
I
wonder,
what
are
the
implications
here
regarding
to
see
groups
and
name
spaces,
for
instance,
with
that,
could
you
can
you
elaborate
on
the
two.
B
B
B
B
So
in
that
case,
the
group
and
the
namespace
only
works
for
the
zen
drivers.
Part,
oh
okay,.
H
B
Okay,
so
in
that,
no
no,
no
typically
the
cooper
equivalent
should
be
running
on
this
dome
zero
vm.
It
is
the
so-called
the
host
it
only
so
cooperator
can
only
see
the
results
on
dom
zero,
not
the
one
for
vm.
I
Okay,
so
it's
a
separate
environment,
entirely.
It's
separated
by
a
hypervisor
like
the
host.
What
I'm
getting
is
the
actual
host
the
cubel
is
running
on
and
the
pods
are
running
on
that's
running
within
a
virtual
machine
and.
E
B
B
So
to
achieve
the
support,
the
supporting
there
are
some
challenges
in.
We
have
mattered
in
the
prototype
work
and
specifically
there.
There
should
be
some
changes
in
distinguishing
different
hypervisor
type
in
build
time
and
resolving
different
software
dependency
for
hypervisors
and
handling
the
health
system,
resource
isolation,
and
then
then
virtual
matching
is
also
something
we
need
to
handle
because
the
host,
the
bare
metal
zen
version
could
be
different
from
that
in
the
leibovert
and
also
vm
status,
monitoring
and
vm
networking.
B
So
we
made
a
proof
of
concept
to
try
the
possibility
and
with
that
prototype
now
we
can
schedule
a
zen
vm
and
connect
his
cluster
and
manage
the
vm
lifecycle,
and
we
can
also
log
in
as
nvm
by
vert
control,
client
and
that
was
tested
on
both
x86
and
r64
and
after
all,
a
proof
of
concept
is
not
complete.
There
are
still
some
areas
we
haven't
been
able
to
cover,
including
the
vm
network,
setting.
J
H
H
Pretty
common
to
use
send,
so
are
you
looking
for
specific
features
which
zen
provides
you
which
kvm
doesn't
have
like
regarding
to
real
term
time
or
something
or
so?
Are
there
some
things
which
you
need
for
your
work
from
zen.
B
Okay,
so,
for
example,
on
a
vehicle
vehicle
system
and
the
robustness
and
stability
could
be
the
top
priority
and,
as
you
know,
that
as
a
type
2
hypervisor
kvm
for
kvm,
the
vcpus
of
a
vm
are,
if
in
fact
like,
contained
or
carried
by
by
the
shrines
of
a
virtual
machine,
monitor
application,
and
that
is
in
application
level,
and
there
is
risk
that
that
application
class
or
something
like
that,
but
zane
is
different
as
we
we
saw
in
the
architecture.
B
The
here
vm
is
running
on
directly
on
the
then
it's
in
hybrid,
so
even
there's
something
wrong
in
in
the
host
in
there.
It
doesn't
impact
the
vm
itself,
so
it
could
be
more
robust
and
stable.
I
So
on
the
chat
I
saw
somebody
ask
alicia,
I
think
yeah,
how
are
you
passing
storage
and
my
question
would
be
also:
how
are
you
tying
this
into
the
network
as
well
at
the
pod
network?
This
then
virtual
machine.
B
So
I
can
share
a
little
bit
our
detailed
analysis
parties
about
the
network
so
for
our
best
thinking
our
best
practice
in
students.
E
B
Is
there
a
chemical,
precise
running
in
regulator,
or
are
you
using
I
about
this
question?
I'm
I'm
sure
that
on
arm
there
will
not
be.
We
didn't
use
a
cumule
on
arm,
but
for
x
86,
I'm
not
sure,
but
I
can
answer
that
we
don't
have
a
cumule.
B
So
we
also
noticed
this
part
so
in
in
coopervert,
as
I
know
that
we
are
watching
a
communal,
precise
to
monitor
if
a
gas
station
is
live
or
not
so
ways
I
forgot
where,
where
is
it
so
for
then
we
need
to.
