►
From YouTube: Kubernetes SIG API Machinery 20190116
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
Okay,
welcome
everybody
to
the
first
2019
sig
API
Machinery,
meeting
Daniel
Smith
is
out
so
I'll
facilitate
here.
Really
quick.
We've
got.
We
got
four
agenda
items
we
had
none
yesterday,
so
these
have
all
been
added
quite
recently,
I
rewarder
it
a
couple
of
them
since
I
wasn't
sure
how
long
the
design
discussion
would
take.
Well,
let's
just
go
ahead
and
get
started.
Jordan
are
you?
Are
you
here
yeah.
B
So
initializers
were
introduced
in
1:7
as
an
alpha
feature.
They
have
not
progressed
since
then.
There
were
quite
a
few
known
issues
that
needed
to
be
resolved
when
they
were
introduced
and
really
very
little
traction
has
been
made
in
those
and,
in
the
meantime,
the
webhook
Admission
extension
points
have,
in
most
cases
removed
the
need
for
these.
B
B
Cog
controller
manager
has
an
alpha
controller
that
was
built
on
top
of
this
function
and
so
to
use
it,
you
would
have
needed
to
enable
the
Alpha
feature
and
set
up
an
initializer
configuration
and
so
removing
this
removes
the
ability
of
that
controller
to
function.
I
think
Andrew
from
a
cloud
provider
was
on
my
call.
I,
don't
know
if
you
want
to
speak
to
sort
of
where
that
controller
is
that
and
what
what
this,
what
effect
this
would
have
yeah.
C
B
C
So
I
noticed
your
PR
removed
the
B
controller
and
the
PVL
controller
in
the
Cloud
Controller
manager,
which
I'm
not
sure
it's
going
to
be,
is
going
to
work
for
all
the
out
providers
right
now.
So
one
of
the
option
is
like
we
can
leave
the
controller,
but
just
have
it
not
check
initializers,
but
something
else,
but
I'm
not
sure
like
how
much
work.
That's
going
to
be
I
have
to
talk
to
the
maintainer
of
those
provider
find
out.
Okay,.
B
B
But
simply
empty
it.
So
today,
if
you
submit
an
object
with
this
field
populated-
and
you
don't
have-
the
Alpha
feature
enabled
it
gets
cleared
and
not
persisted,
and
so
he
was
asking
do
we
want
to
keep
that
behavior
and
the
only
reason
I
can
think
of
for
doing.
That
is
because
of
cube
control,
client-side
validation,
if
you
are
submitting
an
object
that
has
this
initializers
field
set
and
the
manifest
cube
control,
client-side
validation
will
not
complain
because
this
alpha
feature
alpha
field
shows
up
in
the
open,
API
spec.
D
And
not
complained
like
if
you're
using
Q
control
create
F
and
you
have
a
file,
and
it
says
initializer
is
null
or
initializers
empty
array.
It
will
start
failing
when
we
remove
this
field.
So
any
automation
you
have
running
Q
control
create
F
that
somehow
got
the
initializers
filled
in.
It
will
start
breaking
silently
all
over
the
entire
world.
E
F
B
D
The
fact
that
it's
not
persisting
I
think
is
the
I
think
that
limits
it
to
do.
We
want
to
run
the
risk
of
breaking
automation
around
Q
control,
create
for
the
percentages
of
people
that
might
have
gotten
the
initializers
field
into
their
generated
UML.
Somehow
those
uneasy
Peaks
fall
out
anyway
right
because
removing
from
your
country
yeah,
but
the
easy
fix
that
this
is
an
automation
loop
in
get
ops
and
it
silently
starts
breaking
and
my
worry
would
be
I
think
the
risk
is
low.
It
will
silently
break
in
automation.
D
D
You
start
adding
it
to
their
config,
although.
B
D
B
D
That
one's
not
typically
something
someone
would
have
an
application
again
like
this,
is
on
every
object
in
the
entire
system,
the
odds
that
no
one
in
the
community
has
been
like.
Oh
yeah,
reuse,
the
cube,
API
public
API
objects
and
do
it
as
low
we
can.
We
can
test
that
at
least
I'm
only
bringing
this
up.
This
has
the
potential
to
break
people,
silently,
which
is
explicitly
why
we
don't
do
this,
and
even
though
it's
an
alpha
I
mean.
