►
From YouTube: SIG Cluster Lifecycle - Cluster Addons 20190806
B
It's
the
last
last
time
we
met.
We
talked
about
what
a
few
things
we
wanted
from
key
builder,
so
I
went
to
the
key
builder
meeting
to
talk
to
them
and
the
result
was
a
couple
separate
things.
One
was
talking
about
component
config
and
the
context
around
that
was,
you
know
we
think
there's
a
general
set
of
things.
B
You'll
want
to
tweak
about
cube
builder
based
controller,
but
we
also
think
that
each
controller
will
probably
also
want
to
layer,
add
config
to
that
so
I
sort
of
discuss
some
that
use
case
with
with
them
and
and
Sally
and
there's
actually
some
activity
in
the
issue
this
morning.
So
it
is
on
their
radar
to
some
extent,
but
and
if
there
are
maybe
some
more
concrete
use
cases
we
can
provide,
we
give
some
sort
of
straw,
man,
examples
and
I'm
sure
they'd
be
receptive
to
that
I
linked
to
the
issue
in
notes
and.
B
The
other
bit
was
talking
about
pattern
add-ons,
and
this
is
a
thing
that
we've
known
about
for
a
long
time,
is
that
we
want
to
add
a
flag,
cue
builder,
to
spit
out
something
a
little
more
specialized.
In
our
case,
all
the
scaffolding
for
add-ons
and
but
in
general.
The
cue
biller
expects
this
to
be
like
a
common
occurrence
and
solid
mission,
he's
going
to
look
into
sort
of
a
plug-in
model,
though
I
don't
think
there's
any
activity
there,
so
they
just
want
to
poke
it
and
say:
well:
here's
a
plug-in
level
statements.
C
Yeah,
thank
you
for
doing
that.
Jeff
I
wasn't
able
to
attend
them
peeing
last.
Whenever
it
was
time,
then
component
come
the
cluster
API.
Why
don't
you
know
what
I
mean
the
queue
builder
meeting
there?
We
go
I'll
get
there
in
the
end
last
time.
So
yes,
I
will
try
to
attend
this
time.
I
do
agree
with
you
that.
C
Having
a
bunch
of
examples
in
the
code
will
probably
make
it
easier
than
designing
a
generic
plug-in
mechanism.
Some
of
the
words
I
would
argue
thee
for
like
this
is
a
first
step
towards
it.
We
don't
technically
need
to
build.
Integrate
with
coop
builder
right
to
the
coop
builder
is
a
CLI
tool
that
effectively
scaffolds
some
stuff,
but
it
then
uses
a
well
separated
out
controller,
runtime
type
library,
so
we
could
easily
build
coop
builder
add-ons
or
whatever
you
want
composer
add-on.
C
So
whatever
want
to
call
it
and
just
have
it
be
completely
separate
and
I
think
I
think
we
should
talk
about
that
with
them
is
like
what
do
they
like?
Is
it?
What
would
they
sort
of
see?
Is
the
best
option
there,
because
we
do
want
to
unblock
ourselves,
I
suggest,
and
we
could
always
yeah
I,
think
so.
I'm
gonna
try
to
attend
is
it.
It
looks
like
it's
not.
Tomorrow,
though,
is
it?
E
C
F
Just
wanted
to
mention
that,
from
the
SDK
side,
with
queue
builder,
we
I
met
with
them
when
we
were
out
in
San
Francisco
with
the
we
did
like
an
SDK
face
to
face.
So
we
we
met
with
Kuebler
guys
in
Sully
and
Sunil
and-
and
we
also
have
similar
use
cases
with
the
operator
SDK
like
the
ansible
and
helm
operators.
F
D
B
D
B
D
B
B
A
D
D
Ok
I,
the
other
part
of
that
that
I
would
be
really
interested
in
is
that
coup
builder
v2
structures,
api's
very
differently
from
most
crewmen
uneasy,
so
I'm
a
little
interested
in
how
we're
supposed
to
lay
out
the
internal
versus
external
types.
