►
From YouTube: SIG Cluster Lifecycle 2021-10-05
A
A
B
Sure
I
I
everyone
so
some
time
ago,
in
this
meeting,
we
discussed
the
idea
of
basically
making
a
a
project
self-assessment
modeled
on
on
the
on
the
same
assessment
that
the
steering
committees
does
yearly
on
on
every
sig.
The
idea
is
is
not
to
is
to
basically
start
or
restart.
B
Interaction
with
some
project
that
we
are
kind
of
losing
from
our
radar
to
and
at
the
same
time,
to
enforce
the
not
to
enforce,
but
to
promote
the
the
same
best
practice
that
basically
kubernetes
is
asking
us
to
follow
like
keeping
honest
files
up
to
date.
Stuff
like
that.
So
if
you
let
me
share,
I
can
share
what
I
have
prepared
and
then
I
can
great
edit
answers
to
whatever
people
is
interested
in
this
work.
Thank
you.
A
B
Okay,
so
basically,
the
idea
is
is
that
these
are
goals
basically
read
from
the
community
assessment
questionnaire,
and
it
is
that
someone
from
from
each
project
comes
here
gives
it's
email
address,
because
the
questionnaire
is
related
to
an
admin
address
and
then
basically
compile
the
project.
The
the
questionnaire
for
a
project.
So
basically
one
person
for
for
each
project.
B
B
Are
you
keeping
meeting
notes?
Recorders
are
published,
so
this
should
be
really
fast
and
then
there
are
some
other
questions
on.
Are
you
having
additional
meetings,
backlog,
grooming,
some
stuff
like
that.
B
In
order
to,
let
me
say,
assess
how
the
projects
are
are
working
in
the
last
six
months.
Do
you
review
the
update,
you're
contributing
or
your
project,
my
roadmap?
So
this
is
basically
the
the
first
section
which
is
about
community
engagement.
Then
there
is
really
simple
and
offer
free
fast
to
answer.
B
Helps
people
to
onboard
stuff
like
that,
so
is
there
on
boarding
group
path
and
the
answer
could
be
yes.
The
same
of
bernie
is
basically
as
a
project.
What
program
do
you
participate?
Google
summer
code
stuff
like
that,
then
a
little
bit
questionnaire?
Do
you
have
project
contributes
or
from
multiple
companies?
Do
you
have
projects
from
use,
end
user
companies,
and
then
this
question
is
trying
to
assess
if
the
owner
fi
has
been
updated
recently
and
also
which
type
of
upper
base.
B
That
means
basically
what
the
project
is
currently
doing
and
the
focus
is:
is
there
something
that
that
you
want
to
to
highlight
to
to
to
give
a
shout
out
and
yeah
a
set
of
similar
questions
about
the
what
the
project
is
doing,
and
the
last
section
is
about
project
interaction
and
basically
is
about
how
the
if
the
project
is
participating
to
this
meeting
and
and
also
this,
this
great
try
to
figure
out
which
kind
of
interaction
we
have
between
our
projects
so
which
project
is
talking
to
each
other
and
and
yeah?
B
A
B
A
A
Yeah
in
terms
of
coverage,
I
think
this
is
pretty
good.
I
had
some
things
in
my
head
once
you
opened
the
dock,
but
it
was
already
covered.
So
I
think
the
coverage
is
pretty
good,
so
I'm
not
sure
if
anyone
else
has
any
additional
suggestions.
D
No,
I
think
I
think
that
looks
great
one
thing
I
didn't
see:
is
there
a
long?
Is
there
any
form
of
like
long
form
thing
like
anything
else
we
should
know
at
the
bottom
or
like
I,
it's
sorry
they're
hard
for
me
to
see.
Is
that
what
it
says
at
the
bottom.
B
Yeah
there
are
some,
let
me
say,
long
answer
field
where
you
can
write
whatever
you
want,
but
I
I
try
to
keep
them
to
not
use
them
too
much,
because
I
I
I
I
was
looking
for
a
trade,
fair
trade
off
between
making
this
simple
and
fast
to
compile
and
gather
information.
So
there
are
some
open
questions,
but
not
so.
D
Much,
I
think,
yeah,
I
think
it's
a
great
job.
I
think
you
did
a
great
job.
I
think
having
some
open
questions
is
good
just
so
that
there
is
some
channel
if
people
do
want
to
provide
it.
But
yes,
for
the
information
we
want
to
gather
from
everyone
like
multiple
choice.
Check,
share
boxes
is
ideal,
so
thank
you,
yeah.
I
think
this
looks
good.
A
Again,
once
we
gather
these
responses,
if
I
mean
it's
it's
first,
first
of
all,
it's
a
question
of
how:
how
are
we
going
to
contact
some
of
the
providers?
You
know
in
the
past,
not
providers
but
projects.
B
A
But
once
you
get
hopefully
gather
the
responses,
I
think
I'm
guessing
that
you
want
to
based
on
the
responses
and
evaluate
what
action
should
we
take.
