►
From YouTube: Kubernetes SIG API Machinery 20180131
Description
For more information on this public meeting see this page: https://github.com/kubernetes/community/tree/master/sig-api-machinery
A
B
Has
introduced
the
idea
of
a
default
sweating
for
for
admissions,
and
everything
up
to
this
point
makes
sense.
No
behavior
has
changed
at
all
and
now
there's
a
poll
to
have
a
default
list
of
admission.
Plugins
just
get
normally
running,
cube,
API
server
and
in
his
proposal
with
no
Flags
the
admission
control
flag
will
still
be
respected
as
it
is
today
and
Daniel.
B
You
had
said
you're,
not
sure
you
ready
to
turn
that
on
by
default,
I'm
willing
to
have
a
disable
admission
flag
is
because
today
that's
what
it
means,
if
you
don't
specify
anything
and
I'm
willing
to
have
a
default
admission
flag
and
for
this
release,
I
could
even
see
defaulting
it
to
true.
If
you
want
to.
Although
people
whose
clusters
have
no
admission
are
getting
really
unusual,
behavior.
A
B
A
B
B
So
if
he
wants
an
alpha
feature,
he
might
go
to
turn
on
his
alpha
feature
right,
his
off
a
feature,
submission
plumbing,
but
if
there
are
a
lot
of
cases
where
the
admission
plug-in
is
paired
with
another
feature
and
is
required
to
move
something
safe
or
to
make
it
work
as
expected
and
toleration.
This
is
a
good
example
there,
where,
if
you
don't
control
what
categories
your
taints
on
your
nodes
aren't
doing
what
you
think
they're
doing,
and
so
with
a
fairly
known
list
of
included
mission
plugins
and
we
control
those
fairly
tightly.
A
B
A
D
A
E
C
B
A
E
Don't
have
to
use
it
thankfully,
but
I
was
just
you
know
in
an
in
an
abstract
sense,
if
any
of
your
users
that
were
using
it
yeah.
D
F
C
A
link
CPR
or
the
issue
there.
We
came
up
with
inconsistency
in
our
series,
so
we
are
claiming
that
CR.
These
are
likely
different
sources
and
we
realize
that
this
is
not
true,
because
we
persist
everything.
Everything
is
adjacent
block
a
client
sense
and
people
start
to
think
about
these
additional
properties
for
like
API
spec,
where
you
can
is
tweaked
what
huge
yeah
future
fields
look
like.
For
example,
a
field
which
is
not
defined
specified
can
just
be
a
string,
for
example,
and
this
yeah.
Let's
do
this
inside
that
we
do
not
strip
unknown
fields.
C
B
There
are
a
couple
wrinkles
too,
that
I
commented
in
the
pull
request.
The
first
is
that
we
would
only
be
able
to
do
it
if
you
have
to
define
the
open
API
schema
for
the
validation
feature,
which
is
optional
for
customer
resource
definitions,
I
think
that's
an
acceptable
limitation,
but
it
exists
and
I.
A
E
C
B
That
feature
is
optional
kind
of
so
I
write
about
being
a
limitation
I
just
want
to
point
out.
There
is
a
limitation.
The
other
thing
is
that,
are
we
okay
saying
that
you
must
list
all
the
fields
in
the
validation
section
that
in
where
you
define
the
scheme?
I
think
that
I
am
we
do
that
effectively
today,
but
I
wasn't
sure
if
anyone
had
comments
like
no.
If
I
turn
this
on
for
this
kind
of
a
field,
it
won't
valid
you
properly,
ina.
A
I
think
it's
reasonable
to
you're
submitting
a
validation,
I
think
it's
reasonable
to
request
that
that
fella,
that
the
schema
is
complete
like
you're,
not
submitting
the
validation
to
just
validate
a
partial
set
of
fields.
You
should
at
least
list
all
the
fields
that
you
intend
to
set
in
English,
Jima,
I,
think
and
so
I'm
working
on,
like
a
design
for
if
I
ever
get
around
to
it.
I
designed
for
moving
cute
control
apply
like
at
least
some
part
of
it
into
the
control,
plane
and
I.
Think
for
Sierra
T's
to
make
apply
work.
E
A
B
Then
the
next
question
becomes:
what
are
we
gonna
do
about
backwards?
