►
From YouTube: Kubernetes SIG Cluster Lifecycle 20190109 - Cluster API
Description
Meeting Notes: https://docs.google.com/document/d/1Ys-DOR5UsgbMEeciuG0HOgDQc8kZsaWIWJeKJ1-UfbY/edit#heading=h.72kntbx8l6rh
A
All
right,
hello
and
welcome
to
the
Wednesday
January
ninth
edition
of
the
cluster
API
sub-project
meeting
verse
a
clustered
lifecycle,
as
you
might
notice,
today's
recording
was
a
little
bit
different
assume
appears
to
be
having
an
outage
and
so
we're
using
Google
video
chat.
Instead,
hopefully
those
records
and
uploads
normally
to
YouTube,
and
if
you're
watching
this,
then
I
guess
it
worked
all
right.
So
we'll
go
ahead
and
dive
into
the
agenda
timi
about
the
first
thing,
which
is
a
vote
on
moving
the
azure
provider
from
platform
9
over
into
kubernetes
sinks.
C
B
B
A
C
A
A
B
A
I
didn't
it
was
certainly
none
on
the
the
email
thread
that
I
saw
I
think
the
only
reason
we
hadn't
adopted
it
sooner
was
sufficient
sort
of
committers
from
enough
diverse
companies
and
I
think
that
they
they
solved
that
problem.
So
that
was
really
the
big
hold
up
the
first
time
that
they
asked
to
be
adopted.
D
Yes,
right
now,
Q
builder
depends
on
kubernetes
112
to
be
able
to
unlock
the
conversion
webhooks
in
queue
builder.
We
need
to
get
that
wrapped
up
to
communities
113,
so
I've
got
an
active
PR
out
there
for
controller
run
time
right
now
and
I
have
local
change,
sets
for
controller
tools
and
queue
builder,
but
because
of
the
way
the
dependency
chain
works,
I
kind
of
have
to
stay
tuned,
one
at
a
time.
D
A
I
tried
it
to
me,
but
when
I'm
not
talking
and
sometimes
forget
time
yet
I
was
saying
that
it
looks
like
you
assigned
the
the
pr2
tasali
and
he
had
a
question
for
you
and
I.
Think
upgrading
to
1:13
to
get
this
functionality
sounds
great,
assuming
we
don't
break
things
along
the
way.
So
that's
that's
the
main
concern
right.
D
Yeah
yeah
I've
been
able
to
run
everything
up
through
the
Q
Builder
tests.
I
ran
into
a
fun
little
issue
trying
to
get
those
to
pass
because
of
the
K
log
migration
and
it
turned
out.
It
was
kind
of
dependency.
The
vendor
dependency
fund,
who
think
you
boulder
itself,
but
as
far
as
I,
can
tell
there's
no
breaking
changes
that
are
part
of
it.
So
far,.
F
C
D
Yeah
there
was,
they
have
a
document
on
versioning
and
there's.
You
know
they
want
to
maintain
semantic
versioning
much
better
than
kind
of
the
core
kubernetes
objects
that
they
consume,
but
it's
unclear
to
me
because
they're
still
technically
major
version
0,
whether
or
not
that
would
even
really
apply
at
this
point
or
not
so
the
only
change
I
came
in
was
an
additive
change
to
the
rest,
client,
so
I,
don't
think
that
would
have
an
effect
either
way,
but
it'll
be
interesting
to
see
what
happens
going
forward
with
that.
A
Cool
well
definitely
thank
you
for
the
work
getting
that
updated
him
looking
forward
to
be
able
to
use
all
the
fun
new
features.
Previous
here,
DS
I
saw
an
email
thread
the
other
day
that
things
like
like
swagger
definitions
for
CR
DS
or
coming
in
some
from
some
future
release.
Also
so
I
think
keeping
up
with
new
releases
is
great,
because
CRTs
are
definitely
gaining
features
quite
quickly.
E
E
A
E
So
like
right
now,
like
a,
we
kind
of
like
assumed
that,
like
a
closet
and
machines
in
a
single
namespace,
are
all
related
to
each
other
was
one
in
like
you
know
like
do
we
want
to
go
in
this
direction?
Do
we
want
to
have
like
a
high
relationship
like
101
between
maybe
spaces
and
close
through
end
machines,
or
do
we
want
to
allow
multiple
cluster
of
machines
to
live
in
a
single
namespace.
G
E
Yeah
there's
the
current
assumption.