B
D
Well,
if
we
remove
it
from
the
internal
types,
but
it's
still
in
the
swag
respect,
then
anybody
who
updates
an
ecosystem
tool
that
just
happens
to
be
generating
this
object.
Meadow
there
is
internal
type,
no,
the
external
type,
that
we
publish
what
people
import,
that
they
will
stop
writing
I,
don't
think
it
has
a
mid
empty.
So
investor
will
stop
writing
null
and
they'll.
Stop
writing
empty
array.
It
did
have
them
empty.
G
B
D
B
B
D
B
They're,
open
API
generation
doesn't
allow
for
conditional
inclusion.
I
was
trying
to
experiment
with
that.
To
have
some
way
of
annotating
feels
to
be
like
this
field
depends
on
this
feature,
so
excluded
from
the
spec,
but
but
today,
like
every
field,
is
in
the
spec,
regardless
of
whether
the
features
enabled
or
not
and.
I
D
B
A
H
H
F
D
Thing
is
there
anything?
Is
there
anything
we
should
be
aware
of
the
PR
did
another
type
of
context
is
anything
we
should
be
aware
of
in
PR
about
in-flight
things,
and
can
we
link
to
it
if
there
is
or
just
make
a
quick
comment
and
then
can
the
vendor
stuff
be
split
out
into
its
own?
Commit
so
is
the
things
I
had.
The
vendor
has.
H
H
G
J
D
B
One
thing
to
think
about
too
on
the
tails
of
the
last
discussion
is
that
this
is
this
would
be
an
alpha
field
that
would
touch
every
object
and
so
just
kind
of
thinking
through.
What
would
we
do
if
we
didn't
want
this
to
progress
like
out?
There
are
things
we
can
do
now
as
we're
introducing
it
to
avoid
being
in
the
same
position
as
we
are
huge
letters,
so.
A
Maybe
ant
one
you
can
jump
on
on
the
field
deprecation
and
the
previous
one
and
weigh
in
there
and
keep
up
on
the
conversation.
Okay,
great.
Okay,
let's
talk
about,
there
was
a
design
proposal
before
flexible
layered
in-flight
requests,
limiting
distribution.
This
just
came
in
so
I,
don't
think
we're
actually
deciding
on
the
the
dock,
but
we
are
announcing
that
it
exists.
Yeah.
A
F
A
Great
thanks
David
and
it
sounds
like
I
mean
some
of
us
had
been
on
the
cup
that
predated
this
and
it
sounds
like
it's
in
court.
A
lot
of
feedback
from
that.
So
I
guess
the
action
item
for
everybody
here
is:
there's
a
link
to
the
doc.
Please
read
it.
Please
add
comments,
please
think
of
considerations
that
need
to
be
incorporated.
A
K
What
were
you
here?
Hey,
hi,
yes,
I'm
here,
hi
guys,
I'm,
double
I
am
one
of
the
co-organizers
or
co-chairs
for
the
group
LPS,
and
you
can
see
in
this
talk
and
not
stock,
a
link
for
a
pull
request
that
you
want
to
add
working
group
LTS
and
the
Charter
of
the
working
group
IDs,
and
we
need
approval
I
mean,
as
you
can
see
in
the
Charter.
K
K
But
if
you
look
at
historically
hyman
again
different
types
of
stakeholders
uses
of
kubernetes
contributors,
maintainer
zuv
kubernetes
has
different
points
have
come
out
and
you
know
mentioned
that
the
release
cadence
of
kubernetes
is
not
suitable
or
you
know
we
do
not
understand
the
version
it's
Q
or
we
need
wider
versions.
Q.
We
need
API
stability.
K
There
are
various
problems
that
people
have
described
on
different
platforms,
Stack
Overflow
or
you
know,
group
the
mailing
list,
six
meetings,
etc
and
and
I
think
at
one
point,
we
thought
that
it
needs
to
be
working
group
that
creates
a
platform
for
everyone
to
come
and
discuss
the
problems
and
come
out
with
a
solution.
The
working
group,
the
uniqueness
of
this
working
group,
is
that
we
have
a
charter
on
this,
whereas
by
the
working
group,
governance
does
not
require
you
to
have
a
charter.
K
The
reason
for
the
Charter
is,
we
think
the
working
group
needs
we
timely
and
we
need
to
you
know,
have
specific
goals
that
we
can
achieve
and
you
can
go
through
the
Charter
there,
that's
unlinked
on
the
pull
request
and
provide
your
comments
and
yeah
I'm
open
for
questions.
