►
From YouTube: KubeVirt Community Meeting 2021-06-09
Description
Meeting Minutes: https://docs.google.com/document/d/1kyhpWlEPzZtQJSjJlAqhPcn3t0Mt_o0amhpuNPGs1Ls/edit#heading=h.h6h40w9712z
A
A
Okay,
here
we
are
I'll,
give
you
guys
a
minute
or
two
to
fill
in
any
agenda
or
open
floor
items
that
you
may
want
to
put
in
there.
A
While
we're
waiting
for
folks
to
fill
in
agenda
items,
is
there
anybody
new
with
us?
That
would
like
to
say
hello.
A
Looks
like
just
familiar
faces,
we
do
have
a
telephone
call
person.
B
Yeah
this
is
marcus,
thornton,
I'm
away
from
my
office,
so
I
just
dialed
in
via
phone.
A
A
Okay,
well,
let's
begin
daniel
heller
has
the
first
item.
So
how
about
you
start
daniel.
C
Yeah
hi
everyone
I
was
just
wondering
in
while
I
was
looking
at
the
ete
lanes
that
we
have
in
cupert.
C
There
are
a
couple
of
old
1.70
lanes,
and
one
of
these
is
a
storage
lane
which
tests
rook
seth,
and
I
was
asking
roma
moore
about
that,
and
he
suggested
that
we
might
just
streamline
the
six
starch
lanes
and
the
rook
safe
lanes,
and
I
was
wondering
whether
anyone
from
the
storage
team,
for
example
or
regarding
storage,
has
some
ideas,
whether
that
is
a
good
idea
or
rather
not
and
yeah.
I
just
wanted
to
get
some
input.
C
I
was
thinking
about
just
adding
or
enabling
the
rook
steph
within
the
six
storage
lane
and
and
testing
both
together,
or
that
at
least
was
was
was
what
oman
was
suggesting
that
we
should
do
probably
for
all
lanes
that
we
have
like
for
119,
120
and
121
just
enable
the
rook
staff
testing,
also
within
the
six
storage
lane,
and
have
everything
together.
A
So
I
I'd
like
to
add
on
to
to
the
sig
conversation.
Ryan
halsey's
been
running
the
performance
in
metrics
sig
and
he
conducts
a
a
meeting
on
thursday.
D
D
Yeah
I
mean,
I
guess
I'll
speak
for
we
we
would.
If
there
is
demand
for
it,
storage
people
will
be
there
to
listen
and
contribute
so
yeah.
I
mean,
I
think,
that
we
have
been
getting
some
more
contributions
from
community,
so
it
may
make
sense.
I'm
just
you
know.
I
think
at
least
when
I'm
speaking
for
storage
we're
definitely
not
at
the
level
of
q,
bert
but
yeah
I
mean
it
may
be
monthly
or
something.
A
Yeah,
michael
just
feel
free
to
reach
out
and
I'll
get
you
set
up.
If,
if
you
want
to
do
a
monthly
meeting
specifically
for
storage,
I
got
ryan
set
up
and
he's
been
running
with
it,
and
we
see
his
meeting
notes
being
posted
out
to
out
to
the
mailing
list
so
they're
moving
on
nicely.
A
Okay,
okay,
thank
you
and
ashley
has
the
next
topic
regarding
other
levels.
F
Okay,
yeah,
so
shirley
had
reached
out
to
me
and
was
asking
about
the
proper
alert
levels
for
our
different
prometheus
alerts
and
specifically
tagged
me
in
one
asking.
If
it
should
be
critical
versus
warning.
I
was
just
wondering
if
we
have
any
documentation
about
what
level
things
should
be
or
how
we
decide
on.
G
G
Understanding
of
this
sorry,
there's
some
background
noise.
I
think
that's
something
we
need
to
shore
up,
so
we
probably
need
to
have
a
better
understanding
of
how
this
is
used
in
practice.
So
what
warning
means
to
an
operator
versus
critical
and-
and
things
like
that-
is
that
what
this,
so
this
pr
that
you
just
created
is
that
creating
different?
F
The
one
link
shirley
had
created-
and
she
was
asking
me
specifically
about
an
alert
of
if
the
vert
handler
damon
set
doesn't
roll
out.
Does
that
count
as
critical
and
part
of
me
was
like
well,
it's
critical
for
hubert
infrastructure,
but
their
vm
workloads
could
still
be
working.
So
is
it
actually
critical,
and
how
do
we
want
to
define
this.