But
we'll
we'll
see
what
solidi
says
I'll
join
next
week
for
sure
I
have
to
does
anyone
have
the
link
to
the
like
meaning
doc?
For
that.
F
D
F
F
F
D
D
D
There
we
go,
so
this
is
sort
of
a
proof
of
concept
of
using
auras,
which
is
the
day's
project
that
talks
OCI
to
try
packaging
up.
Basically,
just
queue
manifests
push
them
around
on.
This
specifically
adds
some
support
for
customized,
but
the
the
really
high
level
goal
was
just
to
have
an
artifact
that
we
can
talk
about
as
like
an
identity
with
an
identity.
So
I
was
hoping
that
this
has
some
overlap
with
Kubb
add-ons
I.
D
Think
the
specifics
of
like
having
a
tool
that,
for
example,
is
here
in
my
name
space
that
doesn't
matter
like
I'm
soaping
that
we
can
figure
out
what
what
experience
we
want
for
this
type
of
thing,
but
the
basic
is
you
have
a
set
EML
files,
you
push
them
up.
This
builds
an
OCI
image,
manifest
with
layer
media
types
that
are
based
on
whatever
the
actual
content
is
I'm,
so
in
this
case,
I
just
have
one
for
Kuby
em
all
and
one
for
is
specifically
the
customs
file.
D
So
the
idea
is
that
when
you're
parsing
is
somewhere
else
later
with
another
tool,
you
have
a
lot
more
contextual
information
about
it.
But,
more
importantly,
you
have
an
identity
to
talk
about
it
with
other
tooling,
so
this
was
directly
in
response
to
an
issue
in
I.
Think
the
cube
add-ons
get
repo
that
suggested
just
trying
this
out
and
seeing
what
it
looked
like.
So
this
is
what
it
looks
like
I
wanted
to
know.
If
anyone
had
any
feedback
on
this
approach,
this
is
essentially
the
same
approach
that
helm
b3
is
looking
at
taking.
D
So
that's
why
it's
really
interesting
for
talking
about
operators
and
in
add-ons
as
well,
yeah
I
had
a
few
questions
when
working
with
this
was,
did
you
add
support,
or
does
it
implicitly
support
like
directories
like
if
I
wanted
to
package
like
some
smaller
or
like
extending
layers
in
in
sub
directories?
D
D
Yes,
I'm
just
gonna
pass
it
that
example
directory
and
that
that
actually
recursos
through
everything
underneath
exactly
yeah
does
it
follows.
Oh
there's
follow
symlinks
I,
don't
know
how
customize
does
that
it
I'm
just
using
the
go
file,
Walker
and
I.
Don't
remember
off
the
top
of
my
head.
If
that
follows
something's,
not
okay.
Well
I
mean
that's.
That's
probably
a
small
enough
detail.
Okay,
yeah.
E
D
So
you
and
I
had
chatted
about
eventually
building
an
extension
to
customize
or
or
just
maybe
just
patching
customize
to
where
you
could
use
this
in
the
resources
list.
Right,
yeah,
so
right,
I
think
there's
even
a
story
somewhere
in
story
in
issue
somewhere
in
the
customize
get
repo.
It
says
like
hey
what,
if,
instead
of
a
git
repo,
we
could
inherit
from
a
base
that
it's
packages
in
exchange,
there's
already
an
issue
for
that
I
think
so.
But
nice.
C
D
Have
it
at
the
ready
I
could
be
making
it
up
yeah
cuz.
All
we
would
need
to
figure
out.
At
that
point,
I
mean
you
have
pushing
pull
of
these
files
been
working
with
the
registry
and
yeah
I
mean
all
you
would
need
to
do
at
that
point
is
just
have
some
kind
of
ref
that
says:
oh
I'm,
not
working
with
the
git
repo
of
not
working
with
the
local
file
system.
I
need
to
go,
pull
a
no
CI
image
and
then
it
should
even
work.
