►
From YouTube: SIG Cluster Lifecycle - kubeadm office hours 2022-08-31
A
A
A
I
guess,
as
the
original
author
of
the
proof
of
concept
operator
now
also
represents
cluster
api,
who
have,
I
guess,
a
few
use
cases
for
this
operator,
but
fabricio
he's
not
working
on
that
right
now
who
actually
took
the
work,
is
paco
he's
a
qpdm
contributor
from
china
and
basically,
what
we
have
here
is
like
a
short
discussion
in
google
groups.
A
So
he
has
a
like
a
fork
based
on
fabricio's
work.
A
A
Perhaps
we
can
jump
to
the
coaster
api
thread,
which
is,
I
think
he
created
it
here.
Okay,
publisher.
We
want
to
give
some
other
context
about
this.
I
mean
like
we
can.
It
was
from
a
few
years
ago.
I
think.
B
Yeah,
if
I
got
it
right,
paco
started
from
my
very
first
prototype
of
this
thing.
B
I
think
that
one
point
that
we
should
keep
in
mind
is
that
there
was
a
kind
of
art
criticism
to
this
prototype.
It
was
that
it
basically
it
was
not
a
declarative.
It
was
not
following
a
declarative
approach
for
defining
things,
but
yeah.
B
B
Basically,
the
api
that
was
in
this
this
first
prototype
is
was
defined
around
the
idea
of
operation.
I
want
to
do
this
operation
and
this
is
kind
of
imperative.
I
tell
you
the
action
that
you
have
to
do.
Instead,
we
have
to
the
the
one
of
the
feedback-
I
don't
remember,
it
came
from
to
sinclair
or
someone
else
was
okay,
but
the
operator
at
the
end.
It
is
expected
to
drive
the
system
to
a
desired
state
and
our
api
does
not
allow
to
define
this
desired
state.
B
A
Right
yeah,
I
cannot
remember
this
discussion,
but
basically
yet
tim
was
going
for
a
declarative
proposal
for
everything
but
like
to
me,
and
you
have
a
discussion
that
it's
actually
not
possible,
for
example,
certificates
certificate
rotations,
there's
nothing
in
an
api
that
what
what
can
you
pin
inside
an
api
to
say?
Okay,
rotate,
my
certificates,
perhaps
a
date
or
something
like
that,
but
it
was,
I
think
we
got
blocked
on
the
discussion,
mostly
so
yeah.
A
And
I
remember:
you
eventually
removed
the
code
from
the
cube
dm
repo.
I
you
showed
the
demo.
I
think
the
demo
was
absolutely
fine.
Personally,
I
think
the
imperative
approach
is
fine
and
I
think
users
can
consume
it.
The
way
it
is
declarative.
You
know
the
clarity
for
the
sake
of
following
the
kubernetes
patterns.
For
this
thing
you
know
I
was
not
convinced
that
it
gives
us
any
benefits.
B
B
Gives
it
gives
a
lot
of
benefits
because
at
the
end,
when
you
want
to
define
an
operation-
and
you
want
to
move
your
system
from
one
state
to
another,
you
need
to
know
the
current
state.
So
declarative
means
that
basically,
you
have
a
representation
of
the
current
state
of
the
system
and
currently
kuben
mean,
does
not
provide
this
or
provide
this
partially
because
it
has
only
cluster
configuration,
but
it
does
not
provide
node
configuration,
etc,
etc.
B
B
B
I
don't
want
to
block,
let
me
say
what
I
can
say
is
that
how
can
you,
if
you
want
to
tell
oh,
I
want
to
upgrade
my
system?
How
can
you
upgrade
my
system
if
you
know
don't
know
the
version
of
your
system.
B
B
D
D
A
D
Oh
okay,
can
you
hear
me
probably?
No?
Oh,
yes,
better,
better,
okay,
so
I
I
just
said
that
I
am
the
review
stage.
I
do
not
know
what
exactly
I
am
planning
to
do.
The
nasio
is
a
new
project
proposed
in
linux
foundation
by
google
and
other
providers,
mostly
cloud
providers,
and
they
wanted
something
to
do
with
customization
for
the
5g
radio.