G
Yeah,
that
makes
sense.
My
suggestion
is
to
run
with
that
and
maybe
create
a
pr
for
our
monitoring
or
our
metrics
mark
down
in
the
docs
directory
with
what
you
think.
The
recommendation
should
be
we'll
just
kind
of
start
the
discussion
from
there,
because
I
don't
think
nobody
spoke
up,
so
I
don't
think
we
have
a
good
handle
on
it.
A
Okay,
nothing
else,
so
I'm
gonna
go
ahead
and
talk
about
an
event.
That's
coming
up
here
on
tuesday,
tuesday,
it's
early
in
the
morning
I
hit
every
number
except
the
right
number
june.
15Th
1500
utc
david
vassel
is
going
to
do
a
live
stream
with
with
siam
who
is
a
cncf
partner.
A
He
runs
a
a
monthly
live
stream
talking
about
various
kubernetes
topics
and
this
month
is
going
to
be
kubert.
A
So
this
is
been
quite
some
time
in
the
in
the
making
and
we're
finally
gonna
go
forward
with
it.
Hey
david.
Do
you
have
a
quick
link
to
his
youtube
stream.
G
There
we
go
okay,
so
that
should
be
his
youtube
channel
yeah.
I
have
no
idea
what
to
expect
from
this.
It's
going
to
be
interesting,
maybe
maybe
it'll
be
good.
I
hope
it's
good.
A
So
it's
he's
not
just
a
a
random
person
that
just
popped
up
and
said:
hey
guys
how
about
doing
a
live
stream,
so
they
they
approached
us
through
cncf
channels
and
then
through
red
hat.
A
A
All
right,
so
we
also
have
a
kvm
forum
and
kubecon
and
a.
A
And
all
things
open,
kvm
forum
and
kubecon
and
a
our
papers
have
been
submitted,
we're
in
a
holding
pattern
waiting
on
the
council's
decision
same
with
all
things
open.
A
All
things
open.
We
are
developing
the
demo
currently.
So
if
you
have
a
raspberry
pi
and
some
good
internet
bandwidth-
and
you
want
to
participate
in
this
demo-
please
let
me
know
we're
all
the
content
for
this
demo
is
also
going
to
be
built
into
a
blog
post.
So,
even
if
we
don't
get
accepted
into
the
into
the
the
conference,
we're
still
going
to
produce
some
content
to
be
published.
A
A
Dang,
still
no
volunteers,
all
right
open
floor
is
empty.
Does
anybody
else
have
anything
you
want
to
talk.
A
A
I
had
an
idea
that
maybe
we
should
go
through
the
some
threads
on
the
mailing
list.
A
What
do
you
guys
think
about
adding
a
section
for
that.
I
I'm
just
saying
that
pr
to
discuss-
but
I
mean
this
topic-
is
pretty
general.
Maybe
we
can
talk
about
this
first,
okay,.
A
J
Okay,
I
hope
it's
good.
Now.
You've
got
a
happy
medium
yeah,
so
I
was
speaking
about
this
pr
in
the
last
community
meeting
and
so
basically
what
it
does.
J
It
shows
that
pmi's,
with
post
model
cpu
are
being
migrated
only
to
nodes
that
support
this
cpu
model,
and
we
thought
about
how
we
can
test
this
and
then
me
and
vladika
thought
that
maybe
and
we
can
make
a
high
performance
lane
for
like
high
performance
features
that
need
very
heterogenous
cluster,
with
a
lot
of
different
nodes
with
different
hardware
and
stuff
like
that
for
specific
features,
and
I
thought
to
ask
you
if
you
find
this
valuable.
K
J
I
didn't
think
about
the
specific
or
implementations
yet,
but
I
did
think
about
that.
We'll
be
free
to
add
new
nodes
from
a
new
hardware
or
stuff
like
that
and
make
a
very
messy
cluster
for
for
certain
features.
K
E
A
K
C
One
other
thing,
chris,
I'm
sorry.
I
forgot
why
you
mentioned
the
mailing
list.
I
forgot
to
bring
up
and
to
ask
there
was
someone
asked
from
the
mailing
list
about
someone
having
problems
with
python
client
and,
to
be
honest,
I
don't
know
who
is
today
maintaining
this
today
I
does
anyone
else
know.
K
Client
for
cooper,
do
it
yes,
yeah.