If
anybody
wants
to
make
any
comments
on
this
and
I'm
I'm
so
happy
that
I'm
about
to
introduce
this
to
you
all
of
you.
F
A
K
So
yes
go
through
the
Charter
and
let
us
know
you
your
comments
on
that
and
LG
TM
is
what
we
are
looking
for.
We
have
lots
of
the
Charter
contains
links
to
presentations
and
videos
of
what
we,
what
what
the
working
group
wants
to
do
when
we
started
proposing
it,
and
we
have
already
started
meeting
every
two
weeks-
the
calendar
and
white,
etc.
K
If
you
should
join
the
mailing
list,
I've
also
added
the
link
to
the
mailing
list
to
get
Thea
now,
access
to
the
meeting
notes
and
the
recordings
of
the
meetings
and
join
the
meetings.
If
you
have
an
opinion
to
share.
If
and
if
you
want
to
have
the
representation
but
frequently
like
right
now,
we
have
a
survey
that
we
are
planned.
The
questions
for
the
survey
are
in
a
talk
file
and
we
have
requested
comments
from
the
community
we
have
sent.
You
know
Doc's
across
different
mailing
lists,
including
community
staff,
for
comments.
K
D
All
right,
who
had
two
small
ones,
so
one
was
Jordan
I,
just
talked
in
the
background.
There
was
a
an
issue.
We
continue
to
see
issues
where
people
ask
about
load,
balancer
in
front
of
the
API
server
for
inside
the
cluster
I
didn't
know
if
there
was
anybody
else,
who's
falling
along
historically
with
our
attempts
to
add
multiple
connection
backends
to
the
cloud
library,
but
it
is
something
that,
in
the
long
run,
has
some
value
part
of
the
work
we're
doing
with
client
go
around.
D
So
there's
a
couple
of
places
that
have
novel
hooks
around
client
go
to
do
things
like
in
the
cubelet.
We
restart
the
connection
pool
to
get
a
new
client
cert
in
a
few
other
places.
We're
reloading
various
bits
of
client
go
code,
at
least
in
the
next
month,
or
so.
If
anybody
is
tracking
issues
around
Clank,
oh
and
Dyna
dynamicism
changes
that
the
client
go.
Libraries
should
do
I
wanted
to
kind
of
pull
together
the
use
cases
and
move
forward
with
the
lease
stay.
F
D
D
I
want
to
get
that
cleaned
up
and
get
that
inclined
go
and
then
there's
a
couple
of
things
around
transports
where,
if
someone
wanted
to
do
this
kind
of
pick,
a
different
back-end,
we've
tried
to
do
that
as
a
transport
before
it
didn't
work,
I've
been
trying
to
clean
up
trans
kourt's
as
part
of
the
leader
election
step
down
stuff
that
Mike
and
I
were
working
on
so
I
just
want
to.
If
anybody
has
any
use
cases
around
I
know,
dynamism
in
a
client
go
library
and
I
wanted
to
try
and
pull
some
thoughts
together.
D
So
we
don't
drift
too
much
and
the
second
one
was
I.
Had
a
brief
discussion
with
Phillip.
Yesterday,
we've
been
kind
of
poking
around
being
able
to
make
describe
generic
Frost.
It's
one
of
the
last
things.
That's
in
the
couplet,
that's
tied
very
heavily
to
types
and
Jordan
had
kind
of
triggered
this.
You
know
we
were
complaining
about
described
and
he
was
like
it's
basically
just
a
really
crappy
mo
and
so
something
that
Phillip
and
I
were
just
kind
of
brainstorming
off
of
was.
D
Could
we
really
do
server-side
describe
in
a
way
that
is,
is
roughly
a
fairly
generic
mechanism
that
fits
well?
Let's
see
our
DS
an
extension,
so
I
was
gonna
if
people
have
if
people
have
been
thinking
about
describe.
Let
me
know:
I'm
gonna
pull
together
a
quick
brainstorm,
Google
Doc
and
say
that
would
both
six
machinery
just
as
to
start
the
discussion,
maybe
before
a
future
release,
because
it's
the
last
of
the
queue
control
type
specifics
comfortable,
trying
to
go.
A
Okay,
is
there
anybody
that
wants
to
jump
in
right
now
as
interested
in
those
two
things?
Otherwise,
I
think
it
would
be
best
to
just
reach
out
to
Clayton
asynchronously
for
use
cases
or
interest
on
those
I
can.