►
From YouTube: SIG Cluster Lifecycle - Cluster Addons 20191126
A
Hello
and
welcome
everyone
to
the
class
anons
meeting.
It
is
the
26th
of
November
and
it's
the
tools
cube
cond
meeting
I
was
really
happy
to
see
that
Foca
managed
to
meet
up
and
it
sounded
like
you
had
a
good
discussion
there.
Over
lunch,
Lee
said
that
commedia
took
some
notes.
I
haven't
seen
those
notes.
Yet
does
anyone
want
to
summarize,
or
maybe
say
what
happened
there
or
is
everything
on
the
agenda
already.
B
I'd
say
we
had
it,
we
did
have
an
excellent
meeting,
we
had
lunch
and
then
we
broke
it
afterwards
into
a
meeting
room
inside
because
there
were
too
many
helicopters
flying
overhead.
Like
I
was
another
shortage,
military
stuff
and
I.
It
was
pretty
wide-ranging,
but
I
think
we
basically
reached
a
pretty
good
consensus
around
sort
of
like
a
shared
vision
of
what
we're
gonna
do
we
talked
to
Phil
about
a
very
wide
range
of
topic,
so
I
really
hope
we
can
actually
get
those
notes
to
sort
of
young
people's
memories.
B
I
have
already
is
sort
of
already
gone
into
a
blur.
To
me,
I,
remember,
being
very
excited
by
the
idea
that,
like
controller
manager
is
a
like
bundle
of
add-ons
in
a
way
and
I
thought
that
was
mind-blowing.
They
were
seven
right,
but
everything
is
a
bit
of
a
blur.
But
yes,
everyone
else
has
any
recollections.
I
know.
C
We
talked
about
considering
if
we
could
come
up
with
like
a
common
sort
of
status
API
or
these
sort
of
common
embedded
extensions
that
we
don't
have
concretely
defined
and
trying
to
learn
from
RedHat
experience
with
like
giving
those,
but
maybe
try
to
give
very
crisp
guidance
on
some
of
that
stuff.
And
the
other
thing
I,
remember
is
talking
about
the
OCI
cap.
A
Sorry
I
was
trying
to
type
a
couple
of
notes.
It
looks
like
we
have
a
couple
of
new
people
on
the
call
as
well
welcome
to
all
of
you.
If
you
want
to
quickly
introduce
yourselves
or
add
anything
to
the
agenda
that
has
no
yeah
just
checked
since
the
last
meeting.
We
didn't
have
any
other
action
items
so
we're
good
to
go
there
and
Jeff.
You
had
a
document.
You
wanted
to
talk
about
the
truth.
Everyone
can
see
it
on
sherry
as
well,
and.
C
If
you
request
all
proof,
but
I
need
to
figure
out
how
to
actually
make
it
public
so
hi,
this
is
a
I
met,
venue
and
IP
of
his
colleagues
at
Fujitsu.
At
cube.
Con
Vinnie's
joining
us
in
my
top
right,
corner
and
and
Anna's
below
has
done
some
some
PRS
around
config
map
generation
and
the
core
DNS
operator,
and
what
we
sort
of
identified
as
an
opportunity
to
collaborate
on
the
core
DNS
operator,
and
this
is
something
that
we've
we
right
now
have
talked
very
sort
of
library,
driven
and
very
installer
driven.
C
But
we
have
an
opportunity
to
go
kind
of
use
case
driven
here
and
where
these
these
folks
have
a
concrete
need
to
do.
Zero,
downtime
or
near
zero
downtime
for
dns
upgrades,
so
I
think
it's
a
good
opportunity
to
sort
of
collaborate
around
that
and
learn
from
you
know.
What
does
it
mean
to
use
one
of
these
things
in
production
as
well,
so
I
took
a
first
pass
at
outlining
what
I
think
the
gap
is
between?
C
What
we
have
is
a
sort
of
a
functional
operator
with
you
know,
just
the
bare-bones
stuff
that
we
get
from
the
library
it's
based
off
of
to
what
I
think
would
be
something
that
you
could.
You
could
actually
run
in
production,
the
big
dition
that
we
have
between
open
issues
that
are
just
already
about
the
operator
that
we
have
on
github
today
are
adding
a
smoke
test,
so
something
that
actually
runs
and
exercises
the
operator
providing
at
least
a
metric,
and
if
not
a
couple
of
metrics
emitted
from
from
the
operator
and
I.
