►
From YouTube: SIG Cluster Lifecycle - kubeadm office hours 2021-03-31
A
A
A
So
I
see
some
new
music
participants.
Do
you
want
to
present
yourself
to
the
group.
B
C
Maybe
I
I
will
search
for
them
and
link
them
in
in
the
mythic
agenda,
so
you
can
have
you
can
look
at
past
meeting
for
quickly
catching
up
with
the
code
base
and
and
the
contribution
process.
B
C
A
Yeah
in
the
cuba
repository,
I
think
we
have
them
at
the
bottom
of
the
contributing
md
file.
Just
a
second.
A
This
we
have
these
two
instructional
videos.
One
of
them
is
by
me
the
other
one
is
by
fabrizio.
I
don't
think
they
are
that
much
out
of
date.
At
this
point
we
haven't
changed
that
much,
but
I
encourage
people
to
watch
them
both
before
proceeding
with
the
questions.
But
if
you
have
any
questions,
I
think
the
right
place
for
that
is
the
cube
adm
slack
channel
or
the
sequence
lifecycle
slack
channel.
A
A
A
A
There
are
discussions
to
move
things
to
four
months,
which
means
3d
releases
per
year,
but
I
don't
see,
I
don't
see
this
happening
for
the
next
release,
so
we
have
to
assume
that
it's
going
to
be
three
months,
which
may
or
may
not
be
enough
time
for
all
the
features
that
accumulated
for
v1
beta3.
A
As
discussed
before,
the
deprecated
v1
beta
1
should
be
removed.
I'm
proposing
that
we
do
that
in
122.
Costa
api
was
arguably
the
biggest
broker
for
that
due
to
some
complexity,
around
configuration
management
like
do
we
get
any
objections
from
people
on
this
call.
D
A
We
have
to
send
it,
we
have
to
set
it,
but
the
deprecation
happened
in
1
15.,
so
it
was
six
releases
ago.
We
are
hoping
that
nobody
is
using.
It.
A
C
Yes,
so
forecaster
api,
we
have,
we
are
fine.
Now
the
problem
is
fixed
and,
and
the
fix
is,
is
also
backporting
on
the
on
the
v03
branch,
and
so
we
we
can
remove
it.
I
agree
with
justine,
let's
send
another
reminder
in
in
the
mailing
list,
but
it's
just
an
inform
informative,
because
basically
the
application
process
is
started
is
ages
ago
and
it
will.
C
If
someone
come
up
with.
Oh,
I
don't
know
nothing
it,
it
really
diff.
So
we
are
printing
warnings,
it
is
documented.
We
did
everything
possible,
so
we
will
try
to
accommodate
every
ask,
but
really
we
did
everything
in
our
capability
to
make
it
clear
and
not
not
to
the
resurrected
and.
C
A
A
A
So
the
application
of
v1p2,
historically,
we
deprecated
an
api
with
the
release
of
a
new
api,
which
means
that
when
we
introduce
v1
alpha
3,
we
deprecated
v1
alpha
2
immediately.
The
question
is:
do
we
want
to
deprecate
d1b2
immediately
with
the
release
of
v1b23?
My
vote
is
maybe
not
at
the
same
release.
Maybe
we
can
delay
this
a
little
and
for
at
least
one
release.
C
Agreed
we
are
not
in
our
in
in
re
and
as
soon
as
we
have
to
implement
the
conversion
as
when
they
are
in
the
code
base,
then
it
will
be
not
a
problem
to
to
leave
them
there
for
once
cycle
more.
A
Yes,
the
way,
the
way
that
kubernetes
api
is
we
are
applying
changes
means
that
it's
like
a
it's
set
in
stone.
We
reviews
should
happen
on
the
the
v1
alpha,
sorry,
that
we
want
beta
3
cap
changes
and
the
pr's
and
that's
how
we
release
it.
We
it's
out
of
the
you,
know,
outs
out
of
the
door,
and
it's
if
you
want
to
make
changes,
it
has
to
happen
in
a
new
version,
so
but
yeah
we'll
definitely
gather
feedback.
A
A
A
I
think
what
what
I
was
thinking
today
is,
but
I
haven't
looked
if
this
is
appropriate.