D
So
in
that
context
I
was
reviewing
what
are
the
requirements
for
qvarium,
cubelet
and
whatnots
to
enhance
the
crds
and
what
are
we
missing
so
in
that
context
I
was
trying
to
review
everything,
so
I
did
not
have
any
specific
agenda
today.
A
D
A
So
I
guess
I
think
one
big
question
here
is:
should
we
tell
paco
to
if
we
are
going
to
adopt
this
operator
or
something
that
is
going
to
be
under
kubernetes
6
eventually,
should
he
should
we
tell
him
that
hey,
we
have
to
redesign
this
to
declarative
completely.
B
A
Yes
and
you
I
remember
tim
saying
that
hey
we
can
create
this
proof
of
concept
in
the
repository
and
but
basically
later
he
reviewed
and
said
that
it
has
to
be
declarative.
I
personally
agree.
First,
I'm
still
exactly
clear:
how
are
we
going
to
do
certificate
rotation
declaratively?
But
if
paco
has
some
ideas,
we
can
put
them
in
a
cab
like
a
cap.
Iteration
could
actually
rewrite
the
kept,
so
we
can
first
agree
on
it
to
be
declarative.
A
B
B
B
D
B
D
D
So
if
you
say
cubelet
and
one
more,
then
it's
a
challenge,
but
if
it
is
a
drivers,
simply
a
plug-in
with
a
driver,
different
driver.
Maybe
I
will
revisit
this.
I
will
go
to
cluster
api
meetings
and
then
still
get
some
information.
It's
there
in
next
hour,
I'll,
try
to
find
out
some
information
and
share,
but
you
are
right.
One
and
two
is
doable.
Third,
with
respect
to
cluster
api,
we
cannot
say
whether
they
will
be
able
to
use
it
here
on
that.
C
Thank
you,
okay.
Now
my
microphone
should
work.
I
have
a
similar
takedown
for
pizza.
I
think
we
probably
literally
minus
the
way
to
have
a
good
idea.
How
mutability
and
class
api
should
work
if
we
actively
work
on
it,
which
you
currently
not
really
or
not
yet
so
it's
really
hard
to
guess
what
we
actually
want
to
do
and
how
that
operator
could
fit
in
or
not.
A
I
think
that
the
cubanium
users
will
benefit
and
the
basic
user
that
does
not
care.
If
this
is
an
agent
or
demonstrator.
A
I
think
we'll
have
a
like
a
fairly
solid
user
base
of
this
approach,
and
I
also
think
that
the
keep
spray
will
adapt
the
operator
they
currently
have
some.
As
far
as
I
know,
in
terms
of
upgrade,
you
know
managed
upgrades
by
the
high
level
2
and
rotational
certificates.
They
actually
have
some
ansible
script.
A
So
if
we
have
this
operator
that
works
for
kubernetes
users
cube
spray
users,
I
think
it
also
might
work
for
cluster
api
users
that
can
install
it
on
demand.
But,
like
you're
saying,
maybe
some
of
the
cost
api
users
actually
need
something
better,
which
is
the
agent
and
to
why.
A
A
So
perhaps
we
should,
by
you
know,
looking
at
this
discussion
with
backwards
in
the
quest
api
issue.
Perhaps
we
should
consider
whether
the
cubed
m
operator
will,
as
it
is
with
the
demonstrated
approach,
will
have
a
user-based
investor
appear
like
a
completely
custom
solution
that
people
can
install
if
they
want,
or
is
it
like
a
really
bad
idea
to
use
this
event
for
outside
of
kubernetes
and
the
kubernetes
user
base
for
cube
spray
for
costly?
B
Everywhere,
let
me
say
if
I
change
that
I
put
that
of
the
secretary
cyclotech
lead.
B
B
B
B
I
I
kind
of
think
that
before
green
lighting
these
there
should
be
an
agreement
at
seek
level,
so
there
should
be
a
proposal,
indeed
an
agreement
of
city
level
and
probably
not
only
on
the
operator
but
on
the
immutability
story
in
in
in
its
entirety.
A
Yes,
I
completely
agree.
So
what
will
happen
if
somebody
deploys
a
costa
rica
question
today,
maybe
even
with
costa
class,
and
they
apply
the
operator,
basically
deploy
the
kubernetes
operator
on
top
of
this
cluster,
not
in
a
potentially
declarative
model.
