►
From YouTube: Kubernetes SIG API Machinery 20171207
Description
For more information on this public meeting see this page: https://github.com/kubernetes/community/tree/master/sig-api-machinery
A
B
B
B
A
B
A
C
C
B
B
I
figured
that
out,
they
always
seems
all
the
objects
are
injured
or
any
key.
So,
even
though
the
map
from
key
to
object,
it
all
exists
that
the
object
always
moves
its
key,
a
little
which
is
kind
of
annoying.
You
can't
the
object,
yeah
but
I
understand
a
need
of
the
feed
that
the
audience
for
this
was
our
API
objects.
So,
when
you're
trying
to
you
to
store
less
than
that,
the
interest
of
a
certain
space
do
believe.
Yes,.
A
E
C
In
one-eighth
there
are
when
I
was.
It
was
money
to
last
my
last
cycle
trying
to
get
this.
They
were
starting
to
be
tax
on
repos
and
but
before
it
was
like,
you
went
into
the
main
kubernetes
route.
Fellow
and
then
you
look
at
like
whatever
repo
and
you're
like
look
at
the
commit
message
to
seem
like
which
should
be
happy
and
then,
like
all
libraries
necessarily
matter
like
the
point
at
which
they
were
murdered
back
out.
C
C
C
A
A
C
It's
time
in
a
good
user,
other
plans
like
just
actually
vendor
them
into
like
I
mean
that,
between,
like
mean
they're,
like
the
most
sensible
way,
is
to
have
all
the
independent
libraries
and
then
that
we
should
get
vendor
into
Cooper
many,
it's
a
balance.
I
mean
I
want
to
recommend,
gets
hot
out
of
it.
Curious.
C
A
Hard
and
specifically
it's
hard
because,
like
testing
making
sure
that
things
still
works,
yeah
some
people,
don't
we
have
to
make
multiple
PRS
to
to
make
a
PR
to
fix
the
library
make
another
PR.
The
venturi
repository
employee
library
is
importing
other
libraries
which
we
do
and
you
can
have
like
two
or
three
or
four
layers
of
gender
APRs
I,
don't
know
I,
think
practice.
I
don't
have
too
much
sympathy
for
that,
because
I
think
practically
speaking,
you
should
probably
be
making
possible
TRS
anyways.
C
I
mean
it
seems
also
to
give
the
factor
of
like
how
new,
in
the
velocity
of
change
and
Mikey
getting
the
extension
points
there
such
that
it
can
be
funded
it
with
me
and
it
just
like,
with
underlying
libraries
of
it
can
forever
stabilize
worth
your
time.
You
wouldn't
necessarily
have
that
many
changes
going
on
in
the
library
of
themselves
versus
yeah.
A
A
C
A
B
Yeah,
what
were
you?
What
were
you
another
question
this
may
be?
This
is
known
the
documentation
confused
by
where
the
threshold
lies
between
what's
an
API
machine
reading,
what
belongs
and
client
go
and
I
think
I
think
I
know.
The
answer
is
that
you
might
eat
it
when
you're
writing,
something
that
gives
it
to
client.
It
should
probably
be
an
API
machinery
either
below
client,
though,
or
not
love
it.
It's
not
okay.
So.
A
A
This
is
the
screen
on
this
is
the
slide
I'm
got
so
far.
Maybe
y'all
can
help
me
break
this
like.
So
what
is
AP
is
machinery
owned
and
sort
of
concrete
details?
Kubernetes
is
a
hub-and-spoke
api
system
where
the
api
server
is
sort
of
the
hot
on
reasoning.
Clients,
including
like
most
of
the
functionally
speaking
most
of
the
business
decisions,
are
made
in
clients,
so
we
own
the
API
server,
which
is
the
center
of
the
hub.
A
A
We
don't
own
the
operational
aspects
of
Etsy,
D
I
will
say:
my
team
at
Google
is
involved
with
some
of
that
and
then
Google,
but
and
so
we've
got
experiences
trying
to
supply
good
solutions,
and
now
the
qualities
are
ready
header
in
a
similar
position,
but
we
can
it
we're
not
you're,
sorry,
if
that
makes
sense.
We
also
own
the
client
generators.