I
don't
think
today
we
had
months
ago
already
some
users
which
were
using
it
for
ansible
sort
of
thing.
But
since
then,
I'm
not
sure
at
least
no
one
from
us
maintainers
has
is
having
a
look
into
it,
so
they
just
get
for
as
long
as
the
the
tool
which
generates
it
compiles.
A
A
A
All
right,
I
think
we
should
just
go
through
here
and
kind
of
do
it
like
we
did
with
the
we
do
it
bugs.
So,
if
there's
any
email
here
that
hasn't
been
answered,
you
can
take
a
look
at
it.
A
A
He
wanted
me
to
add
the
google
account
to
the
weekly
meetings
and
for
some
reason
this
one
got
sent
out.
I
know
I
said
I
hit
the
do
not
send
invite
button.
It
did
anyways.
A
So
the
the
google
groups
account
got
about
50
declined
messages
from
all
from
everyone.
So
thanks
for
that.
A
We
talked
about
daniel
hiller's
e
to
e
lanes.
A
Oh,
here's
a
good
one
if,
if
anyone
knows
of
any
new
major
users
of
cooper,
let's
get
our
user
list
updated.
A
This
will
apply
to
yeah.
The
users
list
is
in
the
adopters
file
and
we'll
be
using
that
for
the
petition
to
graduate
into
incubating
status.
K
Maybe
it's
an
issue
with
a
new
kubernetes
client
go
version.
I
don't
know.
A
H
Like
I
don't
know
how,
how
does
the
rape
replace
buildup
they're
working
with,
but
I've
seen
similar
errors
when
rebasing
my
kubernetes
master
and
and
not
running,
make
clean
before
running,
build
or
something
so.
D
So
one
thing
that
I
just
want
to
throw
out
there:
I've
seen
other
community
people
talk
about
difficulties,
importing
the
key,
vert
client
go
and
a
lot
of
the
times
they
just
want
to
like.
They
just
want
the
api
definitions
and
none
of
the
really
clients.
So
I
wonder
if
it
would
make
sense
to
do
like
kubernetes,
where
we
have
a
keyboard
api
and
a
convert
client
go.
So
if
you
just
want
the
api,
so
you
can
use
you
know
the
controller
runtime
client
easily.
K
B
Be
a
great
idea,
that's
something
we've
we've
run
into
recently
as
well,
and
you
can
you
can
import
just
the
api,
but
it
also
has
most
of
the
pain.
As
far
as
the
modules
you
know,
figuring
out
all
the
compatible
modules,
there's
a
there's,
an
environment.
B
Added
that
lets
you
pick
like
whether
you
want
to
use
v1
or
v1
alpha
3.,
but
some
of
the
some
of
my
team
members
used
that
and
they
just
kind
of
found
it
a
bit
cumbersome.
I
guess
so
just
be
really
welcome.
K
H
As
slack
was
mentioned
there,
I
I
have
a
question
I
I
I
wanted
to
ask
a
few
times
like
how
do
we
see
the
slack
channel
as
a
responsibility
for
support,
because
I
think
I
also
ask
questions.
Sometimes
you
get
a
response.
I
see
a
lot
of
messages.
People
posting
prs
there,
but
I'm
never
sure
if
everybody
feels
responsible
being
on
the
channel
and
actively
responding
to
it.
Like
I
I
know
I
don't
and
I
should
be
on
there
more
and
I'll.
Try
that,
but.
A
That's
a
really
good
question
and
I
don't
have
an
answer
I'm
happy
to
to
take
that
up
to
with
with
josh
burkus
and
get
some
definition
on.
What's
expected
out
of
slack.
H
Like
right
now,
if
I
have
a
problem,
I
know
the
right
people
to
ask
or
at
least
know
people
I
can
bond
and
if
they
don't
know
they
they'll
tell
me
and
that's
my
shortcut
and
I
I
suppose
more
people
have
that,
but
it's
not
ideal
for
new
people.
Maybe
I
don't
know.
K
H
A
E
A
E
E
I
think
we
upgraded
the
cover
to
to
use,
go
116
right.
Yes,
so
my
id
started
to
do
crazy
stuff.
It
got.
C
N
E
A
K
Yeah
yeah
yeah
yeah,
sometimes
yeah.
So
what
basically
can
happen
is
that
like
tools
used
to
detect
the
minimum
version
and
they
compare
it
with
their
version,
what
they
have
and
then
they
think
everything's
fine,
but
then
it
doesn't
compile
and
it
may
not
even
even
understand
they
go
mod
file
thing
properly
and
stuff
like
this.