We
need
to
use
other
another
way
to
to
find
the
status
of
the
vm,
so
could
probably
buy
some
some
ports
with
the
hypervisor
to
check
the
status
of
a
vm.
I
One
of
the
I
would
say
core
tenets
of
the
key
vert
design
is
that
the
pod
is
the
virtual
machine
and
the
virtual
machine
is
the
pod,
meaning
that
if
the
pod
goes
down
that
we
are
ensured
that
the
virtual
machine
has
terminated,
and
when
I
look
at
this
diagram,
it
seems
like
it's
possible
for
the
virtual
machine
to
technically
outlive
the
pod.
Is
that
is
that
possible?
I'm
not
maybe
it's
not
possible
like
what's
the
separation
here?
Is
it
possible
for
the
zen
drivers,
whatever
isn't
like?
I
B
Okay,
I
see
your,
I
see
your.
Let
me
see.
I
So,
from
a
design
point
of
view,
the
the
reason
that
makes
us-
or
at
least
I'll
speak
for
myself
uneasy-
is
that
we
don't
have
full
visibility
into
what's
actually
occurring.
So
we
don't
know
if
we've
for
sure
terminated
a
virtual
machine.
The
cluster
thinks
that
it's
terminated
because
the
pod
is
gone,
but
the
reality
may
or
may
not
match.
B
To
check
the
status
of
the
vm
if
the
port
has
been
cured
for
some
some
reason.
I
Yeah
yeah,
I
would
just
think
through
that
and
see
if
that's
feasible.
Oh.
H
H
I
wonder
how
complete
the
integration
with
kubernetes
will
be
at
the
end
like
right
now,
really,
as
david
said,
the
whole
vm
is
in
the
part
there
is
nothing
outside
of,
and
this
means
that
all
monitoring
and
reporting
structures
from
cuba,
from
the
cubelet
work
as
expected
and
for
the
scheduler,
we
request
resources
from
the
scheduler
for
our
part
and
they're
really
consumed
one-to-one,
and
it
matches
really
what
the
vms
uses
using
on
every
aspect.
H
B
In
our
current
prototype,
we
haven't
touched
the
this
problem
really.
E
H
That
that
what
david
said
what
I
said,
two
of
the
core
drivers
for
the
current
design
from
cubert
how
it
is
today.
B
Yes,
we,
our
previous
type
of
work,
focus
on
the
feasibility
to
manage
mainly
managing
the
life
cycle
of
vm.
So,
yes,
we
haven't
done
deeply
about
this
resource
management.
B
B
H
H
Was
that
a
question
to
me:
yeah
yeah
yeah,
it's
just
late,
just
by
coincidence,
lately
some
work
on
real-time
kvm
started,
which
may
be
interesting
for
automotive
to
it's,
just
something
I
wanted
to
highlight.
Maybe
that's
interesting
for
you.
B
Uh-Huh,
okay,
I
will
check
that,
but
as
as
our
as
heavy
interest
in
zen,
so
some
of
our
software
stacks
are
have
been
fixed
to
then.
So
how
then
is
a
a
great
requirement
for
a
work?
So,
of
course,
we
will
also
be
interested
in.
I
I
I'm
I'm
curious
if
kevert
is
gonna
be
used.
Is
this
more
of
a
qe
effort,
so
testing
of
some
sort
of
automotive
virtual
machine,
or
is
it
actually
going
to
be
something
that's
production,
so
some
of
the
things
that
we
pointed
out
might
not
matter
if
this
is
just
trying
to
start
up
a
bunch
of
zen
virtual
machines,
because
there's
some
sort
of
validation
that
needs
to
occur
and
the
production
environment
is
totally
different.
B
I
I
think
I
was
asking
more
of
what
do
you
have
any
more
information
you
can
share
about
your
use
case
for
zoom
and
whether
this
is
testing
or
is?
Are
these
production
like
workloads
or
is
this
some
sort
of
workload
that
just
needs
to
go
through
his
validation
process?