D
Recursively
right,
it
should
I,
don't
know
if
customized
puts
any
limits
on
that.
This
is
probably
the
extent
to
which
I've
played
what's
customized
its
packaging
stuff.
With
this
and
and
testing
it
out,
but
I
think
the
only
limit
is
that
when
so,
there
is
an
interesting
option
and
customized
build
that
I
just
noticed
today,
and
that
I
think
is
how
it
does
the
security
model.
It's
called
custom
is
build.
D
D
Our
uses,
the
Hashi
corp
library,
where
you
specify
the
git
repo
and
then
like
you
can
even
provide
sub
directories
and
all
that,
so
we
would
have
to
invent
something
similar
to
that,
but
for
OC,
I,
interesting,
yeah
I
didn't
know
that
they
don't
support
like
a
struct
right,
which
is
like
hey,
like
I,
want
my
thing
to
come
from
here
and
then
I
want
like
these
details
about
how
to
unpack
it.
D
They
use
like
a
long
string
that
needs
to
be
parsed
in
order
to
get
to
the
right
place
when
I
built
this
in
bash.
Originally
I
was
just
like
parsing
the
pre
six
and
then
matching
it
against
like
scope,
EO
prefixes
I,
just
like
had
a
list
of
them
that
I
exported
and
then
like
I,
basically
had
a
if
statement
outside
of
customize.
D
D
D
Think
so
my
follow-up
questions
about
this
work
would
be
like
what
is
this
particular
groups
opinion
on
its
value
to
the
problems
that
are
trying
to
be
solved
here.
I
brought
it
up
not
just
because
of
the
issue
and
the
Geary
Poe,
but
also
like
I,
believe
it
was
Justin,
give
a
demo
awhile
ago
of
the
coop
builder
declarative
pattern,
which,
in
my
mind,
is
like
also
a
way
to
collect
up
some.
You
know
files
and
apply
them
to
a
cluster
in
a
specific
way.
So
the
part
that's
missing
from
this
is
like
the
operators.
D
Side
like
this
is
really
just
talking
about
packaging,
but
you
can
pretty
easily
imagine
like
an
operator
that
takes
a
reference
to
one
or
more
of
these
and
does
something
useful
with
them.
So
that's
kind
of
where
I
was
thinking
this
related,
but
if,
if
I'm
totally
off-base,
that
would
be
a
good
time
to
talk
about
it.
D
C
I
mean
this
is
certainly
compatible
with
the
ideas
of
the
of
the
add-on
operators
as
a
source
of
the
the
manifests
in
the
same
way
that
it
is
coop
builder
I
have
noticed,
I,
don't
get
why
people
wanna
do
this,
but
I
know
a
lot
of
people
want
to
do
this,
so
I
think
that's
I.
Think
for
that
point
of
view,
like
there's
no
reason
not
to
support
it.
Yeah
I
think
like
there
are
certain
advantages
to
having
containers
right
like
they
are
understood
and
they
are
version
to
an
extent.
C
D
Just
the
existing
infrastructure
portion
of
it
is
quite
valuable
because
when
you
talk
about
using
git
or
directories,
the
get
thing
for
most
people
is
way
more
maintainable
and
it's
gonna
encourage
better
practices.
But
then,
when
you
talk
about
marrying
a
git
repo
with
a
kubernetes
cluster,
like
that's
a
that's
a
very
like
imposing
requirement
on
some
of
these
workload,
some
people
don't
use
git.
You
know
it
probably
has
to
go
through
like
an
enterprise,
security,
audit
and
all
that
but
like
in
order
to
have
a
kubernetes
cluster.
D
You
have
to
have
a
registry
like
it's
just
not
gonna
work
unless
you're
like
pushing
Tarbell's
to
every
node
and
so
somebody's
already
has
the
infrastructure
to
distribute
these
packages.
