►
From YouTube: SIG Cluster Lifecycle - Cluster API 21-05-19
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Hello,
everyone
and
welcome
to
the
cluster
api
office
hours
meeting
today
is
may
19th.
We
have
busy
agenda
today,
but
before
we
start,
I
see
some
new
names.
Does
anybody
wants
to
say
hi.
B
C
Hi,
my
name
is
merrick,
I'm
not
really
new,
but
I
took
a
long
sabbatical
for
a
little
while
and
anyways
yeah
just
getting
back
into
things.
So.
A
Awesome
welcome
folks
anybody
else
like
to
onion
and
say
hi
before
we
start.
A
All
right
so,
let's
actually
get
started
then
I
just
wanted
to
do
a
quick
status
check
on
the
0.40
milestone,
given
that
next
three
weeks,
probably
like
we
should
start
like
understanding
like
where
we
are
at
for
the
release
and
probably
with
zero
four
zero,
which
is
super
exciting.
So
we
have
nine
open
release
blocking
issues
I
see
most
of
them
have
a
life
cycle.
Active
and
br
is
open
this
one.
I
think
we
don't
fabrici.
Did
you
take
a
look
at
this.
A
Okay,
if
I
remember
correctly,
this
shouldn't
be
a
problem
anymore
for
kcp,
even
though
we're
not
hashing
the
the
fields
anymore,
but
we're
comparing
them
like
those
values
one
by
one.
So
it
should
be
fine
to
fix.
A
A
A
Okay,
work
in
progress.
I
saw
pr
for
this
one
as
well.
It's
it's
not
linked,
but
yeah.
This
is
ready
for
review
hey.
If
the
prs
creates
a
seal,
like
you
know,
should
be
approved
right
away
there.
You
go!
Oh
yes,
so
these
two
actually
like
related,
I
added
more
support
on
this.
I
don't
know
if
marcel's
on
the
call
I
wanted
to
close
4337
in
favor
of
this
one.
A
A
We
don't
have
this
functionality
in
kcb,
so
what
I'm
proposing
on
this
issue
and
I've
been
working
on
it,
but
I've
been
blocked
by
one
other
thing
that
I
wanted
to
call
out
is
that
we
change
instead
of
saying
just
infrastructure
template,
which
is
a
little
bit
very
generic.
It
doesn't
even
say
that
it's
from
machines,
we
changed
this
struct
to
b,
to
have
machine
template
and
then
the
metadata
you
needed,
then
the
infrastructure
which
is
just
reference
as
it
is
today.
A
The
conversion
web
book
will
take
care
of
restoring
this
field
because
it
was
not
going
to
exist
in
alpha
3
and
convert
the
infrastructure
field
and
then
the
object
meta
planning
to
pass
all
those
annotations
labels
both
on
machines,
the
infrastructure
machine
and
the
cubed
m
config.
So
all
the
options
will
get
objects
will
get
those
those
labels
and
annotations
that
you
specified
in
here.
A
Perfect,
I
see
some
plus
ones
in
chat
as
well,
so
I'm
working
on
this
locally
I've
been
100
blocked
on
something
I
want
to
talk
about
is
there
is
so
apparently
conversion
web
book?
We're
not
running
an
m
test
all
this
time
and
found
that,
because
of
the
breaking
change
that
just
introduced
with
that
pr,
and
because
the
validation
web
book
was
like.
A
Oh,
I
you're
not
running
conversion,
so
the
new
field
is
actually
nil
and
I
had
to
track
this
down
up
to
m
test
and
controller
runtime
and
there's
a
bunch
of
fixes
that
needed
to
go
in
there's
one
more
pr
that
I'm
waiting
for
to
get
nurtured
before
cutting
zero.
Nine
zero
beta
five,
which
should
potentially
100,
hopefully
fix
this
issue.
There
were
like
a
bunch
of
things
that
needed
to
be
done
so
being
blocked
on
that
I'll
fix
that
one
first
and
then
open
a
pr
with
the
changes
for
this
one.