For
instance,
if
some
of
the
responses
are
saying
hey,
we
haven't
updated
our
owner
files
like
is
there
a
forcing
function
that
the
group
should
perform
on
these
projects
so
that
they
can
grow
their
stuff
more
often
or
like?
That's?
Obviously,
we
can
get
the
feedback,
but
the
actions
after
that
are
more
difficult.
I
guess.
B
B
What
what
can
happen
at
the
after
this
form,
get
compiled
by
the
project
is
that
the
project
itself
comes
back
and
asks
us
how
to
do
things,
and
then
we
can
help
or
provide
example
how
things
are
managed
in
other
projects
or
otherwise
we
we
can
we
as
a
seek.
We
can
go
back
to
the
project
and
and
and
start
our
conversation,
for
instance,
what
if
we
see
that
a
project
is
basically
alone
something
stand
alone
in
in
this
community?
Not
not
not
interacting
with
all
the
other
projects
is.
B
That
is
that
something
that
we
want.
Is
that
something
that,
at
least
in
my
opinion,
we
should
go
and
check
with
this
project.
Why
are
you
not
interacting
with
the
others,
because
at
the
end
we
want
to
solve
the
problem
of
cluster
life
cycle
management,
and
what
we
have
in
mind
is
is
just
is
a
vault
run
of
tools.
B
Basically
is
a
compositional
tools
that
work
together
and
if
there
are
some
projects
that
are
reinventing
the
wheel,
maybe
that
we
are
not
doing
our
role
as
a,
let
me
say,
as
a
coordination
in
a
proper
way,
but
I
don't
think
that
is
for
it
is,
let
me
say
it
is
in
a
different
way.
This
is
not
meant
to
enforce
it.
It
is
meant
to
start
a
discussion
and,
let's
see
together
with
the
project
where
this
discussion
leads.
A
I
agree
we
can
potentially
decide
what
we're
going
to
do
with
the
data.
You
know
one
side
effect
of
basically
one
trigger
for
this
collection
of
data
is
that
steering
wants
us
to
keep
our
super
projects
in
check
and
they
also
want
to
know
if
they
update
the
owner
files.
If
they
do
not,
if
we
say
hey
our
super
projects,
do
not
update
donor
files,
we
obviously
have
a
governance
problem,
so
it's
a
bit
difficult
to
gather
the
data,
but
later
not
suggest
any
action.
A
B
A
Right
so
I
mean
grant
pd
access,
you
can
also
grant
the
access
to
to
be
suggesting
if
they
want
to
make
edits
and
we
can
go
from
there.
But
do
do
you
envision
publishing
this
before
or
after
cubeco.
A
Okay,
I
added
one
rather
interesting
topic,
which
is
about
downgrades.
I
see.
Occasionally,
users
are
asking
even
outside
of
cube,
adm
usage.
They
are
asking:
how
do
you
downgrade
a
community
sequester
and
I
don't
think
we
have
documentation
for
that
in
the
kubernetes
website.
A
Basically,
there
are
some
blog
posts
out
there
how
we
can
do
like
row
outs
and
downgrades,
but
it
feels
like
people
are
bootlegging
downgrades
and
I'm
not
sure
that
some
of
our
higher
level
projects
support.
I
know
that
costa
rica
does
not
support
it,
I'm
not
sure
about
cube
spray.
Potentially,
my
guess
is
that
they
do
not
spawn
just
you
have
a
do.
You
support
in
cops
that
I
don't
know.
E
We
certainly
don't
recommend
it.
I
I
think
we
don't
block
it.
D
So
it
probably
will
work
if
kubernetes
works
right.
If
you
go,
if
you
downgrade
by
four
versions,
it
won't
work.
If
you
don't
grab
a
one
version,
it'll
probably
work.
A
Okay,
like
can
you
clarify
what
is
that?
What
is
the
ux
rather,
do
you
have
a
command
or
like?
Do
you
just
declare
it
if
you
change
a
version
or
how
do
you
do
it.
D
Yeah
you
the
way
upgrades
work
is
you,
you
basically
specify
the
new
version
you
want
to
go
to
and
an
upgrade
is
when
that
version
is
higher
and
so
a
downgrade
you
would
just
specify
a
lower
version.
We
don't
validate
that
or
enforce
that.
So
that
would
that
would
you
know
the
user
is
able
to
specify
whatever
they
want.
D
D
I
would
expect
that
most
people
that
do
a
downgrade
do
it
very
rapidly.
In
other
words,
they
get
one
one
of
their.
They
have
three
control
plane
nodes,
they
do
one
and
it
fails,
and
so
they
go
back
down
or
they
they
do
the
upgrade,
and
then
they
discover
very
quickly
that
things
are
not
healthy,
but
I
imagine
once
people
are
running
it
for
a
while
they
do
not
downgrade,
but
in
theory
it
should
work
from
the
sort
of
kubernetes
perspective.
D
As
long
as
you
go
down
by
one
minor
version
and
it
sort
of
has
to
work
because
of
the
multi.