So
this
is
like
open,
API
spec,
which
makes
it
all
possible.
A
The
generation
stack
loops,
a
client
repo
with
the
gem
I
I've,
heard
chatter,
about
like
starting
up
a
client
sig,
which
maybe
we
could
actually
roll
those
over
onto
and
and
if
that
happened,
I
think
we
would
still
own
a
generation
the
next
step,
but
for
now
I
think
our
civic
basically
owns
the
generators.
Also
a.
A
A
There's
two
main
things:
I
have
in
mind
when
I
guess
we
do
own
some
of
the
things
inaccurate,
there's
two
main
things
I
have
in
mind:
one
is
the
the
there's
some
fooling
around
eating.
Actually
not
eating
cake
gets
back
into
the
starter
generator
program,
so
we
own
that
actually,
and
we
also
own
the
the
go.
Client
has
its
own
special
generator
on
that
yeah
in
the
infirmary
generator.
But
that's
that's
not
and
I.
Guess
in
some
kind
of
a
client
thing,
yeah
most
view
most.
A
The
generation
I
think
is
fine,
is
the
is
the
serialization
code
that
turns
Lee
goes
friends
into
a
wire
format.
I
think
that's
the
best
reasonable
case
for
code
generation,
but
the
like
the
informer,
Jinbo
client
in
like
like.
Wouldn't
that
be
so
much
nicer
with
with
trials
you
know
generous
and
like
the
deep
copy
can
like.
Most
of
the
reason
that
we
have
is
the
door
doesn't
have
a
constant
keywords:
yeah.
A
C
A
So
the
reason
why
I
think
namespaces
is
under
our
jurisdiction
is
there's
no
one
else
endure
and,
and
if
and
well,
multiple
people
have
in
the
last
couple
weeks
floated
this
idea
by
me.
If
we
were
to
separate
out
the
API
serving
stack
nicely
people
people
would
come
to
me
like
I
want
to
do
kubernetes
style.
Api
is
without
a
kubernetes
cluster,
and
if
you
look
at
what
what
does
it
take
to
provide
the
kubernetes
api
style?
A
You
need
the
garbage
collector
and
you
need
namespaces
because,
like
if
you
want
to
just
run
series,
for
example
their
name
space,
you
need
the
name
space
controller
to
like
start
up
and
shut
down
those.
So
I
think
I
think
the
name
space
controller
is
actually
part
of
the
general
api
she's
very
doing
that
we
have
it.
A
Yeah
I've
been
suggesting
to
people
if
you
want
to
experiment
with
that,
just
run
a
stock,
kubernetes
api
server
and
use
the
runtime
config
bug
to
turn
off
all
the
api's
that
you
don't
want
to
use
I
think
it
should
wear
a
band
tried
it.
It
should
be
pretty
cool.
Let's
see
what
else
do
we
own?
We
own
the
API
style
guide,
the
the
patterns
like
and
status.
A
Yeah
and
that
adds
up
to
about
25%
of
the
kubernetes
codebase
some,
but
not
all,
that
is
due
to
the
vast
quantities
of
generated
code,
but
I
think
it's
also
important
to
say
what
kubernetes,
what
the
API
machineries
sig
is
not
because
there's
a
lot
of
confusion,
not
in
github
issues
we
sort
of
near
the
top
of
the
list
of
six,
and
we
have
the
word
API
in
our
name.
Therefore,
clearly
beyond
to
look
at
your
API
review,
actually
know,
there's
a
separate
group
for
doing
these
every
use
and
stick
architecture.
A
Unfortunately,
I'm
in
both
groups,
so
I
still
might
have
to
look
at
your
Eve
every
view,
but
the
sig
API
machinery.
Doesn't
we
don't
own
those
controllers,
although
Walter
percent
here
is
looking
into
sort
of
horse
front
here
to
the
controller
manager
in
DB,
but
still
that
might
mean
that
we
own
the
framework
in
which
the
controllers
are
running,
but
we're
still
not
going
to
own
a
particular
controller.
A
You
do
consult
a
lot
on
controllers
like
there's
a
bug
in
that
of
the
deployment
controller.