A
We
should
just
update
this
section
here,
so
the
cap
has
a
section
if
you
want
beta
3,
it
already
has
some
things.
I
honestly
do
not
remember
when
we
added
this
fabric.
Do
you
recall,
with
the
optional
tag.
C
C
C
So
that's
mean
that
if
someone
like
cluster
pi
is
going
to
embed
our
types
in
some
form
into
our
crd,
he
should
not
have
a
problem
and
also
we
have
to
make
our
types
more
serialization
friendly.
So
when
we,
when
we
basically
output
our
types
to
to
yaml,
we
should
we
should
avoid
to
have
empty
field.
The
folds
stuff
like
that,
so
that
the
the
output
should
be
the
shorter
possible,
and
so
the
user
basically
has
to
only
set
what
is
it?
What
is
he
careful?
A
A
It
would
be
possible
with
another
feature
which
is
to
skip
the
add-on
phases,
and
if
you
skip
the
arrow
phases,
you
can
then
upgrade
the
same
cluster,
which
will
basically
the
upgrade
command
will
detect.
If
you
don't
have
the
add-ons
present
and
it
will
skip
it,
I
I
don't.
I
don't
agree
that
we
should
make
this
explicit
in
the
api.
So
for
british,
what
do
you
think
I
should
remove
this?
Maybe.
C
I
I
don't
know
I
I
think
that
having
so
providing
a
a
clean
way
to
avoid
to
install
that
dawns
is
more
and
more
a
frequent
use
case.
C
So,
in
my
opinion,
we
should,
if
possible,
it
should
go
in,
because
if
I
think
that
you
know
that
now
there
is
kubernetes
that
replace
both
proxy
and
and
kubernetes
sorry
and
car
dns.
There
is
also
andrea
that
that
are
from
wire.
That
allows
the
same
so
we
have
to
make
it
possible.
C
I
we,
then
we
should
discuss
in
the
cap
if
we
we
make
it
possible
through
through
skipping
phase
or
through
declaring
explicitly.
I
I
want
to
to
not
install,
I
don't
type
noon
or
something
like
that.
D
Agreed
especially
like
that,
like
from
from
what
I'm
seeing
all
the
companies
or
users
building
on
top
of
kubernetes,
are
rolling
out
their
own
add-ons
management
because
for
it
for
each
use
case,
you
end
up
having
some
specific
requirement
that
you
cannot
generalize
so
agreed
that
we
need
to
enable
keeping
any
add-ons
that
get
installed.
A
Yeah,
what
long
term
it
feels
that
cube
idea
will
be
completely
agnostic
we
have
to,
but
even
if
we
use
operators
or
some
plug-in
methods
for
add-ons,
we
still
the
the
default
cluster
will
still
have
to
deploy
them,
but
to
make
it
easy
for
the
basic
user
to
install
a
cluster
without
carrying
that.
What
is
q
proxy?
What
is
coordinates
and
so
on.
A
So
it
feels
like
the
more
advanced
use
case.
Is
the
companies
and
the
providers
that
need
to
skip
the
default
address,
but
I
I
still
I
don't
really
like
to
include
a
specific
field
to
skip
a
particular
addon.
I
think
that
we
should
just
use
the
phases
and,
like
it's
already
requested
in
this
particular
form,
to
our
skip
phases
via
the
api,
like
that
yeah.
C
Well,
I
I
think
that
let's
try
to
keep
so
we
can
discuss
on
implementation,
detail
and
separated
meeting.
So,
let's
try
to
to
define
what
are
the
themes
that
we
want
in
the
v1
alpha
3
api.
So
one
is,
let
me
say,
crd
friendless.
C
The
second
one
is
skippable
addons
and
skippable
phases,
and
then
we
will
discuss
in
which
form.
A
Okay,
I
can
leave
this
like,
so
I
don't
think
it
talks
about
the
introduction
of
new
fields.
It's
just
a
general
way
of
opting
addons.
So
I
you
know,
skipping
phase
is
already
supposed.
A
A
C
C
C
A
It
was,
I
think,
we
deprecated
in
118,
so
we
considered
it
as
like
an
implementation
detail.