Compatibility
of
what
we
have
today
in
order
to
be
backwards,
compatible,
go
behind
and
you
would
have
to
make
this
optional
now
we
have
to
default
off
right,
but
if
we
wanted
to
make
a
break
here
as
one
of
those
we
made
a
mistake
in
beta,
he
could
make
it
not
default.
A
G
B
I'm,
okay
with
it,
but
my
tolerance
is
fairly
high.
Yeah
means
you
kind
of
did
something
wrong
to
begin
with
yeah.
A
I'm
not
sure,
if
I'm
not
sure
how
bad
that
is
so
so
one
thing
I've
been
thinking
about
is
generally
having
a
storage
yeah.
Actually
it's
listed
in
my
section
down
there
having
a
component
that
makes
sure
the
storage
version
for
objects
is,
is
correct
because,
right
now
you
change
the
storage
version
and
it
doesn't
actually
change
anything
and
that's
the
GMs
people
happen
to
read
or
write
the
object.
I
think
that
is
the
sort
of
thing
that
could
that
that
mechanism
could
be
used
to
fix
this
as
well
like.
A
A
B
A
Yeah
I
I
think
the
ideal
setting
is
on
like
we
should
strip
by
default.
If
you
provided
the
open,
API
spec,
but
we
don't
want
to
break
people.
Maybe
this
is
something
we
could
change
over
the
the
like
like
we
could
change
the
default
when
we
move
from
beta
to
GA.
Maybe
that's
yeah,
just
my
personal
spoke
yeah.
So.
D
A
Not
validate
on
reads:
no,
we
don't
validate
out
really.
We
can
still.
We
can
still
warn
people
like,
even
if
they've
chosen,
not
a
strip,
we
can
still
warn
people
that
when
we
go
to
GA
we're
gonna
start
stripping
by
default
and
your
object
is
going
to
be
modified
in
a
way
that,
maybe
you
don't
expect.
D
G
C
B
You
actually
make
use
of
this
ability
when
upgrading
pieces
right.
So
if
you
add
a,
if
you
add
a
field,
and
you
want
to
make
sure
that
no
one
was
squatting
on
the
field
before
right.
So
if
you
had
a
feel
that
says,
let
me
do
super
awesome
privilege
to
think
you
want
to
be
sure
when
you
upgrade
your
server
to
start
using
that
new
version
that
no
one
has
already
at
field
trip,
and
so
by
strapping
fields
in
advance.
You
sort
of
got
that
for
free.
D
I
can
give
you
guys
a
brief
introduction
and
then
we
can.
We
can
talk
on
the
things
that
we
have
in
versioning
design
right.
Yes,
the
versioning
sierra
versioning
weeks
ago.
The
idea
of
the
OSS
validation,
sir,
the
co2
versioning
design
that
we
decided
that's
good
way
to
go
is
using
web
hooks
for
converting
all
your
days
to
new
ones.
D
There
are
two
point
of
conflict,
sir,
as
I
see
in
the
comments
that
people
have
there,
one
is
reusing
the
existing
web
hook.
Api
is
an
implementation.
I
think
people
don't
have
any
problem
with
the
implementation
implementation
part,
but
there
are
propositions
using
the
same
API
that
you're
using
for
other
web
books
like
mutating
and
validating
for
conversion
web
hooks.
D
D
B
B
A
B
I,
don't
so
so,
if
you
look
at
what
the
differences
are
between
the
API,
is
there
about
90
different
right
like
if
you
look
at
the
fields,
there
are
enumerated
many
in
there
we
can
go
through
the
list,
but
basically
no
field
matches
between
what
admission
does
and
what
conversion
logically
does
right.
Conversion.
Oh.
A
E
E
B
A
A
B
Guess,
I
don't
see
one
web
hook
really
anywhere
in
our
stack
right.
We
started
out.
We
already
had
two.
We
added
a
third
we're
talking
about
adding
a
fourth
I'm.
Ok
with
that,
if
we
want
to
trust
you
combine
specific
things
for
a
custom
resource,
specific
webhooks,
maybe
we
only
have
one
today
we
could
talk
about
what
the
next
one
would
be.
It
would
probably
be
something
like
defaulting,
but
I,
don't
so.
A
A
C
I
I
think
I
mean
we
don't
guarantee
how
troops
we
retells
developers
to
implement
that
properly,
which
we
have
no
guarantee
whatsoever.
So
we
have
to
trust
in
anyway,
and
I
will
do
videos
I,
don't
think
it's
so
it's
if
you
want
to
make
modification
rotations
incongruity.