I
think
is
one
really
big
benefit
yeah
and
the
other
side
is
like
if
the
other
mechanisms
for
like
taking
a
static
set
of
animal
and
or
manifests
and
applying
them
to
the
cluster,
tend
to
involve
mounting
them
into
an
image
of
some
sort.
That
does
something
with
them.
D
So,
like
the
declarative
code,
builder
pattern,
for
example,
and
so
this
sort
of
could
provide
a
way
to
do
that,
a
little
more
generically
and
in
my
mind
the
interesting
part
is
having
an
identity
that
exists
outside
the
cluster,
like
being
able
to
talk
about
a
set
of
things.
They're,
not
local,
to
the
one
cluster.
C
C
I,
don't
know
if
we've
actually
built
it
yet,
but
this
could
certainly
be
another
back-end
I
feel
like
don't
don't
take
my
objection
to
source
and
respect
I'm,
just
like
I
guess
like
if
so,
if,
if
this
it's
not
like,
there
are
valid
reasons
for
doing
so
and
I
don't
see
any
reason
why
this
is
incompatible
with
anything
like
this
could
be
another
back-end
alongside
HTTP
for
pulling
these
manifests
from,
even
in
the
whether
it's
Cube
builder
or
whether
it's
a
declarative,
add-on
pattern.
I
think
that's
I,
think
yeah
great
I.
D
It's
pretty
pretty
thin
stuff
like
when
we,
when
we
talk
about
a
lot
of
the
other
motivations
like
Jeff,
merge
the
patch
support,
and
you
can
imagine
that
you
would
probably
want
to
have
you
know
the
support
for
people
specifying.
Oh
I
want
the
secret
included
in
my
deployment
or
I
want,
and
this
config
map
to
be
user
specified
like
those
kinds
of
things
are.
They
have
supports
for
assembling
the
structures
and
it's
not
templating.
It's
like
actual
indirection
that
is
specified
and
documented.
So.
E
D
D
D
The
the
thing
that
was
interesting
about
using
the
docker
build
was
that
you
could
actually
like
like
go
from
one
customized
package,
and
then
you
could
like
layer
on
top
of
it
using
the
file
system.
I,
don't
know
if
that's
like
really
a
great
abstraction
to
be
adding
in
to
like
customized,
but
it
was
there
and
interesting.
So
yeah
I
did
talk,
sis
and
CI
folks
about
this
one
of
the
they're,
a
bunch
of
different
ways.
You
can
like
put
data
into
noci
image.
The
way
this
particular
one
is
doing.
D
It
is
taking
each
mo
file
and
putting
it
in
to
a
separate
layer,
and
that's
so
that
on
the
back
end,
you
can
sort
them
by
digest
and
get
some
deduplication
if
you're,
storing
a
lot
of
packages
that
happen
to
share
similar
the
same
set
of
manifests
for
some
overlap.
If
you
start
talking
about
layering
like
a
file
system
on
top
and
you
pretty
much
it's
possible
that
you'll
have
some
overlap,
but
it's
a
lot
less
likely
I
think
so
that
was
really.
B
D
D
D
D
D
The
only
other
thing
I
had
was
that
there
was
some
interesting
feedback
from
having
the
add-on
installer
be
reviewed
by
the
koobideh.
I
meant
a
nurse
does
soon
I.
Don't
think.
Ilya
is
on
the
call
here
right
now:
Justin
you're,
obviously
a
cops
maintainer
I,
don't
know
if
you've
had
a
chance
to
look
at
the
add-on.
Installer
POC,
but
we
are
we've
said
before
that.
We're
interested
in
some
shared
implementation
details,
so
I
guess
I
could
probably
share
my
screen.
Yeah.
D
I
refresh
this
page,
so
Fabricio
has
been
like
the
main
person
reviewing
this
for
inclusion
into
Covidien,
and
he
had
a
bunch
of
questions.
One
of
the
main
ones
was
just
like
how
can
kuba
DM
using
the
library
control
the
coop
config?
That's
in
use
and
I
I
wasn't
super
positive
like
what
the
best
UX
for
the
library
is
here.