I
am
I'm
curious
to
know
like
if
the
current
the
group
here
like
has
wants
to
make
this
kind
of
official
or
because,
like
right
now,
it's
more
like
an
assumption
than
actually
like
a
role,
and
so
because,
like
we
have
upcoming
work
in
cos
decade
of
this
around,
this
will
be
great
to
make
it
somewhat
of
visual
I.
A
Think
that
I
think
I
saw
her
DAC
on
the
line.
I
think
hardik
has
also
mentioned
in
the
past
that
in
their
experience
developing
gardener,
they
found
it
really
useful
to
have
multiple
clusters
in
the
same
namespace
and
in
particular,
because
when
you're
referencing
secrets,
committees,
resources
can
only
reference
secrets
and
within
the
same
namespace.
And
so,
if
you
had
say
multiple
clusters
that
we're
using
the
same,
you
know
project
or
account
or
whatever,
at
a
cloud
provider.
You
don't
have
to
duplicate
the
credentials
to
access
that
account
into
multiple
secrets.
C
B
F
F
Should
talk
to
like
I
know,
they're
people
that
do
API
reviews
I,
don't
know
that
we
can
get
their
input
in
general
on
this
issue
and
then
just
like
Cass,
Cass
and
I
over
like
our
whole
API
and
maybe
help
us
on
like
some
of
the
spec
forces
stages,
questions
which
I
know
we've
sort
of
rattled
on
in
the
past.
I
think.
B
B
F
D
So
I
just
wanted
to
add
that
one
of
the
challenges
that
we're
gonna
face
if
we
sit
there
and
try
to
eliminate
the
cluster
per
namespace
and
that
being
the
link
and
the
namespace
being
the
link
between
machines
and
clusters
is
that
there
are
users
that
only
want
to
make
use
of
the
machines
API
independent
of
the
cluster
object.
So
if
we
do
create
a
link
between
the
two,
we
need
to
ensure
that
that
link
stays
optional
and
doesn't
become
a
mandatory
link.
At
that
point,.
A
Right,
I
think
there
are
a
couple
of
ways
we
could
do
that
one
we
could
put
an
explicit
link
between
the
two,
but
another
is
we
could
do
something
like
label
selectors
right
to
match,
which
things
go
together.
Right,
I
think
that's
a
pretty
standard
way.
The
things
are
collected
in
groups
in
kubernetes.
It
would
be
a
little
bit
more
flexible
and
it
would
probably
also
be
quite
prone
to
error,
because
you
could
pretty
easily
create
machines
in
a
namespace
that
didn't
match
the
label
selector
for
a
cluster
and
then
are
they
orphaned.
G
The
other
thing
that
came
up
during
that
discussion
was
cluster
man
using
our
backs.
So
you
can
say
this
group,
this
user
group
can
operate
in
this
namespace
in
this
particular
namespace
and
as
a
result,
they
can
be
prints
for
this
cluster.
Only
so
that
that
was
another
thing
that
that
was
discussed.
I'm
just
collecting
everything.
A
That's
in
the
mention
that
there
are
really
other
examples
in
Corelli's
of
sort
of
Singleton's
and
you're,
saying
basically
like
the
enforcement
of
you
can
only
have
one
cluster
in
a
namespace.
If
you
try
to
create
more,
we
will
reject
it.
There
aren't
really
other
places
in
like
the
kubernetes
resource
model,
where
we
see
that
pattern
used.
A
F
F
We
just
wouldn't
use
it
to
describe
all
like
the
complete
configuration
of
the
cluster,
so
I
think
Rob.
You
suggested
a
labels,
a
label
selector
and
a
cluster
which
selected
a
group
of
machine
deployments.
No
problem
like
I,
don't
think
that's
a
big
deal
for
anyone,
so
I
think
I
think
that
I
don't
think
that
would
be
a
blocker
I
think
that
the
people
that
don't
necessary
want
they
wants
to
start
with
just
a
something.
F
A
That's
true,
I
think
the
other
use
case,
they're
sort
of
two
use
cases
there
that
Jason
was
alluding
to
one
is
to
adjust
adopting
machines
per
the
API,
but
the
other
is
something
talk
about
putting
the
machines
and
the
cluster
resources
into
different
places
and
having
a
link
between
resources
that
are
not
within
the
same
sort
of
career.
Nettie's
resource
universe
doesn't
really
work
right.
A
E
A
Right
I
think,
right
now
we
just
assumed
same
namespace
same
cluster
and
so
I
think
as
a
provider
wanted
to
make
that
a
little
more
nuanced
and
say
you
can
put
multiple
clusters
in
a
namespace.
They
would
also
have
to
have
a
way
of
linking
them,
and
if
we
decide
the
way
of
linking
them
is
labels
provider
would
implement
that
trip
at
the
same
time
right
and
verify
that
sort
of
backwards.