Of
course,
it's
possible
to
forbid
it.
If.
E
A
E
No,
so
the
only
other
so
I
do
agree
that
for
validation,
it
should
be
done
by
the
admission,
control
and
conversion.
Maybe
it's
its
own
series
I
will
play
devil's
advocate.
I,
know
an
argument
that
this
is
kind
of
your
kind
of
mixing
what
books
all
over
the
place
like.
For
example,
if
you
just
scope,
CR
DS,
it's
like
we're
doing
validation,
we
use
the
general
one
if
you're
doing,
conversion
use
the
CRT
specific
ones,
and
so,
while
I
think
that's
the
correct
approach
right
now,
it
would
have
to
be
heavily
documented,
I.
Think.
B
Yeah
I
mean
I
think
the
good
news
there
is
that
I
don't
think
that
there's
any
user
that
would
say
to
themselves.
You
know
what
I
really
want
to
do.
I
want
to
own
my
own
special
conversion
for
how
I
convert
between
apps
v1
and
a
have
to
be
1,
beta,
1,
yeah,
I.
Think
we
wouldn't
let
people
register
I,
guess
what
I'm
saying
is
I,
don't
think
users
will
ever
accidentally
say
hey.
This
is
the
thing
I
want
to
do.
Let
me
pleasure
gonna.
Do
it
they'll
probably
say
gee
thank
God
I.
D
A
A
The
second
one
is
about
maintaining
the
storage
version.
Yes,
sir,
and
and
I
just
I
know
you
put
some
detail
in
your
document
about
it.
I
think
it's
a
general
problem
that
we
need
to
solve
for
all
API,
so
we
should
just
make
sure
that
the
CR
needs
work.
The
way
our
native
resources
work,
so
that
when
we
solve
this
for
everybody
C
or
these
take
advantage
of
it,
yep.
C
I
mean
my
suggestions:
to
not
I
mean
we
won't
have
this
mechanism
for
quite
some
time
for
one
two,
three
versions,
and
if
we
design
something
now,
should
we
let
the
developers
of
see
IDs
yeah?
Well,
it's
this
this
migration
code.
They
will
do
it,
they
will
touch
and
make
a
loop
over
all
objects,
so
the
implements
at.
Do
we
really
want
that
I,
wonder
whether
we
can't
just
implement
it
in
our
facilities
and
it's
better
anyway
in
the
beginning,
so
we
can
make
experience
with
it.
I
I.
A
B
Stefan
was
actually
suggesting
that
he
you
take
another
look
at
something
like
Oh,
a
DM,
migrate
storage
that
we
have
to
in
OpenShift
that
actually
goes
and
updates
all
your
storage
versions
for
you.
Oh
you
have
this
already.
We
had
it
since
three
one
yeah
we
had
what
we
had
to
see
through.
There
was
an
issue
where
we
linked
you
to
do
the
implementation.
It
is
as
simple
as
a
read
and
write,
write
this
on
a
script
or
is
it
a
consider
controller?
It's
just
a
simple
script:
yeah.
B
C
So
there's
one
reason
why
I
also
want
to
see
that
if
I
delete
for
Tom
the
storage
version
or
just
DD
version-
and
we
we
clean
up
everything
at
this
times,
this
deletes
version.
It's
still
available.
It's
a
storage
in
super,
told
us
and
told
us
he
said,
but
it's
not
valuable
anymore.
So
we
need
an
equation
for
that.
So
it's
more
than
just
having
a
controller
and
having
the
API
exchange
and
server
independent
from
that.
A
So
I
so
I
actually
have
a
fairly
detailed
design,
doc
that
I
haven't
published
on
this.
So
it
sounds
like
I
need
to
publish
it,
and
we
should
debate
there.
Whether
this
is
whether
the
approach
I
have
in
mind
is
the
right
one
or
you
have
an
easier
one.
I
think
I
do
want
an
approach
that
requires
zero,
zero
operational
burden,
so
I'm.
F
B
A
B
Somebody
aggregates
my
dream.
Well,
so
let
me
go
through
those
cases
right.
So
if
somebody
aggregates
an
API,
the
new
API
server
will
be
storing
at
its
configured
storage
version.
When
you
upgrade
that
one,
you
have
to
run
your
script
right,
it'll
update,
but
once
you've
updated
the
server
since
the
storage
version
stays
the
same
for
the
life
of
that
server.