We
still
gave
it
a
ga
duplication
timeline,
but
if
copy
cannot
adapt
for
122,
this
means
that
we
should
delay
the
removal
until
the
next
kubereum
api.
C
Technically,
it
is
possible
because
we
are
in
we,
we
plan
to
get
v1
alpha
4
in
in
june,
if
I'm
not
around
the
scene,
and
that
means
that
we
will
get.
We
want
alpha
for
in
time
for.
A
A
D
So
for
this
one
we
might
want
to
protect
what
sig
cloud
provider
is
doing.
I
know
for
the
labels
that
they
had
a
backfill
strategy
where
they,
depending
on
the
version
they
added,
they
had
the
beta
and
ga
labels.
A
So
basically
they
are
the
in
parallel,
multiple
annotations
and
they
phase
out
the
autonomous
slowly.
D
Yeah-
and
I
guess
the
trick
part,
the
tricky
part
here-
is
they
needed
to
care
about
downgrading.
B
D
And
that's
what
like?
That
is
one
of
the
I
guess
cons
of
version
and
annotations
and
labels
versus
when
you
have.
D
A
A
Does
he
talk
about
alpha
yeah?
We?
I
guess
we
have
to,
but
this
is
like
a
separate
process,
because
if
we
remove
custom
status,
the
graduation
of
the
api
server
advertise
address
notation
becomes
the
responsibility
of
this
cap.
So
we
shouldn't
close
the
original
issue.
I
guess
we
should
just
remove
the
cost
status.
A
And
it
requires
more
changes
incubation
to
clean
it
up,
but
it's
like
for
brits
who
said
it
seems
doable
if
costa
api
were
okay
with
that,
I'm
hoping
that
other
projects
do
not
depend
on
it.
Actually,
like
keep
spray,
I
need
to
maybe
check
with
them,
but
that
this
this
is
maybe
like
a
separate
email
but
yeah.
Let's,
let's
have
a
follow
up
here.
It's
it's
kind
of
tricky,
so
execution
plan.
A
I
I
honest
honestly,
wanted
to
break
the
whole
the
whole.
Basically
addition
of
v1
bit
3
into
multiple
pr's.
A
I
wanted
to
remove
the
v1
beta
1
and
addition
the
other
copy
of
v1
bit
3
based
on
v1
b22.
In
the
same
pr.
So
this
is
1pm.
We
remove
v1
bit
1,
we
add
a
like
a
copy
copy
v1p23,
that's
the
the
base
of
the
new
api,
and
after
that
we
cannot
start
applying
some
of
the
high
priority
changes
for
which
what
do
you
think.
C
I
agree,
I
kind
of
suggest
that
you
break
this
into
step,
one
remove
and
the
other
odds,
because
the
those
pr
are
usually
big
ones,
but
this
is
up
to
you,
but
but
I
totally
agree.
We
first
have
to
have
two
to
f.
Maybe
one
alpha
three
in
place
equal
to
v,
one
alpha
two,
so
the
conversion
part
is
really
simple,
and
then
you
start
modifying
v,
one
alpha,
three
and
and
and
and
changing
conversion.
A
Actually,
I
tried
it
yesterday,
but
it
was
with
three
commits.
I
ended
with
three
commits
the
first
one
is
the
addition.
The
second
one
is
the
removal
and
the
third
one
was
the
adoption
of
unit
tests,
because
our
unit
is
very
sketchy.
We
have
we
test
like
against
v1
beta1,
and
then
we
test
against
v1.
We
have
to
test
against
the
conversion
migration
to
v1
beta3.
A
So
I
am
not
sure
if
you
are
okay
with
just
single
pr,
with
three
commits
that
I
got.
C
I
don't
have
a
problem.
The
the
only
problem
is
that
if
the
pr
is
too
big,
it
takes
time
to
to
review
it
and
also
it
take
time
to
address
all
the
comments
so
it
at
the
end
it
takes
longer,
but
it
is
up
to
you
for
me.
It
is
fine,
also
a
pr
with
three
comments,
because
they
are
three
clean
separated
actions,
so
it
is
acceptable.