E
So
go
ahead.
Sorry,
quick
question:
did
you
say
that
was
the
case
in
controller
runtime
or
in
cluster
api
that
the
conversion
web
hooks
weren't
running
actually.
A
Controller
runtime,
so
if
you're
using
mtest-
and
you
think
you
have
your
conversions
web
book
running
they're,
not.
A
And
if
you'd
like
to
know,
why
is
because
so
when
we
generate
crds,
customize
is
actually
in
charge
of
setting
the
web
book
strategy
in
all
these
fields,
but
usually
with
controller
runtime
in
mtest.
We
just
add
the
folder,
the
basis
folder,
which
does
not
contain
that
conversion.
A
So
how
to
do
a
bunch
of
things
to
to
pretty
much
say.
Okay,
I
have
a
crd
I'll
check
if
the
crt
is
convertible
and
then
enable
the
conversion
web
book,
because
it
controller
runtime,
is
like
just
deals
with
unstructured
data.
This
was
kind
of
challenging,
and
but
we
should
be
almost
there
so
glad
we
found
there
before
zero
four
got
released,
though
that's
one
plus.
A
What
else
do
we
have
here?
I
think
I
think
we
we're
all
done.
They
should
have
two
pr's
open,
so,
okay,
any
other
questions
from
zero
four
zero.
Before
we
move
on
the
next
topics,.
E
This
year,
did
you
already
cover
that
second
one,
the
adapt
cluster
still.
B
A
Okay,
perfect,
it
seems
like
forbidden,
has
more
like
a
bunch
of
things
to
do
here,
so
cool
so
stephanie's,
a
saying.
Are
we
implicit
in
block
until
continuing
time,
release
zero
nine
zero?
Yes,
we
are
dependent
on
that
actually
like
another
little
psa
controller
runtime
right
now.
If
you
try
to
upgrade
the
mtest
environment
on
kubernetes
120,
it's
actually
not
gonna
work,
because
120
has
dropped
the
support
for
insecure
ports
which
support
controller
runtime
was
it
has
been
using.
A
There
is
a
pr
from
solid
that
should
fix
this,
but
it
might
take
a
bit
before
yeah
there's
a
there's,
a
whole
refactor,
so
there's
more
changes
coming
our
way.
So
we
need
to
wait
until
this.
It's
pretty
big
spr
emerges,
so
we
are
blocked,
but
at
the
same
time
like
we
could
probably
understand
if
we
want
to
hold
on
into
this
refactor
and
lose
at
0
10,
and
I
can
chat
with
the
controller
in
time.
Folks,
if
the.
D
Thank
you
so,
quick
psa.
We
have
some
fading
tests.
The
failing
test
is
the
one
that
upgrades
from
the
121
to
v122,
so
the
latest
kubernetes
version
from
ci,
and
there
is
an
issue,
and
this
is
already
assigned
to
some
contributor
to
to
get
fixed.
Basically,
we
need
to
enable
enabling
conversion
to
the
v1
beta3
types
into
recently
introduced
in
kubernetes
v12.
D
I
pause
for
if
there
are
questions,
okay,
let's
move
on
so
in
david
mcclure,
just
got
merged
at
our
pr
that
had
the
epv6
supporting
company
really
happy
and-
and
we
have
already
appeared
in
test
infra
for
basically
creating
a
non-blocking
pr
job
running
a
capability,
ipv6
mode.
E
E
Okay
and
then
I
wanted
to
know
for,
for
providers
like
cabsie
who
already
added
support
for
ipv6
single
stack.
Are
there
any
required
changes
once
we
adopt
this
version
of
cluster
api
or
we
keep?
Is
it
mostly
changes
to
kept
z.
D
E
Right,
okay,
but
it
I
know
at
one
point:
there
was
like
a
discussion
of
whether
it
would
it
should
be
like,
like
auto
detected,
which
ip
family
we
were
using.