F
F
A
So
Vince,
why
don't
you
at
least
start
the
issue
with
that
first
question
and
we
can
say
like
like
link
back
to
the
old
issue
and
say
I
think
here
are
some
reasons
why
we
might
want
to
reverse
that
decision
at
this
point
sounds
good.
Thank
you
all
right
not
to
take
the
spotlight
off.
You've
got
the
next
one
also,
which
is
requirements
to
become
an
approver
I.
Think
we
briefly
touched
on
this
before
the
end
of
the
year.
A
A
What
sort
of
criteria
do
people
sort
of
want
to
see
in
those
and
those
people
that
are
getting
promoted
so
that
we
all
agree,
hopefully
mostly
objectively,
that
that
is
it's
the
right
decision
to
promote
and
we
don't
become
sort
of
biased
and
promote
either
people
like
ourselves,
people
from
the
same
company
etc,
and
we
have
sort
of
the
the
most
unbiased
and
objective
system
in
place
that
we
can
get
to
sort
of
create
new
approvers
cuz.
We
certainly
don't
want
to
get
in
the
case
where
it's
impossible
to
have
a
new
approver.
A
It's
so
hard,
there's
so
many
barriers
to
reach
that
we
don't
get
anymore,
because
eventually
we
will
have
none,
but
we
also
don't
want
to
make
it
too
easy
right,
because
we
want
to
have
sort
of
the
feeling
among
that
people
in
the
community
that
the
the
approvers
sort
of
have
the
best
interest
of
everybody
and
are
sort
of
making
the
right.
This.
B
I
can
tell
you
what
we
do
for
kuba
DM
and
it's
worked
out
pretty
well.
We
basically
submit
a
PR
with
a
very
long
time
out
and
see
see
all
the
people
who
are
going
to
be
out
of
rotation
as
well
as
ad.
The
people
who
we
want
to
add
in
the
rotation
give
time
for
folks
to
give
a
time
out
of
like
a
couple
of
weeks
for
folks
to
chime
in
if
they
want
to,
but
we
do
often
rotate
people
out,
because
what
will
happen
is
we
don't
do?
B
This
is
look
at
assigned
issues
and
they're?
The
PRS
will
then
just
lag
in
perpetuity
forever.
Just
because
that's
how
the
way
the
bot
works
and
there's
no
reason,
you
know
we,
we
often
tell
them
like
we'd,
happy
to
have
you
back
if
you're,
if
you
have
time
to
work
on
these
things,
but
the
rotation
policy
that
we
followed
has
worked
fairly
well
and
we've
done
a
fair
amount
of
rotations
now
across
several
release
cycles.
B
A
B
It's
yeah,
it's
typically
self-evident
right,
like
it's.
A
self
selection,
because
they're,
actively
involved
and
engaged
in
the
project
approvers
are
typically
a
higher
bar.
You
know.
Adding
people
to
your
list
is
totally
legitimate
if
there
reasonably
active,
but
if
they
are,
if
they
have
a
stake
in
their
full
time,
devoted
towards
this
and
are
clearly
eating
aged,
it
becomes
self-evident
if
they
should
beer
to
prove
or
none
like
when
we
added
little
mirrors,
no
question
everyone.
Everyone
who
is
already
an
approver
said:
yes,
please
get
a
little
weird
to
be
an
approver.
A
Okay,
so
maybe
we
should
just
do
that
with
Vince's
PR
and
instead
of
having
it
it's
not
linked
here,
but
I.
Remember
seeing
at
some
point
instead
of
having
it
just
add
him,
and
we
can
update
that
to
also
kick
some
people
out
see
see
all
those
people
specifically
link
it
here
and
let
people
know
that
if
they
feel
like
they
should
also
be
added
as
a
reviewer
or
approver.
They
should
sort
of
piggyback
on
that
same
change
and
we'll
set
a
timeout
for
a
couple
weeks.
How's
that
sound
I.
F
E
F
Reviews
I,
don't
know
how
we
would
do
that
with
approvers,
but
I
think
we
can.
Maybe
we
start
with
like
scripts
that
looks
at
like
who
has
done
the
most
reviews
or
the
most
comments
or
whatever
it
is
like
any
way.
Anyone
can
write
the
no
GTM
comments,
even
if
it
doesn't
mean
anything,
and
so
we
can
look
at
the
people
who
actually
do
that.
There's.