So
it
can
be
very
confusing,
like
pretty
much
like
edward
described.
A
A
Great
okay,
we're
at
7
36.
I
think
we
did
a
a
good,
a
good
shot
at
checking
the
mailing
list
to
make
sure
things
have
been
addressed.
A
K
K
K
Let's
see,
did
we
always
start
from
the
newest
to
the
oldest
one
or
yeah?
I
think
that's
the
way
it
works.
Okay,
okay,
so
we
have
a
new
one.
Why
don't?
I
do
word
ctl
restart
test
vm.
K
G
Correct,
I
think
it
can
even
be
simpler
than
that
it's
for
live
migration,
for
example,
there's
times
we
have
to
have
two
pods
at
the
same
time,
so
we
have
to
use
a
generated
name,
yeah.
H
H
G
K
G
H
G
K
G
P
I
think
this
is
actually
a
very
good
point.
That
is
why
we
want
to
have
a
different
path
like,
for
example,
in
my
use
case
right.
If
you
use
path
ip,
then,
if
you
do
a
reboot,
the
ip1
can
change
right,
but
if
you
do
a
restart,
it's
going
to
move
to
a
different
path
in
ip
change,
I
mean
in
the
case
of
using
power
ip.
P
So
actually,
I
think
this.
I
personally
think
there
would
be
that
nice
idea
if
we
could
restart
within
the
same
path,
but
have
a
different
setting,
though,
which
I
don't
know
if
it's
also
maybe
not
possible,
for
example,
when
I
use
restart
right.
What
I
do
is
if
I
need
to
update
the
cpu
or
the
memory
and
then
to
make
that
to
be
taking
effect.
I
need
to
restart
in
that
case,
is
it
even
possible
to
do
a
we
think
like
we
keep
it
within
the
same
path.
K
I
think
right
now,
it's
not
possible
from
the
community.
We
need
his
perspective
to
extend
the
resources
during
the
run
time
for
report,
but
there
is
the
cap
accepted
for
a
pretty
long
time.
I
think
some
people
from
ibm
provided
the
cap,
which
allows
shading
the
changing
requests
and
limit
level
during
the
runtime
of
a
bot.
P
Okay,
then
thank
you
for
that
answer,
because
I
was
also
thinking
this,
but
I
I
say
that
if
I
change
the
resources
it
looks
like,
I
have
to
use
a
different
path,
and
then
I
have
you,
you
start
to
to
get
that
applied,
that
change
applied.
G
Are
you
so
you
mentioned
the
pod
ip
changing?
This
is
one
path
that
we
could.
You
know
reboot
within
the
same
pod,
to
keep
the
ip
for
a
little
bit
longer.
You
still
impacted.
I
mean,
of
course,
if
like
a
node
failure
happened
or
the
pod
just
or
the
gas
crashed,
or
something
like
that,
you
would
get
a
new
ip.
G
So
are
you
concerned
about
still
needing
to
handle
that
scenario,
or
is
it
okay?
In
some
situations,.
P
I
was
just
thinking
that,
if,
in
general
right,
if
from
our
own
user
experience
like
general
vm
right
when
you
restart
it,
you
expect
things
keep
the
same
ip
and
things
like
that
right.
But
the
thing
is
this:
that's
something
expect
this
this
guy
right,
who
found
this
request,
oh
surprising,
that
I
restarted.
P
So
I
think
he
made
some
sense
hey.
His
comment
makes
sense
if
you
think
about
that
behavior,
so
it
would
be
nice.
If
really,
we
can
really
keep
everything
to
be
look
the
same
with
him.
I
put
we
restart,
but
on
the
other
hand,
I
can
see
that
if
we
allow
you
to
modify
resources
and
restart
to
apply
this
new
configuration,
then
maybe
that
make
it
impossible
to
give
the
same
part.
That's
just
my
comments
here.
K
I
think
keeping
the
product
is
a
good
example.
We
have,
there
should
definitely
come
at
some
point
with
ctl,
something
like
we'll
see
to
restart
and
read
ctl
with
a
a
queue
level
reset
and
restart
with
a
acpi
reboot
from
outside,
and
there
the
qmo
process
would
still
be
alive.
The
whole
time
just
the
vm
inside
would
be
restarted
or
rebooted
from
outside,
with
more
or
less
force,
and
this
makes
definitely
sense
to
have.