That's
is
that
the
way
that
ended
up
being
adopted,
there's
no
action
from
the
user
right.
D
Yeah
exactly
so,
when
you
come
or
your
cup
d
cluster
and
you
define
your
seeds,
then
the
cluster
basically
adapts
to
what
you
define.
There
are
some
discussion
upstream,
and
I
just
want
to
to
to
raise
the
point.
Maybe
a
link,
the
the
cap.
D
There
is
a
cap
on
going
discussing
the
possibility
to
basically
move
away
from
the
current
way
that
that
it
exists
today
for
configuring
seeds
that
is
basically
setting
cluster
apis
flags
and
and
move
to
a
different
way,
which
is
based
on
on
on
a
new
api
types
that
is
being
introduced.
D
The
the
cap
is
still
being
discussed,
but
yeah
there
is
there
is
that
this
could
impact
networking
in
some
future
release
of
kubernetes.
A
Awesome
thanks
folks.
Is
there
any
other
question
for
forbids
on
these
two
topics
before
we
move
on.
D
If
I
may,
I
won't
also
give
kudos
to
matt
that
basically
immigrated
almost
our
test
to
to
go
testing
and
yeah.
Thank
you
very
much
for
this.
A
Awesome
sagar:
let's
get
started
on
this
one.
F
Yep
so
like
yeah,
I
just
wanted
to
yeah
just
introduce
this
proposal
that
we've
talked
about
a
bit
but
yeah.
I
hadn't
opened
the
proposal
document.
Yet
this
proposal
is
about
cluster
class
and
creating
management.
Apologies
for
creating
multiple
clusters,
so
the
the
proposal
is
yeah
ready
to
review.
F
I
I
saw
like
a
couple
of
folks
that
did
go
in
and
like
reviewed
the
proposal
but
yeah
I
I
urge
everybody
to
like
go
ahead
and
take
a
look
at
the
proposal
and
leave
some
questions
or
review
comments
as
they
seem
fit.
F
Maybe,
like
I
can
take
a
minute
to
talk
about
what
cluster
class
is
so
cluster
class
is
basically
a
a
way
of
like
or
is
like
a
collection
of
templates
which
can
be
used
to
create
multiple
cluster
objects
easily
without
having
to
create
all
of
the
individual
objects
that
are
required
today
before
creating
a
cluster.
So
this
is
just
a
way
to
like
make
the
experience
of
creating
a
cluster
much
easier
for
end
users.
F
Yeah.
I
guess
that's
that's
all
I
would
say
about
this
for
for
more
information.
I
guess,
like
we,
I've
added
most
of
the
details
here
in
the
proposal,
so
please
go
ahead,
take
a
look
at
it
and
let
me
know
if
you
have
any
questions
online
or
offline
yep,
that's
about
it.
A
Awesome
thanks
agar.
Are
there
any
questions
about
cluster
class
or
comments
feedback.
D
Yeah,
I
I
want
to
thank
sagar
and
also
echo
his
request
to
to
take
a
look,
because
maybe
if
you
show
the
example
of
a
cluster
class
of
or
manage
class
this,
I
think
that
this
will
be
a
huge
step
forward
for
for
the
user
cluster
api.
So
please
take
a
look.
A
Yeah,
why
don't
we
show
like
what
a
cluster
would
look
like
here?
So
the
idea
is
to
make
the
cluster
object
a
little
bit
more
useful
in
this
cycle,
so
we're
proposing
to
have
manage
section
of
the
cluster
where
you
could
define
everything
about
your
cluster
and
then
like
over
time,
would
manage
like
the
life
cycle.
All
these.
A
These
objects
are
fields,
so
in
this
case,
like
the
cluster
itself
would
be
defined
in
a
class
which
you
could
see
here
as
a
collection
of
references
which,
probably
all
are
familiar
with
like
we
that
we
defined
today
in
a
bunch
of
places,
and
then
we
would
have.