Where
does
properly
diff
the
replica
stuff
that
is
made
with
its
back
and
it
can
make
a
lot
of
replicas
sets
under
certain
very
rare
conditions.
So
so
my
suitemates
spend
a
lot
of
time
talking
to
the
aptitude.
A
B
A
There's
there's
too
many
people
with
clusters
out
there
in
the
world
for
us
to
either
exhaustively
help
everybody.
Unfortunately,
we
don't
own
the
what
I'll
call
downward
facing
extensions
mechanisms
so
like
CNI
CSI,
the
other
ones
like
the
runtime
interface,
the
storage
interface
and
networking
interface
that
cublas
uses
to
integrate
with
clusters.
I
am
I
in
my
mind,
I
have
extension
mechanisms.
F
A
F
F
A
There's
a
yeah
there's
some
tension
here
between
dynamic
and
statically
compiled
things,
so
the
informer
generator
is
a
way
of
generating
a
type
safe
Informer
that
that
takes
a
specific
type
to
the
end.
You
don't
have
to
do
any
like
type
ass
or
anything
you
just
type
back.
We
have
others,
so
the
kubernetes
api
system
is
kind
of
unique
in
that
we
have
a
common
metadata
format
that
all
of
our
objects
support.
So
that
means
it
makes
sense
to
consume
objects
that
you
don't
know
the
concrete
details
of,
but
you
do
understand
the
metadata
enough.
A
So
in
fact
our
garbage
collector
has
code
that
generically
acts
on
this.
So
it's
a
good
thing
to
look
out
there
yeah
yeah
generally
like
either
either.
You
know
statically
a
compiled
plan.
What
objects
you
want
to
operate
on
or
you
don't.
So
if
you
know
what
objects
you
want
to
operate
on,
you
can
run
the
country
and
former
generated
a
client
generator
and
you
get
the
static
type
tape
code.
Don't.
A
A
E
A
A
layer
on
top
of
an
untyped
library,
so
even
you
can
look
at
the
thing
that
it
makes
directly
the
the
sheraton
former
business
is
shared
in
the
sense
that
it
lets
multiple
controllers
running
in
the
same
process
share
a
single
copy
of
the
objects
if
you're
watching.
So
it's
from
memory
efficiency
to
be
I.
C
A
Yeah,
the
so
the
informer
stuff
does
have
a
it.
Doesn't
client
ID
indexing
if
you
can
do
arbitrary
flexi,
depending
on
what
you're
doing
that's
that
can
be
massively
inefficient
for
this
established.
If
you
download
hundreds
of
thousands
of
pods
to
find
like
period
limits,
that's
that's
not
great.
We
would
like
the
surfer
to
do
that
for
you,
unfortunately,
for
a
better
or
for
worse,
it's
not
super
easy
with
it
wasn't
super
easy
with
it
wasn't
possible,
with
that
TV
queue
to
keep
like
atomic
index
is
up
to
date.
A
A
Yes,
so
all
of
our
selectors
are
linear
and
a
number
of
objects
in
the
namespace
of
that
particular
kind,
because
it's
becoming
base
so
selectors
and
labels
are
such
that
you
could
index
the
variables
separately
behind
them
into
inefficient
query,
probably
like
log
and
the
number
of
labels,
or
something
like
that.
Yeah
I'm
sure
why.
A
Well,
let
me
let
me
let
me
just
note
that
API
server
is
an
H
8
system,
so
there's
multiple
AJ,
there's
multiple
API
servers
that
are
hopefully
all
serving
the
same
sort
of
data
or
less
yeah,
so
we
socratic
related.
So
we
depend
on
EDD
to
provide
the
to
do
the
heavy
lifting
there
with
with
wrap
in
HK
systems.
You
need
to
turn
on
or
on
breathe,
so
that
all
of
your
API
servers
will
even
on
reads
give
you
the
same
the
same
thing
that
makes
it
tricky
to
busy
indexes
in
an
API
server.
A
A
Api
server
so
I
don't
think
it's
impossible,
but
it's
definitely
not
simple,
and
so
the
sort
of
the
reason
that
I
am
wanted
to
do
I.
We,
we
actually
do
some
caching
in
API
server,
because
we
basically
catch
all
the
watch
with
watch
everything
in
the
API
server
and
redistribute
that
and
watches.