K
O
That
one's
for
me,
yeah
just
for
for
ci.
K
K
K
I
guess
it
makes
sense
to
ask
for
a
few
use
cases
I
could
think
of
some,
but
it
would
be
interesting
to
see
that
what
we
already
have
from
the
from
the
pot
level
and
from
the
actual
consumption
level-
I
I
could
think
about
exposing
this
metric
for
getting
an
idea
on
how
much
is
used
to
the
system.
But
I
have
the
suspicion
that
maybe
there
is
already
enough
provided
through
kubernetes
metrics.
You
know
fukuniti's
metrics
I'll
ask
for
more
details.
Does
that
sound
fine.
H
K
K
K
K
K
K
Ayah,
this
is
probably
an
issue
with
going
from
from
from
the
non-transitional
from
the
going
from
the
transitional
to
the
non-transitional
weird
diode
driver.
K
K
L
K
L
L
L
K
G
Yeah
I
commented
on
this
on
the
per
user
issue.
K
Okay,
so
this
is,
is
this
not
for
the
this
is,
is
this
for
the
over
forward
launcher,
but
not
for
the
the
run
not
for
the
control
plane?
Okay,.
K
G
Yeah
there's
some
confusion
in
the
in
the
discussion
about
that
as
well.
So
I
think
it
makes
total
sense
to
have
a
global
default,
because
if
people
have
multiple
run
times,
then
at
least
one
of
them
would
certainly
make
sense
for
cube
verb,
maybe
not
all
of
them
yeah
that
makes
sense
the
per
user
one.
G
Maybe
if
somebody
finds
that
useful,
the
thing
is
during
that
discussion
yeah
that
discussion,
I
think
it
started
being
apparent
that
the
per
user
run
class
or
whatever
it's
called
possibly
was
being
desirable
because
of
a
side
effect
with
the
run
class
or
whatever
it's
called
runtime
class.
You
can
set
scheduling
information
on
it.
G
So
if
a
pod
has
the
runtime
class
name
set,
then
it
will
inherit
the
runtime
classes,
scheduling,
affinity
and
node
selectors
and
everything,
so
that
was
being
used
to
kind
of
do
some
odd
stuff
here,
which
I
don't
know
if
it's
desirable
or
not.
G
B
Okay,
okay,
yeah
yeah.
I
was
a
little
confused
by
that
as
well.
It
sounded
like
it
was
to
be
used
as
kind
of
a
shortcut
so
to
speak
so
that
you
didn't
have
to
put
you
know
your
node
affinity
on
your
on
your
vmi
itself
or,
and
then
there
was
the
whole
thing
about
device.
Plugins
and
well.
Is
it
going
to
fix
the
right
because
they
they
wanted
to
use
runtime
cloud
to
fix
the
hardware?
B
I
guess
I
think
so
for
for
the
use
case
I
had
in
mind-
and
I
think
it
was
mentioned
on
the
mailing
list
was
you
know,
I
think
the
big
one
that
people
are
going
to
have
is
they
may
be
running
kata
in
their
clusters
and
they
don't
want
to
run
qvert
on
kata.
So
you
know
having
that
kind
of
a
global.
K
That
sounds
crazy
yeah.
I
think,
maybe
that
right
away,
maybe
maybe
yeah
making
it
a
list
when
you
configure
it
so
that
you
can
have
or
an
exclusion
list
or
something
maybe
both,
but
I
could
also
I
mean
I
don't
know
how
realistic
it
is,
but
in
theory
we
could
have
cryo
on
the
pile
on
part
of
the
notes
and
stock
on
another
part,
and
it
would
work
for
both.
I
don't
know
if
this
is
really
irrelevant.
G
G
K
G
Would
start
with
that,
and
then
we
have
a
use
case.
That
really
sounds
good
for
why
we
would
need
a
per
vmi
class
runtime
class
name.
Then
we
can
consider
it.
That
is
like
an
iteration
on
the
global
default.
K
K
K
K
N
Q
K
K
K
N
A
Yeah,
thank
you,
roman,
for
handling
that
it's
always
good
to
have
an
expert
drive.
The
bug
scrub.
A
So
that
concludes
our
our
meeting
for
the
week
and
I'll.
Let
you
guys
go,
and
I
will
take
two
minutes
from
you
this
time,
instead
of
giving
you
30.
well
we'll
finish
up
here
and
enjoy
your
week,
see
you
next
time.