A
Yeah,
I
I
mean
I
the
fact
that
we
have
a
third
commit
that
adapts.
The
unity
is
even
better
because
if
I
combined
the
unit
test
updates
into
the
api
removal,
for
instance,
is
going
to
be
even
more
difficult
to
review.
A
A
A
C
I
I
think
that
it's
possible.
I
I
like
the
idea
that
we
apply
changes
so
in
separated
pr
it
changes
is
a
separated
pr.
It
should
be
possible
to
to
have
them
in
parallel,
maybe
a
little
bit
of
rebasing
game
when
one
merges,
but
this
is
doable.
C
What
what
what
is
what
I
think
that
we
should
do
is
to
define
priority
for
the
changes,
so,
let's
see,
for
instance,
having
a
don's
configurable.
We
consider
this
p0,
and
that
means
that
this
is
something
that
we
we
really
want.
Everything
is
else's
should
be
p1,
and
that
means
that
we
do
our
best
to
get
into,
but
is
not.
C
A
Yes,
I
agree,
so
I
this
is
the
next
point,
which
is
basically
the
list
of
open
issues
for
the
qb
repository
that
are
attacked
with
the
kite
api
change.
A
I
think
we
started
accumulating
that,
like
you,
can
see,
some
are
very
old
from
2017
is
the
oldest
2019
as
well,
but
basically
I
went
through
these
today
and
since
the
priority
important
zoom
label,
we
use
it
only
for
a
particular
release,
which
is
the
pending
release.
A
I
only
use
the
two
priorities
which
are
backward
and
long-term,
the
which
are
technically
lower
priority,
but
p0,
I
think
we
can
change.
The
priority
was
the
122
cycle
starts.
We
can
enable
the
other
two
priorities
which
are
critical,
urgent
and
important
soul,
which
are
accordingly,
p0
and
p1,
or
should
we
for
which
would
you
think
we
should
start
using
the
priority
labels
wrongly
and
start
labeling
from
now.
C
C
I
was
looking,
for
instance,
at
the
first
one,
the
first
one
I
I
don't
agree
it
is.
It
is
something
which
is
already
possible
so.
A
Sure
yeah
the
list
is,
you
know
it's
not
the
whole
list.
Of
course
it's
just
what
we
have
so
we
we
can
change
this
priority
to
awaiting
more
evidence.
For
instance,
I,
like
we,
we
all
agree
that
support
skipping
phases
is
the
p0.
A
A
A
This,
maybe
I
should
close
this,
make
the
kubernetes
config
easy
to
consume.
A
We
can
talk
about
it
later.
Object,
object,
better
is
object,
matter
is
literacy.
What
do
we
need?
Does
I
need
object
better
for
customize
and
things
like
that.
D
A
C
No,
it
is
not
but
yeah
the
the
ear
idea
is
to
make
our
types
more
customize
friendly,
which
is
for
the
users,
but
I
will
say
this
is
a
p2,
because
it's
something
that
came
out
in
the
past
from
kind.
If
I
remember
well,
but
now
it
is
not
a
problem
anymore
from
for
kind,
and
we
don't
have
a
user
complaining
about
these.
So
let's
say
it
is
a
pdu
nice
to
have,
we
can
add,
object,
meta,
and
there
is
a
reason,
but
is
there
is
not
a
real
demand
for
this.
A
D
And
I
guess
like
if
we
move
to
the
sierra
crd
friendly
option
that
also
that
might
also
drag
down
this
iron
and
bring
in.
A
Okay,
I
have
an
argument
that
all
the
fields
that
are
inside
the
cube,
adm
types,
which
are
basically
features
that
no
longer
work,
including
dns
type,
we
removed
proxy
sorry
cube
dns
support
entirely
and
the
other
one
is
hypercube.
A
C
C
This
is
the
discussion
of
before
how
do
we
implement
dns
skipping
by
skipping
phase,
or
by
being
explicit,
I
want
to
skip
the
dns
or
kuber
proxy.
This
is
something
that
that
we
have
to
figure
out.