So
the
client
side
watch
does
not
resolve
in
the
watch
to
at
CD.
A
We
did
that
for
a
while,
but
I
don't
want
to
go
down
the
road
to
make
writing
a
database
software
I,
don't
want
to
go
down
that
road
too
far,
I
can
avoid
I.
Don't
think
that
that
is
our
competitive
advantage,
so
so
yeah.
This
is
something
we
need
to.
This
is
something
we
need
to
think
about,
and
I
also
don't
think
we
want
to
support
arbitrary
greens
against
API
objects.
A
A
Yeah
that
said,
I'm
I,
if
I,
have
somebody
who's
pretty
interested
in
making
that
generally
efficient.
Let's
talk,
I'm,
not
I'm,
not
like
super
opposed
I,
it's
probably
not
where
I
can
send
my
resources,
but
but
yeah
yeah.
There
is
a
lot
of
demand
for
running
kubernetes
style,
api's
independently
from
clusters,
so
I
think.
A
A
A
A
There's
when
we're
talking
about
rent-a-cops,
there's
two
principal
components
to
that.
One
is
what
you
pay
on
every
note.
What's
the
note
agent
overhead,
because
that
is
multiplied
by
the
number
of
nodes
your
system
becomes
a
very
large
number
and
the
other
number
is
what
you
pay
for
your
control
time.
A
You've
only
got
one
control
plane,
so
you
can
afford
play
a
few
percent
of
your
cluster
size
for
the
controller
and
that's
not
that's
not
a
terrible
deal
unless
you're
on
a
very
small
cluster,
which
case
a
few
percent
makes
it
impossible
Vermont
it
will
few
percent
of
102
eg
busting
less
than
one
quarter
or
yours
you
to
write
code.
So
I
will
say
we
have
a
number
of
cannibal
flags
on
API
server,
which
control
how
big
it's
ashes
are
and
then
you
should
probably
turn
off
some
of
the
caches
completely.
D
A
B
B
A
That's
a
yes,
let's
give.
Another
point
is
because
whoever
has
a
lot
of
flags
and
they're
there.
It's
basically
highly
religious
touch
them.
Yes,
so
I
would
like
to
see
us
in
the
next
year
moved
towards
hitting
away
from
those
flags
and
towards
I
mean
ideally,
a
dynamic
configuration
like
the
web
hooks
where
there's
seriously
API
objects
in
the
system
that
track
the
configuration
for
the
web
book,
and
this
means
you
can
put
like
our
back
permissions
on
them
to
decide
which
of
your
users
can
add
additional
admission.
I
think
there's
some
flags.
A
A
C
C
Kind
of
I
think
it's
also
complex
because
just
roll
or
anything
like
that
and
how
they
implemented.
It's
like
get
the
history
lesson
and
understanding
like
the
linearity
of
that
history.
Lesson
like
you
know,
if
you
see
one
thing
using
this
mechanism
yeah
using
something
else
that
use
kind
of
like
figure
used
by
you
know,
you
look
at
different.
People
who
are
committing
here,
like
I'll,
follow
their
mental
model,
yeah.
A
I
think
you
know
there's
this
like
I,
don't
want
to
document
the
current
state,
because
I
want
to
change
like
a
dozen
things
but
turns
out
we're
gonna
live
in
the
current
state
for
quite
a
while,
so
we
need
to
do
a
better
job
of
documenting.
Even
even
if
we
don't
like
the
documents,
we
have
to
write,
which
is
yeah
yeah.
Definitely,
if
I
were
to
do
this
over
again,
I
would
do
a
few
things
Italy,
but
it
is
what
it
is
and
a
lot
of
people
are
using
it.
So
so
we
always
should
immediately.
C
C
E
D
D
I,
don't
know
how
much
time
do
we
have.
This
is
like
long.
It's
until
12:30,
oh
they're,
like
other
holy
cow
sessions,
run
over
top
of
this
so
like.
If
you
wanted
to
go.
Tell
other
sessions
like
you
need
to
pay
into
the
time,
yeah
and
believe
I
think
there's
another
one,
starting
in
like
two
minutes.
So
if
you
are
planning
on
going
to
another
session.