The
h8
control
plane
scenario,
where
you
can
have
control
plane,
nodes
of
different
versions,
so
that
that
is
it
is
supposed
to
work.
I
I
don't
know
to
what
extent
it
is
actively
tested
and
I
certainly
wouldn't
recommend
it
for
people
that
have
been
running
a
cluster
for
very
long.
A
A
I
guess
in
cops
that
you
say
that
you
don't
recommend
it,
but
if
you
have,
you
are
following
this.
This
document
that
I
opened,
which
is
something
that
kubernetes
does
not
follow,
which
is
there,
is
a
order
of
upgrade
between
cube
controller
manager,
scheduler
controller
quad
control
manager
that
have
to
be
upgraded,
not
at
the
same
time
as
the
api
server
of
particular
machines.
D
We
are,
we
are
not
following
that
order.
No,
it's
very
difficult
with
the
I
guess,
traditional
setup
right,
where
you
have
a
control
plate
machine
that
runs
all
those
components
and
you
bring
up
a
new
control
blade
machine
in
an
immutable
sense
that
is
running
the
new
versions
like
it
makes
it
very
difficult.
So
no,
we
are
not
following
this.
This
order.
A
Basically,
tim
at
the
bus
have
said
that,
because
what
we
most
projects
are
doing
is
that
we
deploy
everything
on
the
same
machine.
We
just.
We
hope
that
we,
the
machines,
are
going
to
get
to
a
healthy
state,
eventually,
like
the
components,
will
restart
many
times,
maybe,
but
they
will
get
healthy
yeah.
I
think
it's
a
good
argument.
I
I
spoken
with
jordan
in
the
past
about
this.
He
said
that
it's
it's
actually
critical
in
some
cases
to
upgrade
to
follow.
B
A
Order
exactly
I.
I
first
thought
that
only
the
kublet
skew
is
a
problem
where
you
know,
there's
also
a
corporate
upgrade
order,
but
jordan
also
said
that
the
the
excuse
between
these
components
can
be
a
problem.
D
D
I
believe-
and
so
I
think
that
was
the
thing
that
that
caused
trouble,
if
I
recall,
but
in
general
they
are
kk,
is
supposed
to
introduce
fields
in
one
version
and
start
using
them
in
the
next,
and
we
should
avoid
using
fields
that
are
newly
introduced
that
allows
for
like
that
means
that
fields
won't
disappear.
D
The
cloud
controller
manager,
honestly,
I
don't
know-
I
also
don't.
I
don't
personally
know
of
anyone
following
this
procedure-
that's
not
to
say
that
no
one
is,
I
just
don't
know
of
anyone.
A
I
was,
I
was
under
the
impression
that
gk
does
it,
because
you
know
back
in
the
day,
robert
was
saying
that
the
kuba
president
was
running
on
a
separate
host
instance,
so
it
allowed
for
this
step
by
step
upgrades.
Maybe
I'm
stating
this
wrong.
D
That
is
not
my
understanding,
but
my
understanding
might
be
out
of
date.
I
I
don't
think
it's
the
case.
I
think.
D
I'm
trying
to
think
whether
anyone
like
I'm
trying
to
whether
I
know
of
anyone
that
in
open
source,
particularly
that
does
this,
I
can't
you-
have
to
have
like
different
control.
You'd
either
have
to
change
your
machine
in
place
or
you'd
have
to
have
a
split.
Actually,
maybe
maybe
cap
n
is
doing
this-
the
next,
the
the
media
guy
immediately
after
this
cluster
apr
provider.
Nested
may
be
doing
this.
I
don't
know
like
because
they're
running
just
deployments,
if
I
understand
correctly
so
they
might
be.
A
I
don't
think
open
shifts
are
doing
it
either
yeah.
But
again,
this
is
we're
going
into
the
details
here,
but
the
reality
is
that
we
don't
really
have
any
guides
for
downgrades
and
anything
in
kubernetes
doesn't
really
support
it.
So
I
was
wondering
whether
we
should
eventually
because
it's
in
our
charter
actually
as
a
world
downgrade,
we
should
start
eventually
supporting
for
others.
I
think.
D
I
think
it's
a
good
idea.
I
think
that
the
so
two
things
I
just
want
to
mention.
Chaos
has
started
running
api
server
only
nodes,
so
this
is
something
in
theory
we
could
do.
We
could
like
run
a
lot
more
vms
that
were
smaller
in
size
and
do
that
sort
of
thing,
I'm
a
little
bit
wary
of
doing
that
like
it's,
not
a
great
match,
for
you
know,
running
separate
vms
is
a
bit
of
a
pain,
but
I
mean
the
alternative.
Is
that
we
start
running?
D
We
start
changing
machines
in
place,
which
is
probably
worse.
The
other
thing
is
this
is
not
just
about
downgrades.
This
happens.
This
problem
happens
during
h.a
upgrades
pretty
severely
and
like
that's
the
one
which
actually
is
going
to
bite
people
more
often,
I
think,
than
a
downgrade
yeah.
That's
because
that's
been
here.