Finally
the
version.
The
kubernetes
version
that
you're
using
as
top
level
so
that
you
could
do.
We
could
do
upgrade
in
a
managed
way
so
like,
for
example,
upgrade
control
plane.
A
First,
before
you
upgrade
worker
nodes,
and
then
you
could
define
your
control
plane
in
here
so
like
set
the
replicas
labels
and
annotations
on
your
control
plane
and
then
just
bunch
of
worker
nodes.
It's
like
you
have
a
class
that
you
that
you
can
define
here
as
a
collection
of
templates.
A
A
There
is
a
bunch
of
future
work
that
that's
planned,
so
one
of
which
so
like
for
the
first
iteration,
probably
gonna,
make
everything
a
little
bit
more
immutable
so
that
when
we
want
to
allow
certain
things
to
be
immutable,
so,
for
example,
can
I
change
a
cluster
class,
go
from
class
a
to
class
b,
almost
as
a
rebase?
So
how
do
we
do
that?
How
do
we
allow
that,
but
also
like
limit
what
you
could
change
in
which
class?
How
do
we
define
compatibility
with
between
two
classes?
A
Then?
How
do
we
extend
the
cluster
class
so
that,
like
you,
could
allow
the
customization
of
certain
fields
in
there
with
variables
and
things
like
that?
How
do
we
adopt
current
clusters
and
how
do
we
observe
clusters
and
also
like
how
do
we
integrate
with
current
machine
health
checks-
and
I
think
mike
also
mentioned
the
cluster
autoscaler?
A
G
Go
ahead
thanks,
yeah,
I
was
wondering
a
lot
of
this
proposal
is,
is
actually
modifying
the
cluster
object,
so
I'm
wondering
what
are
what
changes
are?
So
if
I,
if
I,
if
I
use
the
cluster,
do
I
have
to
adopt,
you
know
these
new
fields
and
the
other
question
that
I
had
was
what
what
is
a
managed
topology.
G
So
if
I,
for
example,
if
I
use
kappa
with
my
own,
let's
say
infrastructure,
I
create
my
own
vpc
and
subnets
and
so
forth.
How
does
that
relate
to
the
like
the
terminology
here?
The
manage
managed
topology.
A
So
the
first
question
is
like:
do
you
have
to
use
this
field?
No,
like
you
can
still
use
all
the
things
you're
using
currently
right
and
second,
you
wanna,
take
the
other
one.
F
So
the
the
other
one
so,
like
your
question
was
around:
how
would
you
be
able
to
like
how
would
you
use
your
own
vpcs
or
your
own
properties
with
with
this
managed
topology?
If
I
understand
the
question
correctly,
yeah,
that's
right,
yeah,
so
like
basically,
if
so
currently,
the
cluster
class
is
like
a
collection
of
templates.
F
So
if
you
plug
in
your
own
fields
into
the
templates
that
are
defined
in
the
cluster
class
themselves,
then
you
could
use
that
cluster
class
to
create
the
clusters
and
thereby
you
can
use
the
the
fields
that
you
want
inside
of
a
managed
topology
and
the
term
managed
topology.
What
it
entails
is
that
the
cluster
controller
will
take
care
of
updating
your
control,
plane,
updating
your
worker
nodes
and
making
sure
that
they
are
up
to
date
rather
than
you
having
to
create
these
custom
objects
on
your
own.
G
A
Awesome
is
there
any
other
questions.
Comments
concerns
about
posture
class
before
we
move
on.
A
All
right,
please,
take
a
look
and
comment:
let's,
let's
try
to
give
us
like
at
least
10
a
week
or
10
days
to
to
get
through
this
document,
resolve
comments
et
cetera
and
then
try
to
move
it
to
a
pr
so
that
yeah
we
can
get
started
to
get
this
merged
and
implemented
spencer.
G
Hey
guys
yeah,
so
I
wanted
to
talk
a
little
bit
events.