That
then
gets
handed
off
to
run.
E
B
Okay,
in
fact,
we
are
the
open
source
team
of
and
we
didn't
create
a
concrete
project
product.
So
the
concrete
use
case
is
not
a
young
side,
but
we
only
work
on
the
upstream
to
and
provide
possibilities
for.
H
A
A
H
There's
one
more
thing:
I've
been
wondering-
and
this
is
like
there
are
lately.
Quite
some
projects
are
coming
up
which
are
less
about
running
kubernetes
as
kubernetes
somewhere.
H
I'm
just
posting
a
link
here,
for
instance,
to
kcp,
because
just
some
thoughts
which
are
coming
up
based
on
stuff
I've
read
lately,
so
it's
getting
more
and
more
interesting
to
to
to,
instead
of
going
directly
and
trying
to
modify
kubernetes
and
its
add-ons
to
meet
certain,
let
me
call
it
more
extreme
needs,
instead
of
that,
it's
just
more
like
of
going
back
to
just
using
the
kubernetes
api
principles
and
creating
your
own
implementations
behind.
Maybe
this
is
also
some
use
case
for
for
this.
H
A
H
So
for
me
so
again,
roman
for
me,
it
would
be
important
to
better
understand
how
something
like
this
would
fit
into
kubernetes,
as
it
is
right
now
like,
as
we
mentioned,
the
points
where
the,
where
the,
where
the
dom
is
actually
not
really
in
kubernetes,
which
has
influenced
potential
influence
on
scheduling
c
groups
if
the
vm
goes
down.
I
think
that
are
really
the
points
for
us
in
kubert
to
understand
to
see
if
this
would
be
feasible
at
all
yeah.
B
H
B
Add
up
more
about
this,
and
maybe
later
I
can
share
a
more
detailed
design,
including
the
topics
we
discussed
today.
I
Yeah
more
information
would
be
good,
I'm
generally
just
I'm
apprehensive
of
how
this
could
be
taking
taken
in
exactly
the
way
this
document
or
this
diagram
is.
I
I
think
that
most
likely,
if
we
find
a
path
forward,
it's
going
to
be
a
way
of
extending
cubert
and
similar
to
what
roman
pointed
out
with
kcp
like
some
sort
of
way
of
taking
part
of
the
kuvert
api
and
control
plane
and
then
allowing
like,
for
example,
the
node
portion
to
be
pluggable
or
something
for
for
people
who
want
to
use
something
like
zen,
it's
difficult
for
me
to
see
clearly
right
now,
yeah.
I
I'll
just
say
my
my
thoughts
looking
at
all.
This
is
that
it's
gonna
be
difficult.
It's
not
a
perfect
fit,
but
yeah,
especially.
H
E
A
E
B
So
sorry,
I
didn't
get
your
question.
Can
you
repeat
again.
E
Yeah,
so
I'm
just
asking
how
it
works.
We
send,
for
example,
for
sharing
storage
between
linux
and
xnvm,
for
example,.
B
Yeah,
yes,
and
in
storage,
we,
let
me
recall
what
we
have
very
fine.
A
Michael,
I
can
comment
on
the
storage.
I
have
a
colleague
at
nasa
who
does
data
collection
on
weather
sensors
across
the
planet.
A
His
his
method
of
dropping
storage
to
virtual
machines
is
probably
very
different
than
what
we
are
all
used
to
for
one
thing:
they
run
a
100,
gig
ethernet
system,
and
so
they
have
virtually
unlimited
network
bandwidth
and
they
want
to
use
it
and
then
the
the
protocol
they
use
is
nfs
4.1
and
the
target
is
also
an
ibm
parallel
file
system.
A
So
they
don't
care
about
doing
pvcs
and
pvs
within
the
kubernetes
system.
They
they
keep
that
as
a
separate
layer.
E
A
So
I
think,
as
I
as
I
see
michael
scrolling
through
his
confluence
page,
it
sure
looks
like
he
has
most.