A
D
It's
it's
it's
that
problem
and
and
yes
like,
there's
no
there's
no
guarantee
when
you
have
three
coupe
controller
managers
that
the
leader
will
be
of
the
the
newest
version
or
the
oldest
version,
whatever
your
guarantee,
whatever
guarantee
you're
looking
for
that
guarantee
does
not
exist,
and
so
the
compounding
problem
is,
if
you,
if
kube
controller
manager,
uses
update
and
a
field
does
has
been
removed,
it
will
just
delete
it.
Sorry,
if
it's
been
added,
it
will
just
delete
it.
D
If
it's
the
older
version-
and
that
was
the
problem
we
saw
back
in
the
day-
I
think
you
know
that's
addressed-
perhaps
by
using
server
side
apply
or
by
using
patch
everywhere.
But
I
honestly,
I
don't
know
to
what
extent
that
is
all
present
and
it's
not
it's
scheduler
as
well
to
any.
In
theory,
any
component.
A
There's
a
there's
a
kip:
there
is
work
in
progress,
I'm
not
sure
for
which
version
it
will
land,
but
there's
a
kept
for
a
controllable
leader
election
where
during
upgrade,
you
can
say.
Okay,
this
instance
is
the
one
I
want
to
lock,
or
I
prefer
to
have
this
sequence.
You
know
of
leader
erection
transitions.
A
Vince.
Do
you
have
a
comment.
F
I
just
have
a
comment
on
the
upgrade
in
general,
like
the
or
like
sorry,
the
downgrade.
I
guess
in
this
case
I
would
say,
like
this-
is
probably
like
a
something
like
that
could
be
potentially
like
a
very
dangerous
like
to
do
on
like
a
case-by-case
basis,
but
I'm
thinking
about
like
a
rollback
instead
so
like,
and
this
is
only
something
that,
like
a
tool,
probably
outside
of
cubitium,
I
guess
could
do.
F
But
I
don't
know
like
how
would
you
do
it
like
in
cupidian
like
directly,
but
we
have
thought
about
like,
for
example,
cubedium
control,
plane,
enclosure,
api
to
support
downgrading
as
in
roll
back
the
version
to
what
it
was
before,
not
just
like
downgrading
to
any
version
so
like
if
the
upgrade
styles
or
something
do
a
safe
roll
back,
but
not
necessarily
downgrade.
If
that
makes
sense,.
F
No,
no,
we
don't
have.
We
don't
have
the
ability
to
do
that
today,
like
we
just
have
thought
about
it
but
yeah.
I
guess
like
you,
would
have
to
preserve
some
sort
of
state
or
I
guess
like
look
at
the
current
state
of
the
cluster
and
say
like.
Oh,
I
still
have
one
of
the
old
control
plane
nodes
instead
and
so
like
maybe
roll
back
is
feasible.
A
I
see
yeah,
but
that
that
is
roughly
what
qbm
does
on
the
mutable
upgrades.
It
preserves
the
like
an
intermediate
state
of
the
manifest.
If
the
upgrade
fails
it,
the
user
can
go
back
to
those
also
the
the
m
command
can
do
it
automatically
for
them
at
least
from
the
first
load.
I
think
the
upgrade
fails.
A
Qb
is
going
to
say:
hey,
I'm
going
to
roll
back
the
old
manifest
because
they're
great
fail
yeah.
That's
that's
what
we
support
today,
roughly
incubation
at
least,
but
maybe
we
should
start
claiming
that
in
general
downgrades
that
are
not
intermediate
during
intermediate
upgrades,
basically
and
supported-
and
maybe
that's
a
point
to
upgrade
to
add
to
the
charter
like
what
do
you
think
about
it?.
F
Yeah
I
mean
if
we
do
have
to
update
the
charter
to
like
have
to
say
maybe
roll
back
rather
than
downgrade.
That
probably
be
better.
I
guess
we
just
have
to
find
the
right
word
right.
D
I
think
that's,
I
think
we
should
people
are
going
to
want
to
want
to
downgrade,
at
least
in
theory,
and
I
think
we
should,
when
we
add
this,
I'm
in
favor
of
adding
this
it's
more
more
truthful.
D
But
when
we
add
this,
we
should
identify
like
why
we
feel
that
we
can't
support
downgrading
just
put
like
some
breadcrumbs
like
you
know,
we
need
to
see
if
we
had
separate
like
a
separate
vm
for
each
for
each
component.
We'd
feel
happy,
maybe
but
whatever
it
is,
that
we
that
blocks
us
like
that's,
we
should
leave
a
leave,
a
breadcrumb
for
future
people
that
want
to
support
downgrades.
I
guess.
A
D
I
wouldn't
say
the
cap
supports
it.
I'd
say
we
tolerate
it.
We
don't
block
it
like
you
know,
I
don't
think
we
have
a
lot
of
testing
if
any
on
it.
I
think
I
think
we
could
try
to
establish
some
testing
on
it.
I
think
that'll
be
a
good
thing
to
do
on
downgrade.
D
If
anyone
wants
to
support
it.