G
We
talked
a
little
bit
yesterday
on
slack
about
this
q
bar
back
proxy
stuff,
but
I'm
not
sure
how
much
it's
going
to
matter
depending
on
how
close
we
are
on
0.4
of
cluster
api,
but
we've
got
a
bunch
of
people
that
are
wanting
to
run
sedero
on
rm64
nodes,
and
so
we've
been
kind
of
doing
the
work
to
uplift
that
and
get
multi-arch
images,
and
all
that
good
stuff
done
kind
of
the
last
hiccup
I
think
we're
running
into
is
is
really
just
with
q
bar
back
proxy
in
in
the
core
components
right
it's
using
0.4.1,
which
is
super
old
and
not
multi-arch,
and
so
anyway,
I
don't
know
what
the
right
answer
is
here
other
than
like.
G
I
think
it
looks
like
it
does
in
the
master
branch
anyway.
So
anyway,
it's
more
just
a
discussion
point
than
anything
I
guess,
but
some
answer
for
arm
64
for
this
would
be.
A
A
A
A
Okay,
I
guess
I'll
ask
the
other
maintainers
cecile
for
free
to,
like
you
are
supposed
to
upgrade.
G
Keyboard-
I
didn't-
I
didn't
dive
that
far
into
this.
I
one
of
our
other
guys
was
was
the
one
checking
this
out,
but
I
mean
it
looked
like
a
virgin
bump
worked
just
fine
like
it
didn't
seem
to
really
break
anything,
but
there
may
very
well
be
things
that
I
don't
know
that
you
guys
may
be
able
to
spot
totally
right.
E
But
I
think
if
we
like,
I
would
start
opening
a
pr
and
seeing
exactly
how
much
changes
are
needed
if
it
is
just
a
version
buff
and
we
have
forward
testing
that
to
make
sure
that's
backwards
compatible,
then
I
think
it's
fine,
that's
my
view
on
it.
Okay,.
G
Yeah
I
I
can
so
yeah
I
mean
it's
like
I
said
it's
already
fixed
and
master,
so
I'll
I'll.
Just
try
to
do
a
cherry
pick
to
release
cr3
and
kind
of
see
what
that
looks
like.
D
As
far
as
I
know,
it
is
just
a
version
bomb,
at
least
in
master.
We
don't
add,
we
don't
have
any
problem.
I
I'm
just
wondering-
and
this
is
a
question
for
spencer.
D
G
Have
yeah
we've
been
telling
people
just
a
straight
use,
cluster
ctl
and
the
upstream
core
stuff
right,
like
the
latest
release,
so
yeah
I
mean
we
can
certainly
tell
people
to
patch
it.
It's
just.
You
know
not
great.
D
No,
no,
I
I
got
it
if
you
are
using
upstream
stuff.
This
is
annoying
for
the
people.
G
Yeah,
okay
yeah,
like
I
said
I'll
I'll,
pull
that
change
into
and
release
year.
Three
we'll
see
what
it
looks
like
see
how
much
stuff
falls
over
and
then,
if
we
want
to
punt
it
till
zero.
Four,
that's
fine.
H
Macon
yeah
thanks,
I
just
bringing
up
again
this
issue
I
raised
came
a
few
weeks
ago
is
cube
our
backpack
proxy
required
and
do
we
need
to
keep
including
it.
H
Yeah,
I
I
think
just
for,
for,
like
a
provider
distributing
cap,
cappy
like
it'd,
be
great
to
have
one
less
thing
to
distribute
patch
build.
So
I
I
I'm
yeah,
I'm
just
wondering
if
this
is
something
we
can
even
rip
out.
A
So
the
last
comment
was
from
right
here:
it's
like
that.
It's
only
used
for
metrics.
I
don't
necessarily
want
to
remove
and
then
we
can
use
metrics.