I
I
don't
have
a
an
idea
now,
but
I
I
don't
want
to
to
obtain
in
the
idea
that
we
are
only
skipping
chases,
because
I
don't
have
yet
to
think
of
the
problem
that
that's
all
I
want
to
think
a
little
bit
about
about
it.
A
Oh
yeah,
you
just
pass
skip
phases
and
the
phase
that
you
want
to
skip
is
add-ons
slash,
dns
and
that's
it.
Even
if
you
pass
a
coordinates,
configuration
images
and
things
like
that,
it
will
be
skipped.
C
C
So
if
I'm
a
consumer
of
the
kubernetes
config-
and
I
wanted
to
understand
if
a
cluster
is
a
core
dns
or
not,
I
have
to
to
check
for
the
for
the
for
for
the
config
map,
okay,
which
is
on
the
and
direct
way,
and
also
as
some
counter,
let
me
say
indication
that
basically
people
installing
their
their
own
version
of
their
don.
They
cannot
use
the
same
name
of
the
config
mac.
So
there
are
pro
and
cons,
and
I
want
to
work
around
this
to
find
a
better
solution.
A
Okay,
okay:
this
is
going
back
to
the
skip
adults
problem.
I
see
yeah
all
right
this
one,
how
much
time
70
minutes?
Okay,
this
one
I
this
is
a
very
low
priority.
In
my
opinion,
this
is
related
to
basically
tagging
the
api,
so
that
k
work.
A
If
somebody
uses
the
cube
adm
api
with
the
recent
version
of
kwak,
if
they
print
our
api,
some
of
the
fields
will
be
hidden
like
fields
that
contain
secrets-
and
this
is
this-
like
a
nice
to
have
that's
why
I
I
have
this
priority
as
as
priority
backward.
I
would
say
this
is
like
a
p3.
A
A
C
A
A
A
Yeah,
okay,
we,
this
is
optional,
completely
graduate
patchwork,
so.
C
C
I
don't
know
that
that's
mean
graduating
the
feature
stuff
like
that.
If
you
want
to
keep
this
in
the
radar
for
me,
it
is
a
p3.
It's
not!
I
I'm
not
aware
of
people
of
user
asking
or
companies
asking
for
this
feature
so.
A
C
Okay,
let's
bring
let's
start
with
p3,
and
then
we
bring
nadir
first.
If
there
are,
if
there
is
a
use
cases,
then
I'm
happy
to
bump
the
priority.
A
Yeah,
I
think
the
use
case
was
that,
if
somebody
is
deploying
some
manifests
of
the
control
plane,
there's
no
way
for
them
to
customize
them.
Declaratively.
A
A
A
A
We
currently
pass
a
flag
to
the
couplet
so
that
the
kubrick
does
not
the
corpus
cable.
The
garbage
collector
for
images
does
not
remove
the
specific
version
type
of
the
pulse
image.
So
that's
like
a
couplet
feature.
That's
why
we
care
for
the
couplet
for
the
pulse
image
in
the
couplet,
but
long
term,
like
I
say,
is
going
to
be
a
cri
implementer
thing
for
air
gap
scenarios
on
the
side
of
cuba.
Adm
we
care
about
the
pause
image
because
users
might
want
to
pre-pull
it
now.
C
D
Constant
in
that
we've
had
eight
news
in
the
past
for
cluster
api.
Regarding
this,
the
the
issue
here
is
that,
if
you
make,
if
you,
if
you
have
an
ova
that
works
with
a
given
kubernetes
version,
but
you
want
to
have
still
want
to
have
like
a
different
pause
image,
you
end
up
having
to
build
another
ova.
If
you
want
to
specify
a
new
pause
image,
because
today,
if
you
use,
for
example,
container
d,
you
have
to
configure
that
through
the
timer
file.
D
So
ideally
we
should
be
able,
like.
I
know
I
don't
know
if
we
can
do
that
through
the
for
the
extra
arms
that
we
have,
but
if
like
that
seems
to
be
at
least
legitimate
use
case.
A
Yeah,
you
can
do
this
with
extra
rx
the
cobra
configuration.
I
don't
think
it
supports
the
pause
image.
Yet
maybe
it
does.