I
think
that'd
be
great
if,
if
a
neutral
wants
to
come
along
and
say,
hey
we're
gonna
architect
in
a
different
way
and
we're
gonna
support
downgrade,
I
think
that's
a
great
feature.
You
know
for
some
new
tool
like,
but
don't
see
us
doing
it
immediately
in
payoffs.
F
It
probably
does
make
sense
not
to
block
it
necessarily
but
like
rather
than
like.
F
If
there
is
something
in
the
chart
that
says
like
we
will
support
upgrades
or
like
downgrades,
maybe
that
that
could
change
like
to
say
like
we
might
support
safe
rollback,
but
we
won't
support
like
fully
downgrading
a
cluster,
like
I
mean
you're,
all
kind
of
on
your
own,
if,
like
you,
have
like
a
stable
close
and
you
just
want
to
like
go
two
or
three
versions
behind
you
know
have
fun,
but
but
I'm
saying
like
from
a
sick
perspective
from
our
tools
like
we
probably
don't
want
to
share
ourselves
in
the
foot
right
like
we
want
to
like
do
safe
operations
and
support
safe
operations,
so
that
that's
like
that
was
like
my
comment,
I'm
not
saying
like
we
should
block
it
necessarily
we're
doing
cluster
api.
F
We
do
block
downgrades.
If
I
run
correctly
or
by
like
rollback,
we
might
say
like
we
might
support
rollback
in
some
of
the
tools.
A
Kuberium
also
blocks
it,
but
users
can
force
essentially
an
older
version.
So
in
a
way
we
already
support
like
downgrades.
D
D
A
Yes,
that
is
true.
You
know
one
one
problem
that
I
see
if
I,
if
I
go
and
try
to
pr
this
charter
document
today,
it
has
to
pass
review
by
the
steering
committee
and
civ
architecture
who
owned
the
version,
skill
policy
and
upgrade
sequence
of
components,
and
if
I
start
to
play
with
the
wording
around
downgrade
and
say
that
it's
not
recommended
my
pr
is
probably
going
to
be
bought.
A
A
I
think
openshift
supports
it.
You
can
downgrade
the
poster,
I'm
pretty
sure.
So
we
are
basically
saying
that
we
don't
recommend
it.
We
want
to
change
the
the
charter
and
basically
some
of
the
kubernetes
claims,
but
it's
actually
possible.
So
I'm
not
sure
I'm
not
sure
this
pr
is
going
to
be
accepted.
Basically.
D
Fine
to
say
that
we
don't
have
any
tools
that
support
that
that
are
designed
to
make
upgrades
safe.
How
about
that.
A
Yeah
yeah,
but
also
this.
This
is
like
intermediate
state
in
the
charter
and
to
my
understanding
we
should
be
documenting
in
the
general.
What
is
the
current?
Like?
Not,
okay,
let's
have
a
statement.
A
Basically,
the
charter
is
responsible
for
defining
scope
of
a
group
and
dow
grades
according
to
everyone
is
for
this
group,
but
we
cannot
really
say
that
here
we
don't
support
downgrades
now,
but
we
are
going
to
support
them
in
the
future
or
something
like
that
or
if
you
want
tools,
you
can
write
them
but
yeah.
I
think
it's
going
to
be
difficult
to
change
the
world
in
here.
Maybe
we
just
need
better
precisions
precision
for
what
we
want
to
state.
B
Yeah,
what
is
at
least
what
I
don't
know
personally
is
at
which
stance
a
downgrade
is
feasible,
because
what
we
have
in
kubernetes
is
a
skew
is
a
set
of
skew
rules
and
these
q
rules
work
work.
Well,
while
you
are,
you
do
upgrades
with
some
caveats.
For
instance,
when
you
are
nha,
you
have
to
make
sure
that
your
components
stay
consistent
with
one
api
server
stuff
like
that.
So
but
they
work
work
for
upgrades,
but
when
I
think
to
downgrade.
B
What
if,
if
I
upgraded
the
storage
version
for
my
object,
change
it,
and
then
I
try
to
downgrade
I'm
not
sure
that
the
downgraded
component
will
be
able
to
read
the
upgraded
start
version,
and
currently
I
don't
have
a
senior
to
recognize
that
the
storage
version
is
already
changed
or
not.
B
So
I
I
think
that
before
okay
now
we
have
downgraded
in
the
charter.
I
remember
last
time
we
reviewed
the
charter.
I
remember
pointing
out
that
we
don't
have
nothing
to
support
it,
but
yeah.
This
is
a
matter
of
wording
at
least
what,
as
far
as
I
understand,
the
problem
is
that
okay,
we
can,
if
the
user
needs
these,
we
can
try
to
really
make
it
possible,
but
at
the
end
we
really
depends
of
the
downgrade
rules
for
each
component
and
in
my
opinion,
those
rules
are
not
stated
in
properly
or
clearly
enough.
A
From
what
I
understand,
the
the
scenario
that
you
mentioned
with
the
storage
version,
it's
actually
supported
by
the
kubernetes
skew
and
the
version,
promotion
and
deprecation
policy.