D
Like
my
review
is
that
there's
there
there's
like
public
document
ECPI
itself,
and
then
there
is
the
like
how
to
what
can
is
the
right
mix
of
architecture
is
in
the
records
of
ways
of
building
fees
like
beholder
is,
and
what
like
there
are.
You
know
like
the
same
look.
The
color
is
great,
but
like
you
don't
so
the
holy
has
one
color
right,
so
you
might
have
liked
the
process
that
you're
running
like
an
operator
or
whatnot.
They
has
multiple
controllers.
I've
run
inside
this
new
process.
You
know
it's
like
okay.
D
So
you
don't
have,
like
parallel
changes,
source
of
objects
for
a
particular
type,
so
these
type
of
architectural
things
like
aren't
obvious
unless
you're
like
actually
reading
the
source
code
for
communities
I
mean
and
then
on
top
of
readers,
from
being
able
to
like
extract
out
the
meaning
of
be
like
why
the
heck
are
they
giving
it
this
way?
Why
are
all
these
like
crazy
things,
a
little
bit,
I
think.
C
Was
like
hey,
we
need
to
do
this
and
it
was
basically
like
look
at
what's
going
on.
You
know,
controller
and
like
try
to
copy
something:
that's
not
necessarily
as
complex
that
cool
and
then
things
like
the
qz
stuff,
like
that
I
think
that
makes
so
much
sense
once
you
get
it
right
and
get
it
then.
Maybe
when
you're
building
new
controllers
you're
also
creating
lots
of
boilerplate,
which
you
know
kind
of
wrap
it
up
in
some
yeah
Italy.
D
A
D
C
D
Reading
their
people,
like
you,
know,
yeah,
the
original
people
were
in
it
did
is
for
surgical
rabies
and
we
figured
out
how
to
do
it,
and
then
they
wrote
their
own
in
that
kind
of
model
when
they
people
so
that
in
the
main
sorted
medium.
And
so
you
have
this
like
I'm
talking
gameplay,
you
know
like
three
projects
or
projects
and
people
start
like
me.
Even
mistakes
that
you,
like
you
shouldn't.
Even
you.
D
A
D
I
mean
like
just
the
architecture
too,
of
like
a
controller
like
this,
like
this
is
what
you
do.
These
are
why
Sheraton
plumbers
exist.
Like
you
know.
This
is
why
I
work
use
exists.
Like
you
know
you
every
once
a
while.
You
see
people
that,
like
controllers,
that
where
they
do
like
a
lot
of
logic
like
in
the
hooks
for
the
for
the
watches
and
you're
like
no.
D
No,
no,
please
don't
do
that,
because
that
will
run
in
parallel
so
like
just
making
sure
that
folks
avoid
those
pitfalls
and
that
kind
of
print
like
sound
architectures,
when
they're
building
things
and
layout
I
also
often
see
the
problem
of
voice.
People
will
write
everything
in
one
controller
and
then
like
they
will.
We
basically
do
a
sync
loops.
That
life
goes
and
does
everything
you
know,
even
though
they're
like
they're,
watching
one
object
and
then
they're
creating
and
monitoring,
like
maybe
several
other
ones
yeah
like.
D
Maybe
they
have
one
CRT
that
creates
five
other
types
of
convenience
objects.
And
then,
whenever
one
of
these
changes
they
have
synced
the
entire
CRT
again,
which
then
syncs
all
of
these,
which
is
not
always
the
ideal
thing
to
do,
partly
because,
like
people
will
do
sync
Luke's
square,
like
all
the
logic
is
sequential
and
they'll
do
things
like
change
things
in
the
middle
of
it
and
then
later
on
in
the
loop
to
the
lake.
You
know
something
else
will
happen
like
they'll
do
something
else
and
that's
contingent
on
the
text.
D
So
they're
very
approach
would
be
like
build
things
in
more
modular
way,
such
that
it
can
be
run
in
parallel,
like
you
have
multiple
controllers
that
run
in
parallel.
Looking
at
the
same
so
like
so,
for
example,
you
know
maybe
watching
a
single
CRD,
but,
like
you
have
multiple
controllers,
that
watch
the
C&C
already
and
then
you
know
right
out
the
individual
objects
that
you
care
about
and
then
because
you're
doing
definitely
I
mean
there
are.