C
C
We
we
have
done
a
few
sort
of
one-offs
where
we've
added
we
added
like
I
added
like
the
patch
functionality
that
looks
like
customize.
You
know:
we've
talked
about
adding
config
map,
name
hashing
that
looks
like
customized,
but
I
think
we
have
the
opportunity
to
just
wholesales
take
features
or
take
the
customized
functionality
and
sort
of
integrate
it
to
that
existing
sort
of
declarative
pipeline
and
get
all
that
stuff.
You
know
for
free,
so
to
speak.
C
A
C
From
my
perspective,
like
the
biggest
unknown
is,
is
exactly
what
customized
integration
looks
like
I've
talked
to
some
folks
and
customized
is
now
available
for
API
integration,
at
least
from
a
level.
You
can
get
the
same,
fidelity
that
you
would
get
by
shelling
out
to
customize
that
that's
published
now.
The
actual
internals
of
customized,
like
I,
want
to
config
map
renaming
and
then
I
want.
C
You
know,
name
fix
up
all
the
name,
references
that
sort
of
stuff-
if
you
just
wanted
to
pluck
out
features
like
that,
that's
still
coming
in
customized
before
or
ODOT
for,
but
I
think
it'd
be
interesting
to
see.
If,
if
what
they
have
today,
with
basically
being
able
to
customize
build,
is
enough
and
I
think
we
can
get
some
stuff
to
work
where.
B
C
C
B
C
D
D
C
Looks
like
they've
come
a
long
way
in
in
making
that
integration
possible,
and
they
now
have
a
really
small
API
to
integrate.
I
think
my
biggest
open
question
and
I
want
to
produce,
like
I
plan
on
spending
some
more
design
time
on
this
in
the
next
week
and
I
want
to
prototype
it
working.
But
my
biggest
open
question
is
just:
how
do
we
imitate
the
file
system
to
to
customize?
C
Because
you
know
we
detach
the
concept
of
loading
the
manifest,
because
we
want
to
be
able
to
source
it
from
anywhere
and
customize
expects
a
file
system
layout.
So
if
you
provide
us
a
file
system
to
load
from
it's
easy
to
say
like
okay,
well,
maybe
use
where
all
the
files
are,
but
I
think
that's
where
it
could
get
messy
and
weird.
So
I
want
to
prototype
that
before,
like
really
going
down
the
path
I.
D
Also
believe
that
there's
some
room
like
if
you're
gonna
build
that
ecosystem
integration
for
loading
right
that
some
of
those
should
then
get
upstream
back
into
the
sea
ice
UX
right
user.
Then
we
go
yeah.
If
you
want
a
source
from
an
OC
I
image,
you
should
go
from
there
if
you
want
a
source
from
an
HTTP
source
and
use
that
as
a
resource
then
like
that
has
some
implications
and
they
may
or
may
not
be
desired
for
the
command
line,
but
we
can
do
whatever
we
want.
Obviously,
as
a
library.
E
D
Was
actually
thinking
for
the
add-on
installer
portion
of
the
API
right,
like
we
want
inline
patches
so
that
those
things
can
be
easily
captured
without
users,
doing
weird
stuff
on
the
file
system
or
having
to
clone
repos
and
stuff,
and
also
an
interesting
detail,
is
that
there
isn't
like
a
like
the
git
clone
is
not
built
and
go.
It's
shelling
out
from
the
customized
implementation
internally.
So
you
shell
out
to
customize
and
customize
shells
out
to
get
which
is
fine,
but
just
as
dependencies.
You
know.
A
D
Yeah
I,
like
just
thinking
of
weird
things
like
oh
okay,
well
I,
could
wrap
the
base.
That's
referenced
in
the
API,
like
with
a
just-in-time
file
system
like
real
customized
directory
right
and
then
post
in
the
patches
there
and
build
the
customized
animals
so
that
the
CLI
tool
behaves
properly
with
these
inline
patches
when
you
exec
it
and
it
starts
to
get.
You
know
like
pretty
ugly
right,
and
so
you
can't
just
like
bolt
on
additional
patches
when
you
inflate,
which
might
be
kind
of
a
nice
command
like
flag
honestly.