So
it
was
already
thought
people
already
thought
about
this
problem,
and
I
was
told
I
think
by
jordan
at
least
some
time
ago
that
this
is
reported.
A
And
the
guarantees
in
a
in
the
promotion
of
apis
basically
guarantees
that
you
can
also
do
a
downgrade.
This
is
my
understanding.
A
So
if
the
version
I'm
trying
to
think
about
this
scenario,
so
the
version
changes
in
this
now
it's
the
version
is
v1.
A
You
know
basically
saying
that
the
client
can
no
longer
read
this
version
from
the
server,
but
then
to
by
the
setting
the
the
api
server
will
down
convert
the
storage
version
to
like
the
one
beta,
one
that
is
still
supported,
and
the
client
will
be
able
to
read
it
or
something
like
that.
Maybe
not.
B
B
A
It
was
using
the
old
upgrade
downgrade
framework.
Let
me
check
how
much
time
do
we
have.
B
Yeah,
maybe
I
I
did
not
research
the
the
topic
deeply,
so
I
I
admit
that
I
I
can
have
legs
of
information.
C
D
I
think
you
know
to
your
point
for
betel,
even
if
and
and
liberia
like,
even
if
the
even
the
policy
of
kubernetes
is
that
we
don't
do
these
things,
like
mistakes
can
happen
and
that's
why
we
need
the
the
tests
right
and
if
we
don't
have
the
test,
which
I
don't
think
we
do,
we
certainly
don't
have
tests
which
include
cloud
controller
manager,
upgrades
and
downgrades
like.
Then
we
should
not
assume
that
it,
it
all
works
perfectly.
D
D
So
it's
it's
a
complex
sequencing
problem,
but
yeah
I
feel
like
until
we
have
tests
we
shouldn't.
We
shouldn't
be
saying.
This
is
something
that
that
we
that
we
expect
to
work
like
there
probably
should
be
a
cavity
in
there
being
like
hey.
This
is
not
only
other,
no,
no
tools
that
we
know
of
them
want
this,
but
there
are
no
tests
that
verify
it.
So,
even
if
you
found
the
magic
tool,
there's
no
guarantee,
it
will
actually
work.
A
Yeah,
I
completely
agree
if,
basically,
I
was
telling
users
for
a
very
long
period
of
time
that
downgrades
are
not
recommended
they're,
not
really
supportive.
That's
pretty
much!
Also
what
cuba
adm
says.
We
have
this
in
the
documentation
of
kubernetes,
so
I
think
we
we
can
continue
charter
wise.
We
can
continue
owning
the
potentially
the.
A
The
whole
scope
of
simplifying
downgrades
eventually,
but
I
think
I
agree
that
we
should
continue
saying
this
to
users
that
there
are
no
guides.
You
shouldn't
downgrade
pretty
much.
D
I
really
like
I
agree.
I
think
I
really
like
the
the
notion
of
a
of
a
rollback
or
whatever
we
whatever
we
said.
I
think
you
know
vince.
D
I
think
you
said
rollback,
but
that
that
to
me
is
the
the
scenario
which
we
should
be
able
to
support
and
I
think
it's
so
much
users
care
about
like
I
go
halfway
through
an
upgrade
or
an
upgrade
plus
five
minutes,
and
I
realize
that
everything
is
falling
apart
and
I
need
to
like
push
the
red
button
and
get
back
to
safety,
and
I
understand
that
I'm
not
gonna
run
this
for
a
month
and
then
go
back
and
I'm
certainly
not
gonna
go
back
by
two
versions,
but
if
I,
if
I
if
something
goes
really
wrong,
I
want
that,
but
the
ability
to
like
eject
and
go
back
to
where
I
was
you
know
an
hour
ago.
A
Yeah,
like
I
said
I
don't
know,
what
is
the
state
of
cube
spray
for
coaster?
Api
maya,
the
city
is
a
vincent
fabricio.
You
want
to
potentially
support
this
robux
scenario.
F
Potentially,
yes,
it
needs
a
cap
upstream
and
it's
probably
just
going
to
be
cupid
yen
control
plane
we
have
just
mentioned
like
here
and
there
like,
because
you
know
like
upgrades-
could
fail
in
a
number
of
different
ways
as
well.
So
it
makes
sense
to
support
a
rollback
scenario
like
exactly
have
deployments
do
or
similarly.
A
Let's
see
right,
I
mean
it's
going
to
be
nice
to
have
definitely
does
anybody
else
have
comments
for
this
topic.
We
should
probably
move
on.
F
Yeah,
so
I
just
wanted
to
do
a
call
out
like
I
was
looking
at
the
add-ons
project,
so
there
is
like
a
topic:
that's
coming
up
in
boston.
I
think
cecile
brought
up
last
week
as
well
during
this
office
hours,
and
we
have
upcoming
changes
in
kubernetes.