If
not
all,
of
the
data
points
we
need
for
creating
a
design
proposal,
you're
scrolling
through
there
pretty
fast.
So
I'm
just
like
getting
snippets
of.
A
H
B
B
On
the
document,
the
questions.
H
B
B
J
H
Yeah,
for
instance,
so
the
the
scheduler
and
cubelet
know
how
much
resources
are
used
and
reserved.
So,
but
this
this
would
not
be
reflected
anymore,
because
either
you
would
request
with
the
request
on
the
vm
or
the
part
request
from
the
dom
zero,
where
kubelet
is
running
the
ram
and
from
the
dom?
U,
which
makes
it
double
or
you're,
not
really
reflecting
the
reality
at
all.
From
your
request,.
J
K
Also
with
allocating
resources
so
same
way
with
when
we
allocate
dedicated
cpus,
for
example,
so
I
wonder
how
this
dom
you
will
get
that
specific,
dedicated
cpus
that
you
got
from
the
cpu
manager.
I.
B
B
B
K
Oh,
I
was
sorry.
I
was
just
saying
that
normally,
when
we
start
the
vm.
E
K
We
integrate
this
with
the
with
the
pod,
and
the
pod
has
different
qs
levels.
So
when
the
qs
level
is,
for
example,
they're
guaranteed,
then
they're
going
the
resources
that
we
allocate
for
that
are
guaranteed
as
well,
and
the
one
who
allocates
these
resources
is
the
cpu
manager
and
the
node,
and
then
it
would
assign
specific
cpus
for
that
port
to
use,
and
that
and
these
cpus,
these
specific
cpus
are
the
ones
that
are
going
to
be
used
for
the
vm
yeah
in
pinning
mode.
K
So
I
mean
when
we're
taking
this
model
as
an
example,
I
wonder
how
the
dom
view
will
know
which
cpus
to
use
on
the
node.
B
Okay,
that's
that
is
the
zen
hypervisor
determined
does
that
have
to
determine
which
domain
use
which
cpu
and
it
also
demands.
If,
if
all
the
cpus
are
shared
among
all
the
domains
or
just
assign
someone
specifically
to
some
vms.
K
This
is
the
key-
I
guess
in
all
of
this,
because
the
cpu
manager
will
lose
control
here.
Cpu
manager
is
the
one
that
allocates
resources
in
the
cluster
on
the
node.
Oh.
B
H
A
So
I'm
showing
50
minutes
after
the
hour.
Michael,
if
you
don't
mind,
I'm
going
to
close
this
topic,
we
have
a
few
more
small
items
to
talk
about
before
the
close
of
the
meeting.
Yeah.
A
Thank
you.
This
has
been
an
excellent
discussion
and
the
proof
of
concept
looks
looks
really
exciting.
Thank
you
and
look
forward
to
seeing
more
of
this
of
this
topic
so
definitely
keep
an
eye
on
on
the
notes,
I'll
get
the
notes
sent
out
and
as
well
as
the
chat
log.
So,
okay.
B
A
B
A
Okay,
a
couple
bullet
points.
Roman,
if
you
don't
mind,
I'm
going
to
talk
about
yours
also,
we
just
had
a
couple
events.
We
had
a
kubecon
n
a
last
week
and
david
vossel
and
I
ran
through
our
presentation
and
we
had
I
lost
track
because
we
were
both
talking
and
demonstrating.
A
Last
time
I
looked,
there
was
45
attendees
and
we
ended
up
with
about
six
questions,
and
so
it
was
a
very
active
presentation
and
stu
and
I
did
all
things
open.
Yesterday,
coincidentally,
we
presented
cooper,
running
arm,
64
virtual
machines,
and
although
we
weren't
able
to
present
like
a
bitcoin
miner
running
on
arm
64
or
a
minecraft
server,
we
did
run
virtual
machines
and
we
had
about
30
people
attending,
and
we
also
had
about
five
questions.
A
Kubecon
n,
a
was
not
recorded.
All
things
open
was
recorded
and
we'll
soon
as
we
get
that
recording
we'll
get
that
posted
up
to
youtube.