C
D
D
D
C
D
D
D
You
know
and
then
like
having
to
patch
that
out
and
then
all
of
a
sudden,
it's
like
you,
have
this
very
simple
need
and
while
it's
possible
to
patch
out
it's
like
okay
well,
now
I've
got
to
go
like
four
foot
manifests
somehow.
So
it's
not
super
hard
to
do
you
write
a
bunch
of
you
write
some
files
and
then
it
references
it
might
be
yeah
doing
that.
Programmatically
feels
weird
yeah.
C
E
You're
talking
to
the
people
who
are
in
the
CUDA
operator,
which
is
like
a
meta
operating
for
installment
operators
and
they
have
a
good
investment
market,
so
they're
a
discussions
like.
Can
we
extract
just
your
test
framework
and
make
that
very
clear,
I,
don't
know
what
I'm
in
it,
but
there
they
were
beginning
to
talks
and
I.
Think
that's
where.
C
B
A
C
E
So
this
is
just
a
very
quick
design
doc,
for
maybe
what
is
super
minimal
MVP
on
cluster
representation?
I
don't
know.
Maybe
this
is
something
that
overlaps
with
a
lot
of
things.
We've
been
thinking
about
sort
of
seems
like
they've
already,
not
as
much
but
just
see
what
everyone
else
thinks
is
around.
E
So
the
basic
idea
is
the
first
luck.
That's
all
happened,
this
a
cuter
animal
is
installed,
so
this
doesn't
really
affect
any
of
the
later
installation
partners
I'm.
In
fact,
the
operator
writing
part.
Really.
This
is
kind
of
like
after
it's
all
there.
The
idea
is
that
you
would
put
a
label
on
any
component
of
your
add-on,
so
in
this
case,
in
prefixes,
with
this
label
prefix
and
then
an
identifier,
and
so
you
don't
create
these
objects
of
all
yourself.
A
E
Least,
that's
not
the
like
standard
workflow.
Instead,
you
label
your
objects
and
the
controller
for
this
generates
these
for
you
and
then
at
its
most
basic
level.
Let's
just
select
at
what
the
components
are
and
maybe
surface
their
status
conditions,
and
this
is
just
like
here's,
the
things
that
make
up
without
here's
the
status
of
those
things
at
a
very
high
level
overview
of.
What's
going
on
in
your
cluster,
we
had
some
other
ideas
really
ways
to
extend
this
in
the
future.
E
Like
maybe
you
say,
hey
here's
how
to
extract
some
like
really
useful
metadata
from
the
resources
that
you're
selecting
over,
so
that
you
also
have
that
in
a
status
and
we're
still
like
not
talking
about
writing
these
resources
yourself,
at
least
not
for
this
in
the
shoulder
puzzle.
And
then
you
know,
if
you
want
to
start
doing
more
complicated
aggregation,
maybe
there's
a
way
to
register
data.
That's
another
thing
that
happens!
That's
what
some
ideas
around!
What,
if
I
want
a
tough
little
thing?
E
E
B
A
E
On
this,
our
install
element,
so,
let's
see,
and
how
can
I
know
whether
or
not
they're
healthy,
so
I,
don't
think
this
is
dressed
as
some
of
the
things
you
talked
about
like
give
you
an
overall
red
or
green
or
yellow
agree
on
my
Allen,
that's
something
if
you're
going
to
build
on
top,
but
this
is
like
a
first
step.
Hey
we
already
have
is
all
all
this
stuff
in
the
cluster
and
we
just
cleared
all
in
one
place.
D
Recall
the
so,
this
approach
is
using
this
single
API
group
version
of
mankind
approach
where
we
have
this
general
API,
and
then
we
can
write
a
single
controller
right
on
that
type,
which
is
appealing
I'm
a
little
curious
about
how
we
could
I
guess.
We
could
have
fields
that
are
general
enough
in
here
similar
to
some
of
the
cluster
API
types.
E
My
suspicion
is
that,
like
what
api's
are
included
is
like
AB
standard
interviews
when
I
rapped,
and
so
we
just
if
we
find
the
CID
can't
we
just
do
this,
but
I
kind
of
like
the
idea
of
like
we
can
figure
out
what
stuff
is
important
to
expect
by
letting
people
write
their
own
definitions
of
what
is
important.
Six-Packs.
E
D
E
Yeah
I
noticed
where
you're
going
with
that,
but
I
guess
we
could
have.