F
There
are
going
to
put
even
more
things
out
of
three
and
like,
for
example,
cpi
csi,
and
I
guess
cnn
is
already
there,
but
as
these
out
of
three
atoms,
which
are
required
to
have
a
functioning
cluster
grow,
and
there
is
like
a
question
of
like
how
do
we
generally
support
add-on
management
within
cluster
api,
given
that
the
goal
of
this
project
is
to
support
to
create
a
functioning
kubernetes
cluster
with
cni,
like
our
quick
and
dirty
solution,
was
to
add
support
for
cluster
resource
set,
which
was
honestly
logistic
pathways
like
it
was
just
the
quickest
way
to
install
yaml,
not
necessarily
to
cni
but
yamo
when
a
cluster
is
created.
F
So
I
guess,
like
one
question
was
like
here,
I
didn't
understand
the
status
of
the
adams
project
and
also,
like
I
didn't
understand
how
to
consume
it.
If,
like
it's
very,
you
know
it's
being
maintained,
but
if
it's
not
maintained
should
we
consider
climbing
it
should,
or
you
know,
if
folks
are
stepping
up,
then
how
do
we
make
sure
that
we
can
consume
it
sooner
because
it's
123.
D
I
can
probably
take
this.
I
think
the
it
is
very
much
we
are
trying
to
get
into
chaos.
First,
I
don't
think
anyone
is
intending
for
the
project
to
be
exclusive.
In
other
words,
if
you,
if
you
have
another
path-
and
you
want
to
pursue
it-
no
one's
going
to
say
hey
you're,
breaking
the
rules,
the
the
we
are
trying
to
solve
it
in
cluster
add-ons
and
you're.
Also
welcome
to,
like
you
know,
bring
different
approaches
to
that
forum.
It
looks
like
we
will
get
it
into
chaops
fairly
soon.
D
We
do
not
yet
have
a
good
answer
for
I'd,
call
them
cloud
resources
in
particular
permissions,
and
that's
why
I
think
we
talked
a
little
bit
with
cecile
about
last
week,
where
we,
you
know
referenced,
possibly
using
like
case
the
kubernetes
config
connector
on
gcp
and
the
amazon
ack
amazon,
something
connector
for
kubernetes,
perhaps
on
amazon
like
some
sort
of
some
sort
of
way
of
describing
im
permissions
using
that
sort
of
thing.
D
But
honestly,
we
don't
have
a
good
answer
for
that
yet,
but
we
are
trying
to
get
add-ons
into
consumable
in
projects
in
ways
that
make
sort
of
existing
users
happy
and
like
maintain
the
existing
security
guarantees
that
they
have,
which
is
sort
of
almost
impossible,
but
we've
we've
threaded
the
needle.
I
think,
in
terms
of
being
able
to
support
updates
without
being
without
changing
the
sort
of
scope
dramatically
of
what
you
know.
D
What
runs
on
my
cluster
as
it
were
in
terms
of
cluster
api
you're,
perhaps
in
a
bit
better
state
in
that,
if
you're
willing
to
run
an
operator
that
describes
those
iem
resources
on
the
management
cluster,
then
you
have
a
nice
place
from
which
to
configure
them.
D
You
know
what
I
mean,
so
that's
so
you
can
configure
iem
from
the
parent
from
the
management
cluster,
but
we
don't
have
that
and
then
we
can
also
have
add-ons
that
run
either
in
the
cluster
or
in
the
management
cluster.
So
it's
up
to
you
or
add-on
operators
that
run
a
new
place,
but
I
think
I
think
that's
it's.
We
haven't
the
the
place
of
first,
where
we're
trying
to
get
it
in
first,
purely
because
we
sort
of
stole
that
on
kubedium
is
into
chaos.
D
Cluster
api
would
be
super
interesting
and
might
be
able
to
help
a
lot
with
the
with
the
I
am
type
resources,
and
so,
if
you
wanted
to
try
to
get
it
into
cluster
api,
I'd
love
to
talk
about
that
events,
but
if
there's
other
alternatives
you
want
to
pursue,
I
have
no
objection
to
that.
The
the
general
goal,
I
think,
is
if
we
try
to
get
all
these
things
described
as
yammer
manifests.
An
operator
can
install
them.
D
Ideally,
they
will
be
very
simple
to
install,
but
I
am
is
certainly
a
wrinkle
in
that
simplicity.
F
Okay,
thanks
justin
that
that's
like
really
useful
context
there.
The
impermission
specifically
like
I
guess
like
those
like
mostly
aws,
related,
but
I'm
sure
that
other
cloud
providers
have
other
things
that
need
to
be
configured.
It's
something
that,
like
the
the
provider
itself,
like
will
kind
of
guide
the
user.
So
we
just
as
far
as
a
cluster
api
we
just
have
to
give
out
or
like
create
apis
to
install
these
add-ons,
and
so
then
you
know
like,
but
there
are
apis
in
this
project
as
well.
F
So
I
guess
like
if
these
are
not
meant
to
be
exclusive,
maybe
what
we
can
do
is
to
solve.