But
I
want
to
make
sure
that
not
that
we're
using
metrics
today,
because
we're
actually
really
bad
at
it,
but
but
yeah
so
like
if,
if
you're,
okay,
to
like
just
open
up
vr
and
like
to
see
you
know
by
removing
it
what
happens,
that
would
be
great
because
we
have
extensive
endurance
so
as
well.
H
Right
yeah
it
just
from
I
mean
for
my
understanding.
It
just
is
doing
some
our
back
guarding
around
who
can
access
metrics
but
you'd
still
be
able
to
access
metrics
without
it.
It
just
would
wouldn't
necessarily
be
doing
an
arbac
check.
Okay,
I'll
I'll,
open
a
pr
with
it
ripped
out
and
see
what
happens.
I
E
Yeah,
I'm
just
gonna
say
regardless.
I
don't
think
we
should
remove
it
in
0.3.
If
anything,
we
should
remove
it
in
the
main
branch
which
doesn't
solve
spencer's
issue
for
now,
but
I
think
we
should
have
both
in
parallel.
A
That
cool,
it
seems
the
funnest
is
saying
each
other,
like
captain
dentist,
actually
patched
the
manifest
to
get
the
metrics
so
yeah.
C
A
A
Cool
fabrizio,
you
have
the
next
two
items.
D
Requests
for
basically
review
the
work
for
upgrading
thruster
cattle
and
make
it
support.
The
the
new
model
of
multi-tenancies
do
remove,
does
mean
basically
move
credential
that
could
be
global
credential
with
a
secret
that
is
hosted
in
the
provider.
Namespace.
The
the
work
is
ongoing
and
what
is
really
important
is
that
the
provider
implements
take
a
look
at
listed
at
the
change
of
of
the
control
for
cluster
cattle
move.
D
So
there
should
be.
Let
me
say
what
is
working
today
will
continue
to
continue
to
work.
We
are
only
supporting
new
additional
features,
but
it
is
a
change
of
contract,
so
I
would
like
people
to
if,
if
not
having
a
look
at
the
pr,
at
least
having
a
look
at
the
at
the
changes
in
in
the
markdown
document
that
defined
the
documents,
the.
D
Okay,
so
if
there
are
no
questions
about
this,
I
want
to
report
back
basically.
Last
week
I
discussed
about
all
the
machine
integration
in
cubic
mark
and
after
I
I
went
to
seek
discoverability
to
present
the
project,
the
the
prototype,
a
demo
prototype-
and
maybe
I
could
demo
this
year
next
next
week,
but
yeah
the
the
feedback
were
really
positive
and
basically
sig.
Scalability
is
really
interesting
to
the
to
prototype,
together
with
the
cluster
api
team.
D
A
first
stress
test
job
using
cluster
api,
and
this
is
a
huge
opportunity
for
this
project,
because
basically
this
allows
us
to
get
feedback
from
scalability.
They
will
use
cluster
api.
I
guess
at
scale
that
that
we
never
used
before
and
also
this
will
allow
cluster
api
to
become
more
relevant
for
the
kubernetes
test
senior
release
senior.
A
A
Thanks
for
praying
too
and
really
awesome
work
thanks
so
much
for
pushing
this
through
with
the
other
things
as
well.
I
Hey
yeah,
sorry,
for
the
day,
I
just
wanted
to
say
that,
based
on
the
last
meeting,
I
like
I've
addressed
some
of
the
like
I've
addressed
the
comments
so
far
in
that
talk
and
made
some
changes
in
the
proposal.
I
still
have
to
add,
like
I'm
working
on
adding
some
architecture
diagrams
so
I'll.
Do
that
and
just
wanted
to
say
that
if
you
get
time,
please
give
it
another
review
that
will
be
really
helpful.
A
Thanks,
thank
you.
Let's,
let's
definitely
take
a
look
and
I
think
soon
we
might
be
able
to
maybe
open
up
a
pr
on
this.
If
there
are
like
no
more
outstanding
comments.