You
could
be
a
part
of
multiple
add-on.
It's
great.
So
you
know
I,
guess
two
pillars:
yeah
beeping
I
have
a
the
ingress.
You
know
it's
not
to
say.
If
it
were
is
near
me
because
I
may
be
investigating
you
to
you.
Both
undress
yeah.
E
E
D
We
we
could
say
that
every
add-on
gets
a
unique,
install,
ID
or
any
number
of
things,
or
ask
the
user
to
provide
one
if
they
want
to
differentiate
between
multiple
installs
in
the
ingress
controller.
One
is
the
great
example
of
that
use
case.
Where
often
people
want
ingress
is
for
different
zones
such
as
internal
or
external,
or
something
that's
geo,
sensitive.
D
Very
very
interesting-
and
this
is
a
little
bit
of
a
different
approach.
Then
we
had
that
ad
hoc-
or
we
had
those
in
face-
face
to
face
conversations
about
using
some
of
the
similar
properties
of
the
CR
DS
and
then
something
outside
of
the
DL.
That
would
then
provide
a
user
experience
for
different
CR
DS
right
that
have
things
like
common
fields
and
a
similar
category
or
label
there.
D
B
Certainly,
it
looks
like
we
still
have
what
I
didn't
want
was
I
didn't
want
one
CRD
for
all
like
this
is
complicated,
so
there's
this
has
both
the
CRD
for
custom
add-on
and
upper
addon
CRD,
which
is
what
I
really
wanted.
I
wanted
her
out
on
CRT.
No.
This
is
like
a
pattern
for
like
a
design
pattern
for
like
how
do
we
say
all
of
these
lines
are
instances
or
subclasses
of
a
hast
wrought
on
in
this
case,
and
yet
we
need
to
figure
out
what
your
right
pattern
is.
B
This
looks
like
a
reasonable
good
pattern
for
that.
I
yeah,
like
someone
else
brought
up
like
a
category
thing,
I
think
on
the
CR
DS,
which
I
think
I
was
certainly
mentally
bookmarked
and
will
now
bring
out
the
tab
to
inspect
I
like
the
idea
of
some
sort
of
like
other.
There
are
lots
of.
There
were
like
sub
resources.
E
D
You
run
that
logic.
It
probably
is
a
good
performance.
Consideration
like
if
you've
got
like
imagining.
You
have
a
thousand
add-ons
right.
It
would
suck
to
run
add-on
CTL
and
have
to
compute
statuses
for
every
one
of
those
things
all
the
time,
whereas
if
you
have
a
controller
like
it,
basically
you've
got
caches
all
the
way
down.
B
I'm
starting
to
worry
less
about
the
today's
implementation,
as
it
were
like
we
can
make.
If
we
decide
that
like
this,
we
can
do
something
we
get.
Something
is
possible
today
and
suboptimal,
but
can
be
made
like
much
more
efficient
in
future.
That's
the
sort
of
thing
we
can
fix,
but
but
this
is
also
a
nice
approach
in
that
it
doesn't
require
any
change
that
doesn't
require
that
work.
E
B
To
me,
I'd
want
to
see
the
use
cases
like
race
before
you
know.
That
would
be.
That
would
be
my
like
I'm
just
a
little
unsure
as
to
what
we're
really
gonna
do
with
it
right
like
I,
don't
secured
me
exactly
how
you
just
interact
with
it
from
an
installation
point
of
view
from
a
monitoring
point
of
view,
and
that
to
me
would
be
the
I
in
order
to
answer
that
question
correctly.
I
would
need
to
understand
that
better.
E
D
I,
just
like
somebody,
who's
operated
clusters
like
from
that
kind
of
persona
and
experience
I,
certainly
have
been
frustrated
with,
like
the
number
of
things
that
get
installed
into
coop
system.
It
becomes
like
a
massive
mess
right
like
anytime.
I
want
to
install
some
serious
infrastructure
component
I'm
like
how
can
I
name
space
this
thing
so
I
keep
it
separate
like
from
everything
else,
and
then
that
experience
of
running
helm
list
in
your
cluster.
It
provides
so
much
value
for
somebody
who's
managing
the
cluster
to
like
actually
understand
what
is
running
inside
of
it.