You
know
the
problem
like
separately,
for
not
just
for
cluster
api
to
give
the
those
apis
to
be
pluggable
and
then,
as
these
projects
mature
over
time,
we
can
look
to
integrate
with
each
other
later
and
make
with
application
one
api
versus
the
other
or
vice
versa,
and
we
can
just,
I
guess,
convert
and
converge.
D
Yeah,
I
think
I
mean
I
think,
that's
fine.
I
think
if
we
have
an
operator
first,
like
let's
say
the
the
external,
the
aws
cloud
controller
manager,
you
should
you
should
be
able
to
run
that
operator
in
the
management
cluster
or
in
your
child
clusters.
I
don't
know
as
you
as
you
see
fit,
so
I
think
there's
a
lot
of.
D
F
Okay,
I
mean
yeah,
we'll
take
we'll
take
a
look
at
the
apis
like
there
are
like
described
in
here.
It's
mostly
because
the
question
was
like.
Should
we
create
add-ons
api
in
cluster
so
class
api
add-ons
api?
F
I
guess
to
to
install
and
like
make
it
plugable
as
well,
because,
like
I
could
see
like
folks
who
wanted
to
use
helm
or
they
want
to
use
k
app
from
kerbel
and
vmware,
or
they
want
to
use
customize
or
whatever,
there's
a
lot
of
thinking
that
like
it
needs
to
be
done
here
for
sure
but
yeah.
I
guess
like
we'll
gather
more
information,
and
then
we
can
circle
back
once
we're
more
mature.
D
Yeah,
I
think
that
the
for
installing
the
or
for
managing
say
the
aws
cloud
controller
manager.
What
we
have
been
pursuing
is
the
idea
that
that
should
be
its
own
crd
like
strongly
typed,
and
that's
because
it's
sort
of
a
piece
of
software
that
we
understand
well
and
or
that
understand
the
stuff
well
and
wants
to
expose
like
certain
bits,
no
function.
D
Excuse
me
no
functionality,
but
the
sorry
for
the
the
installation
of
the
operator
itself.
That's
a
lot
more
confusing,
so
I
don't
know
there.
We
might
need
a
package
concept.
Maybe
I
don't
know.
F
Yeah,
I'm
sure
we'll
know
more
once
like
we
actually
get
into
it,
but
now
it's
like
really
hard
like
good,
to
understand
like
how
this,
because
the
landscape
chain
is
changing,
so
I
guess
like
let's
learn
and
we
might
duplicate
some
work,
but
we
should
definitely
chat
more
like
as
things
make
progress
on
both
sides.
A
Cops
is
doing
that,
but
also
I
have
a
general
comment
that
if
you
want
to
know
what
is
the
state
of
also
cecil
asked
about
this
like
last
time?
A
If
you
want
to
know
what
is
the
state
of
darwin's
project,
the
best
way
is
to
just
join
the
meeting
like
I've
been
guilty
of
not
being
able
to
join
this
meeting
because
I
have
internal
conflicts,
but
I
will
definitely
want
to
follow
this
more
closely
and
that's
basically,
I
know
that
lee,
who
is
one
of
the
active
people
in
this
project,
now
works
for
vmware,
but,
for
instance,
if
vmware
has
interesting
add-ons,
I
don't
think
lee
should
be
the
only
person
that
is
involved.
A
If
microsoft
has
interests
in
addons.
Please
send
your
people.
People
should
join
the
media
and
discuss
these
topics.
Maybe
we
can
help
with
reviewers
from
the
cops
pr.
Basically,
from
my
perspective,
the
whole
effort
is
needs
more
people
to
drive
it.
That's
that's
the
main
problem
here.
D
That's
true,
I
mean
it's
part,
it's
also
my
fault
like
we.
We
have
to
get
the
first
one.
We
have
to
get
the
first
one
in
there
like
it's
been
a
there's
been
a
lot
of
chicken
and
egg,
and
I
we
have
to
break
that
deadlock
in
some
way
and
I'm
trying
to
break
that
deadlock,
and
so,
but
that's
not
to
say
other
people
are
not
very
welcome
to
also
try
to
break
the
lock.
But
it's
not
it's
not
entirely
a
tragedy
of
the
comments.
A
A
I
cannot
attend
the
meeting
and
we
kind
of
blocking
ourselves
as
well,
because
because
maybe
just
if
you
get
more
people
joining
the
meeting,
you
can
have
more
ideas
how
to
fix
their
back
problems
and
things
like
that.
So
yeah.
I
think
people
are
the
problem.
We
just
have
to
invest
more
into
this
effort
and
that's
what
we
have
to
do.
D
Yeah,
I
mean
actually
like
vince.
We
should
we
should
chat
and
see
if
there
is
a
way
to
collaborate
and
like
we
should
talk
about
where
how
chaos
is
going
to
do
it
in
detail
and
whether,
like
you,
can
adopt
the
same
approach
and
whether
you
can't,
even
if
you
choose
not
to
right,
hopefully
we'll
learn
from
each
other,
and
that
would
be
super
valuable
for
me
and
also
fun.