B
B
Kind
of
a
step
back
and
and
design
and
decide.
Okay,
where
we
are
going,
is
motivated.
How
much
would
we
want
would
ability
to
to
be
part
of
the
puzzle
at
which
level
et
cetera,
et
cetera,
et
cetera?
So
I
understand
that
that
time,
I'm
really
talking
about
a
grand
vision,
but
that's
it.
If
we
have
a
direction,
then
we
cannot
execute,
but
without
a
direction.
It
will
be
really
hard
that,
at
the
end
of
the
of
the
story,
the
all
the
pieces
fit
together.
A
So
I
wonder
if,
if
we
we
hold,
if
we
decide
to
hold
on
the
cluster
api
contention
around
mutability,
do
you
think
that
we
can
actually
produce
a
design
that
is
going
to
be
doable
in
the
future?
Because
I'm
also
considering
the
the
you
know
contributor
time
he
has
this?
He
works
for
a
cloud
provider
in
china.
They
have
the
use
case
where
they
want
to
easily
upgrade
big
questions
with
an
operator.
A
B
B
So
I
I
understand
that
a
paco
effort
and
it
started
from
a
prototype
that
was
dropped
and
deleted
for
reasons
and
yeah
now
he's
asking
feedback,
we
can
provide
feedback,
but
at
the
end,
if
he
is
willing
to
understand
to
listen
to
feedback,
that's
fine
and
probably
we
can
reconcile
these
to
the
seek
view.
Otherwise
it
is
a
personal
project.
It's
not
a
sig
project.
A
A
B
But
at
this
stage
in
cluster
I
can
tell
you
in
cluster
api
community,
there
is
no
discussion
about
motability
or
there
is
some
discussion,
but
they
are
really
really
necessary.
We
we
are.
We
are
it's
too
early
to
give
an
answer.
A
I
saw
that
jason
defended
the
immutable
approach
with
basically
someone
who
works
for
tinkerbell.
I
think
he
still
works
for
that.
He
basically
said
that
it's
like
much
better
to
rotate
the
hardware
instead
of
doing
multiple
changes
on
machines.
So
I'm
not
I'm
not
seeing
the
use
cases
in
question
api,
but
maybe
the
you
know
use
cases
elsewhere.
A
So
perhaps
this
should
be
our
feedback.
We
should
say
hey
at
least
start
with
the
declarative
pattern,
or
something
that
we
want.
B
Let's
take
a
step,
so
I
think
that
the
idea
of
the
clarity,
in
my
opinion
makes
sense,
but
this
was
a
feedback
given
two
years
ago
to
the
to
the
first
design
of
the
first
very
first
prototype
of
this
thing.
So
if
paco
is
believes,
that
is
that
his
current
design
is
robust,
a
cover,
a
set
of
use
case,
can
be
used.
B
So
I
I'm
I'm,
I
just
reported
of
the
fever
that
I
got
with
the
first
prototype.
I
think
that
this
is
still
valid.
That
is
true,
but
I
I
don't
know
what,
where
paco
want
to
go?
What
is.
B
B
A
Yes,
like
I
said
basically
the
use
case
to
just
have
upgrades,
I
think,
with
the
current
model,
which
is
the
imperative
operation,
and
I
have
not
seen
any
discussions
with
paco
where
we
basically
recommend
to
redesign
this
part
to
be
declared,
but
yes
you're
basically
saying
that,
even
if
he
creates
a
proposal
to
basically
release
this
operator
as
part
of
the
kubernetes
six
under
this
group,
if
people
don't
agree
on
that
that
this
is,
you
know
the
clarity
versus
superintendent,
if
they
don't
agree,
potentially,
we
cannot
accept
it.
B
This
is
this
part
of
getting
a
pro
project
adopted
by
the
sikh.
I
I
think,
let
me
say
we
capo,
the
paco
could
argue
or
one
queen
could
argue
that
today,
cluster
cuban
mean
is
imperative.
B
Okay
and
it
is
imperative
and
it
is
and
also
gives
to
the
users
the
responsibility
to
do
to
go
on
different
machine,
make
the
same
action
in
in
a
given
order,
etc,
etc.