No.
A
B
A
B
A
B
B
A
test
every
every
resource
in
the
system
it
uses
discovery
to
find
them
all
which
allows
to
write
every
resource.
It
reads
that
CD
to
see
what
version
it
was
stored
in
and
any
movement
has
to
be
explicit
in
that
test,
and
so
one
has
to
have
thought
about
it
on
the
de
challenge.
What
if
somebody
adds.
A
A
D
And
last
comment:
I
have
on
ERG.
Versioning
is
I
kind
of
disappointed
that
in
one
day
that
I
will
publish
this.
Nobody
talked
about
the
declarative
conversion.
I
really
like
the
idea.
So
there
is
a
gonna.
Have
there
is
really
half-baked
I
just
want
you
guys
to
know
to
just
look
at
it.
I
don't
want
to
see
us
in
a
position
in
in
a
year
say:
oh,
you
should
have
went
with
the
declarative
approach.
A
Cycles
and
sounds
interested
one
of
these
sounds
interesting.
You
can
thing
offline,
but
I'll.
Just
very
briefly
describe
the
problems.
The
first
one
on
my
list
is
no
note.
Heart
beats,
especially
if
you
have
a
lot
of
images
on
your
notes
or
those
images
have
a
lot
of
names,
because
NCD
does
copy-on-write
every
time
the
nodes
heartbeat,
which
is
ten
times
a
second.
It
makes
a
copy
of
the
entire
note
object.
A
So
if
you
do
the
math
on
ten
times
a
second
times
the
number
of
notes
in
your
system
times
the
number
of
images
plus
the
number
of
image
names
that
adds
up
to
a
great
deal
of
data
I,
don't
know,
we've
been
thinking
about
whether
we
can
split
off
the
just
the
heartbeat
field
into
a
separate
object
or
maybe
move
it
to
a
sub
resource,
or
maybe
maybe
there's
a
change.
We
can
make
to
xee
to
make
this
more
efficient.
I,
don't
know
if
that
sounds
interesting
to
somebody.
A
B
A
B
B
B
A
B
A
A
A
B
I
would
be
fine
if
I
were
trying
to
protect
myself
as
Nene
getting
too
big,
as
opposed
to
trying
to
make
sure
that
one
particular
set
of
projects
didn't
use
too
much
aggregate
CPU.
If
I
were
trying
to
protect
myself
from
NCD
being
too
big
I
would
try
to
look
for
something
different
than
quota
different
than
quota
as
it
exists
today,
just
scaled.
The
clustering
level
I
would
look
for
something
once
I
mean.
A
Yeah
I
think
it
would
be
useful
to
be
able
to
track
the
quota
over
the
whole
the
whole
cluster,
then
the
next
one
on
my
list
is
rate
limits
back
pressure.
This
is
this
is
like
the
previous
one.
Is
people
making
too
many
things?
This
one
is
people
making
those
things
too
frequently
or
making
a
deleting
them
too
frequently
or
just
calls
general,
so
I
know.
We've
definitely
talked
about
this
before,
but
I'm
interested
in
somebody
I.
Don't.
B
A
A
A
F
A
Something
that
we've
been
discussing
a
fair
amount,
which
is
the
control
plane
version,
doesn't
latch
right
now,
if
you
are
doing
a
rolling
update
of
your
control
plane,
you'll
have
some
API
servers
at
various
versions
at
different
versions,
which
me
and
they're
serving
different
versions
of
your
api's.
They
have
different
flags
and
I.
A
A
I
I
would
and
I
met.
You
I
also
think
we
shouldn't
allow
people
to
compare
them
on
single
resources
and
I
have
sort
of
two
reasons.
For
that
one
is
it's
kind
of
lazy
and
you
can
do
whatever
you
need
to
do
other
ways.
The
other
reason
is,
it
does
tie
our
hands
in
terms
of
what
storage
systems
we
could
eventually
support.
I.
A
A
A
H
A
B
B
B
F
B
A
G
A
A
G
A
But
maybe
we
could
change
that
to
a
like:
isn't,
is
within
window
predicates
and
allow
the
allow
the
storage
back-end
to
tell
us
I
think
that's
a
better
way
of
handling
that
so
Jenny
in
case
you
couldn't
hear
Jenny
says
that
the
the
caching
layer
inside
API
server
does
this
in
incorrect
comparison.
But
that
doesn't
surprise
me
like.