B
Also,
we
also
do
grooming
in
Covidien
so
like
if
we,
if
somebody
before
grooming
someone
to
become
an
approver,
it's
pretty
obvious,
like
they're
involved
in
the
PRS
and
they're
dovetailing
with
other
approvers,
so
they're
not
acting
autonomously.
We're
not
trying
to
just
do
it
independently.
These
people
are
actively
engaged
in
working
with
the
other
approvers
to
make
sure
that
they're
they're
doing
what's
in
the
best
interest,
and
they
would
recognize
the
minutia
that
can
exist
between
some
of
the
issues.
A
F
A
A
Tim
was
saying
sat.1
PR
with
a
long
time
out
and
make
all
the
changes
in
bulk,
rather
than
have
like
a
long
series
of
oh
this
person.
Now
this
person,
this
person
now
we're
the
removal
and
sort
of
just
take
it
all
in
one
shot.
I
know
that
sets
the
clock
out
a
little
ways
for
Vince,
which
is
a
little
unfortunate,
but
I
think
it's
probably
the
right
way
to
do
it,
and
you
know
Vince
when,
when
you
go
through
all
these
eum's
pr's,
it
makes
it
much
quicker
for
me
to
approve
them
anyway.
A
What
we're
talking
about
now
is
like
the
owners
files,
which
is
sort
of
a
kubernetes
mechanism
with
in
a
github
repo
to
allow
people
that
don't
have
github
ackles
to
still
be
able
to
use
the
box
to
allow
automation
to
make
changes
to
the
route
to
the
code.
Okay,
so
it
might
be
useful
in
the
sense
that
if
we
could
get
the
automation
to
also
synchronize
like
the
github
groups,
with
what's
in
our
owners
files,
then
useful.
But
if
it's
just
synchronizing
github
groups
with
like
a
yeah
no
file
is
checked
in
somewhere.
A
Then
it's
sort
of
at
a
different
level
than
what
we're
talking
about
here.
It
does
overlap
with
a
section
of
the
owners
files.
It's
like
the
the
alias
for
like
this.
A
cluster
lifecycle
leads
or
the
cluster
API
admins.
Those
are
both
should
be
backed
by
github
groups,
but
the
maintainer
x',
which
is
the
the
bottom
lists,
which
is
the
people
that
have
approval.
That
would
be
separate,
at
least
today,
that's
separate
from
a
github
group.
Okay,.
A
So,
like
recently,
like
Tim
Sinclair
was
added
to
the
github
back
hold,
so
you
could
tinker
with
all
the
UI.
We
did
think
it
hub
like
adding
labels
and
milestones
and
shuffling
issues
around,
but
that's
separate
from
being
lists.
That
is
a
maintainer
or
an
admin
in
the
lots
and
lots
of
different
Eckles.
B
So
we
know
VMware
and
previously,
as
FTO
we
we
are
been
waiting
patiently
for
an
actual
final
cut
of
v1
alpha
1,
and
not
only
that,
but
we
would
like
to
actually
help
drive
forwards,
any
things
that
we
want
to
do
to
get
that
hard
milestone
into
place.
So
I'm
volunteering
myself
to
stand
up
to
help,
do
sort
of
the
scrubbing
of
the
backlog
to
get
the
milestones
in
order.
B
So
that
way
we
can
start
just
like
legitimately
executing
to
try
and
get
a
cut
of
a
release
within
a
reasonable
time
frame,
maybe
a
couple
of
months.
The
reason
being
is
because
we
want
to
cut
a
release
of
Questor
API
AWS
provider,
but
we're
kind
of
explicitly
blocking
that
release
on
the
formal
cut
of
v1
alpha
1,
a
cluster
API.
You
know
it's.
We
want
to
make
sure
that
we're
doing
the
best
thing
by
the
PTO
I
want
to
do
that
in
absentia.
B
So
I
want
to
talk
with
you
Robbie,
as
well
as
on
the
sinkhole,
to
try
and
help
groom
the
backlog
and
try
to
get
into
working
order.
So
we
can
just
go
and
I'm,
not
certain
I.
Couldn't
do
it
purely
by
myself
right
now,
because
there
are
some
issues
that
have
just
been
there:
city
block
that
I
don't
even
know
if
they
even
apply
anymore.
So
yeah
I've
tried.
A
So
one
thing
that
we
talked
about
around
cube
Khan
was
once
we
had
gotten
to
the
point
where
CRD
is
supported.
Conversion
then
like
we
might
as
well
cut
it
release,
because
we
can
can
kind
of
move
people
forward
great
and
that's
I
think
what
we
did
with
like
with
cube
ADM
is
we
could
cut,
do
an
awful
one
and
do
an
awful
two,
because
we
have
the
ability
to
still
make
changes
but
not
break
people's.