D
D
B
Guess
from
that
perspective,
one
thing
I
bring
up
is
that
this
that
that
case
is
similar
to
the
applications
here,
D
right
and
so
I'm,
not
aware
of
the
exact
state
of
the
application
CRD,
but
like
should
we
name
it
cluster
add-on
or
something
else,
that's
our
open.
Should
we
look
at
using
one
of
those
efforts
or
figure
out
what's
going
on
there
but
yeah?
B
This
I
accept
that
and
like
I'd
say,
the
alternative,
which
is
unproven,
is
to
have
some
method
of
identifying
the
CR
DS
and
use
that
as
like
the
this
year,
DS
that
are
applications
or
add-ons
and
use
them
as
grouping
objects.
But
this
certainly
feels
much
more
natural
in
today's
urban
areas.
World
is
where,
until.
E
With
navigations
Y
orthogonal
to
what
we
want
to
do
with
post
Reynolds,
if
they'd
only
lemonho--,
then
it
makes
sense,
but
that
also
has
its
own
instead
of
like,
along
with
it.
So
yeah
I.
Think
particularly
one
of
the
big
differences
was
that
you
have
to
know
ahead
of
time
what
clients
can
are
composed,
so
you're
going
to
tell
the
application,
steer
you
what
kinds
of
objects
to
go
to
court
and
because
this
is
laborers-
and
this
is
like
we'll
find
in
any
objects
and
it's
similarly
the
GC
controller.
That's
looking.
C
Yeah
I
played
a
little
bit
with
the
applications
here
need
and
developing
declarative
library,
and
the
nice
thing
is
like
when
you
are
the
operator
defining
the
object,
like
you
do
know
all
the
types
that
you're
probably
gonna
touch,
but
at
this
point
like
when
I
picked
it
up.
This
was
like
six
months
a
year
ago,
when
I
picked
it
up.
It
was
like:
okay,
I'll
use
this
API,
but
the
controller
is
not
there.
The
actual
support
wasn't
that
interesting
at
the
time.
Maybe
that's
changed,
but
maybe
it
hasn't.
E
E
E
B
That
would
be
a
great
approach,
I
think
and-
and
we
also
like,
we
don't
have
a.
We
don't
currently
have
clear
guidance
on
whether
our
add-on
operator,
CR
DS,
should
be
namespaced
or
not.
But
yes,
if
we
establish
that
guidance
that
typically
they
are
Singleton's
and
certainly
cluster
scoped-
that's
cluster
atoms
and
therefore
they
have
a
cluster
application
that
that
is
a
nice
approach
but
yeah
than
the
like
figuring
out.
E
Oh
one
thing
he
said,
though,
I
wanted
a
convective
I,
think
that
in
the
past
it
was
possible
and
reasonable
to
write
poster
extensions
that
didn't
affect
local
cluster,
but
I
think
that,
with
all
of
the
features
around
here,
he
eats
as
they
are
now
specifically
more
conversion.
Workbooks
I
no
longer
think.
That's
reasonable!
I
think
that
if
you
have
a
controller-
and
you
can
see
scourging
with
all
of
the
current
student
teachers,
pretty
much
opposed.
E
B
D
Yeah
that
that
whole
cluster
prefix
verbage
I'm
realizing,
is
quite
unfortunate
because
it
conflates
with
the
naming
scheme
of
most
other
things.
D
It's
like,
like
somebody
looking
at
these
API
objects,
all
together,
trying
to
figure
out
what
what
is
the
kubernetes
is
like.
Oh
so
like
is
there
an
add-on
and
a
cluster
add-on
and
like
that?
That's
a
scoping
thing.
E
D
I'm,
just
like
trying
to
be
like
very
careful
and
like
thinking
about
the
architecture
and
like
what
this
will
look
like
externally
to
the
newest
person.
That's
trying
to
operate
a
cluster
yeah
because
yeah,
if
we've,
if
we
have
this
parallel
work,
going
on
with
zigge
apps
and
we're
not
talking
to
them
and
we
build
the
same
thing
sort
of
but
like
not,
and
we
have
all
of
these
objects
and
there's
maybe
like
another
CLI
tool
like
so
now.
D
So
the
very
like
preface
in
any
of
the
work
that
we
have
done
in
this
group,
there
were
already
you
know,
tools.
People
were
already
doing
things.
We
talked
about
potentially
adopting
some
of
those
things.