So
it's
not
only
imperative,
but
it
gives
a
lot
of
complexity
to
the
user.
B
B
B
I
think
that
we
have
to
make
sure
that
our
all
the
things
fit
together,
but
I'm
open
to
different
point
of
view.
I
I
I
don't
have
a.
A
So
I
can
comment
on
against
the
google
groups
thread
to
say
that
if
paco
wants
this
to
be
adopted,
he
has
to
provide
the
design
with
the
declarative
approach
and
will
feed
the
feeders
his
api
ideas,
how
to
do
some
of
the
things
that
are
more
questionable
and
actually
purely
imperative
in
our
opinions,
such
as
kubernetes
configuration
changes
like
what
are
we
doing
here,
even
which
configuration
is
so
much
configuration
that
it's
not
clear
how
we
can
store
it,
maybe
a
field
with
a
same
semantic
version
or
something
that
is
not
enough
to
store
all
that
we
need
like,
for
example,
how
do
we
mutate
hcd
flux?
A
A
A
B
B
Should
this
be
a
new
project
in
the
sig,
or
should
this
be
an
experimenting
cuban
meal
if
it
is
an
experiment.
B
We
can
do
whatever
wrong.
It
is
an
experiment.
It
is
clearly
to
the
user
that
there
are
no
guarantees,
support,
guarantee,
etc.
B
A
A
A
It
might
be,
it
might
be
better
to
just
involve
more
people,
because
we
are
running
in
circles
really
it's.
Maybe
the
right
answer
is
just
a
combination
between
the
clarity
of
imperative
and.
A
A
So,
yes,
I
think
we
should
go
to
the
wider
ship
discussion
for
this
one
in
terms
of
getting
this
into
a
kubernetes.
Seek
repository,
compare
contrast
to
getting
into
the
cubed
repo
again,
I
think
we're
just
running
in
circles
like
putting
it
back
into
the
report,
I'm
personally
minus
one
on
that
and
we
can
just
keep
it
in
partners
repository.
It
doesn't
matter
in
my
opinion.
If
the
project
we
dropped
the
idea
entirely,
we
might
as
well
just
keep
it
as
a
fork
on
the
park.
A
I
think
it's
fine
because
essentially
he's
the
only
contributor
now-
and
I
I
don't
see
you
know
any
benefit
of
having
this
in
the
soup
folder
of
cabinet
right
now
and
yes,
we
shouldn't
we
shouldn't
release
it
as
a
repo,
separate
repo
in
the
k6,
until
we
have
design
that
we
agree
on
so
because
paco
is
in
a
completely
different
time
zone.
A
He's
probably
going
to
watch
this
vote
vod,
but
like
so
my
immediate
response
should
be
that
we
are
pointing
to
the
next
meeting
and
we
just
we
are
not
sure
what
to
do
with
the
design
at
this
point
but
yeah,
but
if
he
wants
to,
he
can
use
it
in
his
use
case.
The
way
it
is,
and
nobody
will
complain
it's
just
that
we
as
a
sig,
we
don't
want
to
use
it
the
way.
It
is
right
now.
B
A
D
Yeah,
I
think
it's
the
ownership
issue,
you
know
and
cluster
api
does
not
have
any
direct
answer
to
what
we
want
on
mutability
and
at
the
same
time
we
cannot
literally
be
dragged
into
something
which
we
are
not
coming
for.
A
Well,
we
have
to
hopefully
when
we
open
the
discussion
about
the
operator
in
the
sig
meeting.
We
will
have
some
representatives
from
cluster
api
where
we
can
discuss.
One
of
the
major
blockers
for
this
operator
is
to
also
you
know,
handle
the
quest
api
immutability
problem.
I'm
not
sure
if
that's
what
you
mean,
but
you
know,
hopefully
we
continue
discussion
on
this
part
more.
A
We
basically
need
a
lot
more
discussion,
and
it's
not
only
the
providers
requested
api
that
have
to
participate.
It's
basically,
all
projects
that
consume
kubernetes
have
to
participate
until
we
agree
on
the
design.
D
A
Leave
you
10
minutes
to
go
to
request
api
meeting
after
that
and
drop
a
quick
comment
for
paco
that
will
continue
in
the
discussion.