A
Roman
says
that
there's
several
patch
level
releases
going
back
to
version
40..
We
fixed
a
bridge
bug
that
affected
windows,
virtual
machines,
failing
migrations,
going
back
to
0.40,
that
was
for
libert.
If
custom
liver,
qemu
is
used.
H
A
Yeah
we
we
I'm
so
glad
that
didn't
fail
during
demo
last
week,
because
we
we
demoed
live
migration.
A
Let's
see
zing
says:
there's
a
cadis
dependency
update
schedule.
H
D
Yeah,
it's
just
that
right
now
we
are
using
like
1.20
right,
which
was
like
10
months
ago
was
update
and
in
our
application
we
need
to
import
something.
This
version
is
to
lower
so
which
make
this
important
have
some
conflicts.
So
my
question
is:
if
it's
updated
online
is
needed
basis.
If
there's
no
fixed
guidance,
do
you
mind
if
we
update
the
dependency
and
then
send
the
sale
for
it.
H
I
think
two
points
here:
first,
yes,
please
just
go
ahead
and
update
it.
If
you,
if
you
need
it,
that's
definitely
fine
and
second,
I
think
we
will
soon
have
an
improvement
in
place
where
we
will
release.
In
addition
to
client
to
kubernetes,
we
will
release
a
pure
kubrick,
slash
api
golem
package,
which
just
contains
the
api
and
where
you
can
then
run
client
chen
to
generate
your
own
client
like
for
most
other
kubernetes
dependencies.
That
should
help
making
your
product
project
mostly
independent,
of
the
kubernetes
version
which
we're
using
in
keyboard.
D
Okay
sounds
good
yeah.
I
saw
that
effort,
but
yeah,
I'm
very
happy
to
say
it
and
well
with
for
it,
but
meanwhile
we
will
try
to
see
we
can
update
the
dependency
okay.
Great.
Thank
you.
Thank
you,
and
I
mean
the
next
one
is
also
iris.
This
is
about
the
macro
types
in
eye
plugin,
I'm
wondering
like
what's
the
stylus
of
that
and
repo,
and
if
anyone
is
maintaining
it,
because
that
one
has
even
older
dependency
like
1.19
and
also
we,
we
have
some
issues
when
using
this
magnitude
signal
parking.
D
So
we
are
wondering
if
we
can,
upstream
all
the
fixes
or
improvements
enhancement.
We
have
on
this
branch.
L
L
It
uses
like
very
old
versions
of
pretty
much
everything,
and
but
if
there's
like
an
active
user
base,
looking
for
it,
we
can
just
well
work
together
and
update
pretty
much
whatever
it
is
that
you're
looking
for
in
it.
L
Yes,
like
I
can
promise
like
very
timely
reviews
on
it,
but
I
can
promise
that
they
will
happen.
H
Also,
I
want
to
mention
that
most
of
us
are
on
slack
2
in
kubernetes.
H
A
So
we
have
a
a
lot
of
new
users
this
week
likely
from
our
our
presence
at
conferences.
I
will
make
sure
to
send
out
a
communication
reminder
reminding
everybody
of
our
community
page
on
the
website,
the
mailing
list
and
our
slack
channels.
A
Okay,
that
takes
us
to
the
58th
minute
of
the
hour.
We
are
out
of
time,
so
I
will.
I
will
let
us
skip
the
mailing
list,
review
and
bug
scrub,
which
roman
always
likes
skipping
bug,
scrubs.
A
Oh
michael,
it
is,
it
is
okay.
We,
week
to
week,
we
we
have
to
skip
bug
scrub
or
we
do
a
deep
bug,
scrub
it.
It
varies.
Your
topic
was
very
exciting.
We're
really
glad
to
have
you
here,
so
I
am
going
to
close
out
this
meeting
now
and
return
one
minute
to
you
and
thank
you,
everybody
for
joining
us
this
week.
This
was
a
really
great
meeting
very
lively
and
we'll
see
you
next
week.