D
D
You
know,
you're,
basically
doing
one
watching
this
year
at
eat,
but
you're
just
it's
just
calling
the
books
in
each
controller
and
then
doing
the
stuff
that
can
happen
in
parallel
and
and
then
each
one
of
those
is
fairly
simple
in
what
it
actually
does
it.
Just
like
this.
Okay,
something
changed
in
my
security.
I,
look
at
this
field
and
I've
read
out
this
object.
Yeah,
the
more
sources
you
have
controllers
are
smarter,
just.
D
A
Of
boilerplate
yeah,
okay,
so
please
I,
like
the
points
controller,
is
like
an
example
of
the
sort
of
motor
controller,
watches,
watches
pods
and
it
watches
services,
and
it
writes,
and
once
always
that
the
in
points
controller,
yeah
yeah
it.
It
basically
just
said
the
joint
you
know
find
to
a
pilot's
or
any
time
upon
and
changes.
It
Dirty's
the
services
that
naturally
chemist
services
changes.
A
It
also
agrees
it
and
that's
a
yeah
I,
don't
know
how
much
is
this
changed
since
I
wrote
it,
but
that
that's
the
one
that
I
usually
tell
people
there's
also
like
a
guitar
when
you're
dealing
with
an
object
that
is
in
a
different
API
system
like
like
you,
you're,
like
a
club,
API
yeah
like
starting
to
load
balancers,
it
might
not
be
a
watch.
Nothing
might
not
be
a
convenient
watch
to.
C
C
A
C
A
A
So
one
of
the
things
that
I'm
thinking
of
trying
to
get
someone
to
do
in
the
next
year
is
a
introspection
API.
So
you
can
ask
because
the
garbage
collector
has
a
graph,
basically
every
object
in
the
system
and
what
its
parent
is.
It
would
be
awesome
if
you
could
query
that,
and
it
just
gives
you
back
two
names
and
stuff
or
UI
knees
or
jeans
basins
whatever.
So
that's
something
I.
C
Do
yeah
helps
here
because
there's
lots
of
code
that
we
have
found.
It's
like.
Okay,
go
look
at
a
list
error
for
this
height
for
the
names
base
of
you
know,
then
the
top
levels
he
or
any
object
is
it
go
collect
all
of
those,
and
then
you
know
literally
the
skin
that
list
yeah
looking
for
control
like
it's,
not
something
that
would
be
I
mean.
Obviously
it's
a
yeah
for
loop
constantly
is
Oh.
Although.
A
From
your
description
have
a
little
worried
that
maybe
you
haven't
I
didn't
see
problem
like
if
you
make
an
object
and
then
you
have
to
go
clear:
the
system
to
find
out
what
the
maivia's
other
thing
that
you
made
it's
easier
for
code
like
that
to
make
two
objects
and
then
code
that
produces
a
deterministic
name.
And
then
you
can
figure
out
what
it
would
have
made
without
clearing
the
system
at
all.
Yeah.
D
C
Will
be
like
in
a
dump
stage,
yeah,
and
you
know
your
system
more
than
I
do.
Obviously
no
the
other
way
happening
is
like.
If
you
have
deterministic
names,
you
could
get
like
crust,
it
would
get
left
in
the
system
that
you
know
sometimes,
like
you
know,
hog
will
take
longer
to
determinate
or
something
like
that
and
someone
who
pushes
an
optic
back
in
again
I
mean.
Obviously
you
do
have
deterministic
names.
It.
D
Pushes
Congress
yeah
like
what's
one
of
the
things
lately
I
think
things
like
we
do
plumbing
the
color
and
stuff
heavily
in
their
status,
every
collision
count,
which
is
ports
which
is
actually
included
in
the
hash
that
it
creates
it's
good,
but
it
uses
to
create
here
as
an
input
to
the
hash
that
uses
to
create
the
names
where
it's
represents.
Yeah
I,
wonder
if
we
can
say
500
an
additional
line.
D
Like
they
deployed
like
those
things
they
adopt
represents
which
you
probably
on
your
controller,
don't
need
to
support.