You
migrated
them.
A
I,
don't
remember
exactly
when
that's
coming
to
see
our
DS,
but
it
may
not
be
coming
soon
enough
right
to
meet
your
we'd
like
this.
The
next
couple
of
months
deadline
so
I
think
that
implies.
We
really
want
to
take
a
scrub
over
the
API
surface
and
make
sure
that
we're
we're
really
happy
with
it.
If
you
aren't
going
to
be
able
to
easily
move
fields
around
for
a
little
while
so.
D
The
Canadian
Ischl
conversion
webhooks
landed
as
a
feature
that
you
can
enable
in
113,
so
it's
not
GA
yet
I
think
it's
beta.
We
can
enable
it
in
113
to
hold
up
for
us
right
now.
Is
the
the
webhook
tooling
and
queue
builder?
Isn't
there
yet
for
the
conversion
with
hooks
and
that's
kind
of
what's
driving
the
work
that
I've
been
doing
to
start
updating
the
dependency
cerdo
113.
A
So
now
that's
a
little
bit
closer
than
I
was
thinking.
That
would
be
great
I.
Think
if
we
were
at
the
point
where
we
had
some
tooling
in
place,
or
we
could
say
we
can
move
you
forward.
Then
it's
gonna
be
a
lot
easier
for
us
to
say
here's
a
first
alpha.
We
can
at
least
agree
that
it's
a
good
enough
starting
point
and
we
can
get
you
off
of
it.
A
B
One
thing
too
is
that:
does
it
need
to
be
an
alpha
one
for
the
conversion,
or
does
it
need
to
be
in
an
alpha
right
like
if
we,
if
we
cut
the
alpha,
we
just
need
the
conversion
upgrade
as
you
never
go
back
like
you
traditionally
in
kubernetes
land,
you
never
go
back
and
in
fact,
I
don't
even
know
of
precedent
to
actually
verify
that
object.
Conversion
backwards
actually
works.
So
the
we
would
just
you
know
as
long
as
we
had
the
alpha,
which
could
be
several
months
from
now.
Have
the
conversion
Mew
be
fine.
B
I,
do
you
think
there's
like
a
limited
list
of
how
we
want
to
do
this?
Also,
just
in
general,
like
cluster
API
itself
is
slightly
different
from
the
grooming
for
most
other
things
in
cyclists
or
life
cycle.
Now,
we've
we've
kind
of
spread
the
cuvette
diem
goodness
across
multiple
repos,
and
it
just
makes
it
easier
for
new
folks
who
want
to
get
involved
to
be
to
follow
the
same
standard
pattern
which
I
was
supposed
to
write
up
for
some
of
the
other
folks
who
wanted
to
follow
that
pattern
as
well.
A
No
be
great
also
once
write
that
up.
We
should
talk
about
that
during
the
sake
kneading
and
see
if
we
can
get
no
cops
and
mini
cube
and
everybody
else
following
sort
of
a
similar
pattern.
You
know
if
we're
gonna
apply
it
across.
The
repos
would
be
great
if
the
other
sub
projects
that
are
nominally
fall
under
the
sig
can
slowly
start
to
adopt
those
as
well.
A
Excellent
model
I,
missed
I,
think
that
was
the
end
of
today's
agenda.
I
missed
last
week,
I'm
sorry
I
was
not
feeling
super
great
I
did
make
a
new
meetings
note
stock
for
the
year.
Hopefully,
people
realize
that
they're
looking
at
the
right
doc
and
there's
a
link
at
the
bottom
to
last
year's
notes.
A
They're
still
there
they're
sort
of
an
archive
state
and
I
think
it
was
the
only
other
big
thing:
I
want
to
mention:
I've
been
trying
to
go
through
the
PR
backlog,
especially
the
ones
that
have
been
LG
TM
and
get
things
merged.
Now
that
I'm
sort
of
back
and
checking
email
again,
and
hopefully
we
can
start
to
pick
up
some
speed
now
that
we've
hit
the
new
year,
I
guess
one
other
reminder
for
folks:
the
CFP
deadline
for
the
coop
con
tu
is
coming
up
in
a
couple
of
weeks.
A
A
Alright
well,
hopefully,
my
recording
of
this
meeting
will
be
somewhere
that
I
can
actually
upload
it
to
YouTube
and
we'll
get
it
get
it
linked.
Otherwise,
we
will
see
folks
I
think
next
week
we
have
a
cig
meeting
on
Tuesday
and
then
another
one
of
these
meetings.
Next
Wednesday,
we
will
see
you
then
Thanks
take
care.