But
in
any
case,
if
this
is
moving
on
the
side
of
the
cri
implementer,
then
the
sierra
implementer
configuration
has
to
be
baked
in
the
ova
anyway.
So
it's
outside
of
the
scope
of
kubernetes
if
they
want
to
switch
the
pulse
image
for
container
differences.
C
Yeah,
let's,
let's
rally
on
the
on
the
issue,
but
the
the
the
key
point
for
me
is
that
kubernetes
should
not
configure
any
cri.
C
A
A
A
This
is
the
umbrella.
V1
b23
object
beta.
We
discuss
the
handling
of
extracts.
We
have
six
minutes,
so
this
one
is
that
if
you
pass,
I
think
we
have
an
example
here.
A
If
you
pass
extra
rx
that
are
the
same
cube,
adm
does
not
support
that,
and
it's
going
to
take
the
latest
like
we
unify
them.
A
There
was
a
proposal
to
use,
not
use
a
go
language
map.
The
default
map
object
but
use
something.
That
is
a
a
map
with.
I
think
the
multi
key.
I
forgot
the
name
of
that.
Basically,
let
me
see,
I
think.
A
I
think
this
should
be
a
stucco
or
or
this
one
like
we,
we
can
have
like
a
structure
that
is
contains
a
list
of
these
argument
structures
instead
of
the
map
string
string.
So
what
do
you
think
about
this?
It's
back
walk.
We
haven't
seen
that
many
requests.
A
C
A
I
mean
yeah,
it
can
have
a
separate
kept
even
no
problem,
but
it's
just
I
I
I
will
see
I
will
see
if
it's
very
low
or
what
priority
I
will
see.
If
I
I
can
write
something
up,
even
if
a
google
doc
you
know
not
a
kept,
but.
A
Yes,
okay,
switching
hcd
to
hcdsa
keys.
Currently
we
have
a
fairly
low
demand
and
to
my
understanding,
hcgsa
is
kind
of
tricky
in
terms
of
when
you
start
light
licensing
software
in
a
company,
hcsa
might
be
problematic
for
you,
so
I
asked
some
folks
around
and
enabling
this
by
default
might
be
tricky.
So
actually,
I
would
say
we
should
just
punt
this
item
for
now,
so
I'm
going
to
give
it
a
p4
which
is
awaiting
more
evidence.
Currently,
it's
alpha.
A
A
I
guess
I
guess
we
we
still
want
to
make
this
optional,
so
it's
if
it's
optional,
it
has
to
be
in
the
api.
So
I'm
going
to
ask
nader
about
this
as
well.
I
don't
know
what
is
if
somebody
has
opinions
like
you
can
of
the
law.
D
So
for
for
things
like
stig
and
other
regulations,
the
only
constraint
that
we
have,
I
think
is-
are
the
ciphers
that
we
use.
D
A
No,
we
do
not
support
actually
the
list
of
ciphers.
It's
just
everything
is
hardcoded.
A
There's
no
way,
there's
no
way
to
specify
the
ciphers
the
same
way.
You
can
pass
it
to
the
kubernetes
or
api
server.
If
that's
that's
what
you
mean.
D
Let
me
double
check
I'll
get
back
to
you
on
that,
but
I'm
pretty
sure
we
have
we're
using
that
time
stream.
So
it's
either
we're
not
aware
of
something
or.
A
We
are
out
of
time
by
the
way
I'm
going
to
assign
myself
as
asking
about
this,
and
also
I'm
going
to
ask
what
was
the
other
item.
A
Purchase
to
config,
okay,
I'm
going
to
ask
some
folks
and
maybe
know
more
about
this.
The
last
slide
that
we
had
is
image
for
policy.
I'm
going
to
add
this
as
backward,
and
I
don't
think
we
have
anything
else.
I
mean
if
you
have
anything
else,
we
can
discuss
it.
Please
open
a
ticket
for
the
qbm
repository
and
I'm
going
to
add
this
sp3
as
well.
A
So
no
problem.
We
are
out
of
time
and
I'm
going
to
stop
the
recordings
of
the
meeting.
You
can
continue
to
the
costa
rica
meeting
and
thanks
see
you
again
in
a
couple
weeks:
bye.