Looking
you
probably
don't
want
to
add
that
complexity.
No,
it's
very
complexed
and
like
I,
don't
know
anybody,
that's
ever
used
it,
but
apparently
it
exists
and
it's
important
for
some
reason.
Maybe
for
people
to
do.
Is
it
been
how
to
sew
a
thing
at
one
time?
Maybe,
but
they
don't.
You
don't
emulate
that
Felice
and
I.
C
C
C
C
Okay,
we
talked
about
meaning
what
yeah
good
access
to
the
square
described.
The
big
question
was:
what's
the
use
case,
we
use
do
one
something
like
next
to
a
new
system
like
who
is
your
metrics
are
not
aware:
small
objects
in
the
system
and
anything
that
you
want
something
we
we
try
and
take
I,
so
just
much
more
complex
yeah
as
soon
as
we
provide
some
the
only
back
the.
A
Other
major
use
case
that
I'm
aware
for
people
who
want
this.
Is
you
eyes
which
want
to
display
the
tree?
Show
users
the
relationship
between
the
various
optics?
It's
like
I
make
a
make
a
deployment
I
want
to
know
like
these
replicas
sets
to
managing
these
pods
pods
containers
or
or
like
I
have,
and
it
would
be
ideal
if
it
the
other
way
too,
like
I
had
this
CR
Yi,
which
has
this
operator
and
winning
it
connects
all
the
deployments
it
makes
up
to
this
up
to
the
CRTC.
D
Want
to
grab
database
yeah
the
problem
also
like
kind
of
comes
like
you
make
other
things
are
like
relating,
but
not
related
in
the
sense
that
they
have
like
a
direct
relationship.
Ie
selectors
yeah.
So
like
things
like
how
you
what's,
a
member
of
your
service
is
not
like
immediately
knowable,
particularly
if,
like
I
mean
for
for
normal
objects,
like
deployments
and
services,
like
you
kind
of
know,
because
those
have
selectors
because
their
core
objects.
D
A
C
But
even
if
it
was
just
in
that
in
the
library
that
the
controller
could
use
it,
you
could
just
be
like
here's
a
list
of
informers
or
even
send
of
informers
for
different
types.
Yeah
and
here's,
like
my
top
level
I,
don't
know
like
I'm
gonna.
Give
you
an
object.
Give
me
back,
it's
obviously
hard
because
I
think
it's
going
to
be
you.
B
A
A
Yeah,
so
the
topical
goals
I
think
there
definitely
needs
to
be
a
way
to
have
more
than
one
on
C
book,
I
think
I'm
kind
of
so
I.
You
shouldn't
have
to
have
a
command
line
level
access
to
cube
it
guy
server
here
to
set
that
up
pretty
sure.
But
the
question
in
my
mind,
is:
should
we
go
to
a
completely
dynamic
thing
like
a
admission
look,
or
should
we
move
it
into
like
a
config
map,
the
rest
of
where
the
rest
of
the
fights
pretty
every
should
go?
I
haven't
really
thought
that,
through,
like.
A
B
A
A
A
A
A
A
And
is
that
a
use
case
they,
okay,
that
you
have
I
started
as
well,
I
think
later
I'm
going
to
let
go
multi-tenancy
work,
we
group
went
the
part
like
the
isolation
mechanism.
Is
the
Nuge
base
so
yeah?
There
are
projects
that
I've
seen
where
you
can
fit
in
a
namespace
plenty
of
presents
that
I've
seen
that
like
to
have
cluster
cluster
roles,
and
it
seems
like
you
know
if
you
can
fit
it
in
HBase,
some
I
am
Christmas.
Doesn't.
A
Notice
this
illegally
would
want
to
think
carefully
about
just
need
to
make
sure
that,
like,
if
we're
going
to
advertise
namespaces
as
a
as
a
security
boundary
for
admission
up
folks,
we
need
to
make
sure
that
you
can't
do
something
crazy.
Like
change
the
namespace
of
the
hot
incoming
object
right,
yeah
I
was
thinking
about
like
people
making
like
a
custom
researched
that
pretty
much
was
a
wrapper
around
this
and
then
having
a
controller
pick
up
these
customer
sources.
That
I
knew
you.