That
was
a
complicated
conversation,
pick
some
simple
things
to
move
forward
with.
How
do
how
do
we
keep
it
simple,
but
then
enable
the
the
richness
and
the
flexibility
and
extensibility
that
something
like
the
deployments
api
has
done
for
people
running
all
different
kinds
of
workloads
right.
The
deployment
is
three
abstractions
up
from
the
pod.
D
I
have
the
pod
replica
set
to
deployment,
and,
but
nobody
ever
writes
a
replica
set
these
days
right.
Nobody
is
trying
to
deploy
a
bunch
of
web
servers
with
seven
pod
objects.
They're
gluten
the
deployment
to
run
controllers
to
run
batch
processing
things
that
exit
early
to
run
web
servers
to
run
compute
services
with
no
networking
needs
whatsoever.
D
E
E
So
starting
and
then
they're
soaking
every
is
this
the
same
use
case
that
they
are
considering
or
not,
and
another
Sara
Lee
blog
converting
this
clinic.
This
is
probably
like
day
or
two
of
definitely
just
did
something
I
would
play
with
it
and
just
see
like
is
this
doing
what
we
need?
Excuse
me
miss
motivated
by
the
knees
for
both
of
those
two
approaches.
I
think.
D
E
I'm
gonna,
say
yeah,
this
is
late.
Not
this
is
a
stone's
throw
from
stuff.
We
do
another
one,
so
it's
like
we're
literally
just
like
we
don't
want
to
be
off
in
our
island
anymore.
Do
you
want
to
like
contribute
to
stuff
and
that's
kind
of
where,
like
yeah,
we
can
find
this
multi-subunit
other
people
want
to,
and
so
this
is.
Maybe
this
is
a
small
piece
of
it.
E
D
D
Yeah
I
the
mentioning
like
joining
the
Sega
apps
call
I've
personally,
never
really
done
that.
So,
if,
if
somebody
wants
to
take
point
on
that
and
be
nice
to
them-
and
let
them
know
that
we're
thinking
about
this,
then
we
can
hopefully
open
up
the
door
for
any
kind
of
cross
talk
or
collaboration
there,
so
that
we
don't
end
up
with
something
that
nobody's
coming.
E
D
D
E
D
A
D
D
E
D
E
D
D
D
D
A
A
D
A
D
Creating
and
contributing
meaningful
work
in
this
area.
It's
really
appreciated
this
is
this:
is
real
open
source
stuff,
all
the
people
in
governance,
so
we
just
had
coop
con
I
chatted
with
a
bunch
of
folks
about
add-on
installer,
and
this
has
been
where
I've
been
contributing.
The
most
technical
area
of
contribution
for
me
personally
and
folks
are
positive.
D
They
have
positive
reactions
about
like
having
a
component
config
that
lets
people
install
arbitrary
bundles
of
things
in
some
supported
manner
when
composing
that
things
like
cluster
API
and
so
we've
got
this
PR,
which
I
mentioned
in
the
C
cluster
life
cycle,
update,
which
has
the
majority
of
working
implementation
and
there's
a
few
to
dues
that
I've
listed,
but
in
general
this
should
be
a
pretty
good
starting
point.
The
kuben
and
maintenance
team
is
aware
of
this
feature.
We
had
a
triage
call
where
we
decided
not
to
ship
it
in
117.
D
D
D
People
are
gonna,
look
at
it
and
be
like
oh
hey
can
I
use
that
should
I
use
that
and
for
it
to
actually
do
something.
That
is,
that
provides
value
is
exciting,
so
same
thing,
I
could
definitely
use
help
kind
of
with
the
patching
problem.
If
somebody
would
want
a
pair
program
with
me
on
on
the
patching
different
approaches
to
that
would
be
ven
during
the
library
or
doing
that
weird
file
system
thing
we
were
talking
about
earlier
or.
D
There
are
definitely
areas
where
it's
unclear.
What
the
best
way
forward
is
thought
about
using
the
inline
patch
but
developing
the
inline
patch
feature
into
the
API,
so
that
Kubb
idiom
has
a
simple
interface
to
default.
A
custom
version
of
that
patch
when
it
installs
the
API.
What
am
I
saying
when
it
installs
the
add-on
it
can
build
a
in
a
and
add-on.
Api
instance
write
that
inline
patches,
the
qu
proxy
config,
that
kind
of
stuff.