D
I
think
like
the
most
going
native
thing
would
be
to
provide
like
a
struct
member,
where
you
would
pass
a
coop
config
type,
but
that
requires
the
caller
to
serialize
to
deserialize
the
good,
config
and
toss
it
in
which
is
like
kind
of
arduous
and
like
requires
importing
the
buncha
libraries
and
stuff.
So
then
the
second
thing
I
was
thinking
is
oh
well,
you
could
pass
a
file
name
here
or
we
could
maybe
provide
the
ability
to
do
both.
D
H
I
may
add
something
I'm
trying
to
understand,
which
kind
of
which
level
of
abstraction
they're
Donuts
dollar
we
gave
it
to
could
mean
I'm,
trying
to
make
an
example.
For
instance,
let
let's
think
about
the
proxy
plugins
okay
could
mean
basically
does
two
things.
One
has
asked:
the
user
are
component
configures
and
then
it
passes
to
the
superproxy
config.
G
H
H
What
I
am
trying
to
understand
is
if
Anton,
Estrada,
somehow
I
eyes
to
meet
the
complexity
of
the
or
the
structure
of
the
underlying
at
dawn
or
I,
need,
as
a
cupid
mean
I
needed
to
know
how
the
the
Koopa
proxy
at
dawn
is
implemented.
So
I
needed
to
create
a
page
to
know
the
exact
resource,
name
and
so
on,
and
so
on.
H
That
doesn't
mean
that
I'll,
tied
alkylate
will
be
could
mean
with
the
implementation
of
the
yam
or
form
for
the
Koopa
proxy,
because
if
we
are
to
couple
the
this
is
a
problem,
everyone
every
time,
I
won't
change.
The
other
should
change.
If
the
don't
study,
Mouse
stablish,
an
interface
that
the
cup
was
the
two
things
disease
food-
and
this
is-
and
the
same
applies
not
not
only
for
the
he
need
so
for
the
first
distillation
of
their
dawn,
but
also
for
the
life
cycle
of
death
dawn.
H
H
Having
that
donese
tunnel
will
let
Google
mean
to
easy
switch
that
Don
that
I
installed
by
default?
But
if
I
look
at
this
from
a
Koopa,
the
main
user
perspective,
you
know
they
want
to
call
Kubat
Mini
in
it
and
pass
everything
is
needed
for
getting
the
control
plane
working.
So
the
Cooper
menus
are
should
not
even
know
that
we
mean
is
calling
list
Allah.
D
Yeah
well,
the
user
can
so
right.
Now,
right,
like
the
user,
doesn't
have
to
know
that
Koopa
dam
is
going
to
configure
to
proxy
with
the
proxy
configuration,
but
then
they
can
use
the
Cuban
and
config
file
to
pass
a
new
cube
proxy
configuration
and
then
Covidien
will
plumb
that
in
so
I
think
the
new
UX
would
be
that
they
can
write
an
add-on,
installer
configuration
and
pass
that
in
the
cupid,
cabana
and
config
file,
and
then
all
of
the
behavior
contained
within
that
would
basically
be
executed.
C
Yeah
I
just
add,
I
think
it
is
also
important
to
like
match
the
existing
workflow
so
where
there,
where
there
is
an
existing
cube,
ATM
configuration
file
that
doesn't
specify
the
cube
proxy
addon,
but
does
in
the
add-on
style
and
the
component
conflict
style,
but
does
specify
in
the
directly
in
the
cube
ATM
configuration.
We
would
imagine
that
or
I
would
imagine.
C
The
way
that
would
work
is
prove
ATM
would
sort
of
transparently
convert
that
into
a
cube
proxy
component
config
and
rather
than
executed
directly,
would
instead
execute
the
add-on
tool
as
if
as
if
the
user
had
specified
it.
So
the
user
can
still
specify
you
know
the
the
in
line
in
line
to
proxy
configuration.
They
know
and
love
for
the
options
that
are
available
today
and
they
it
is
executed
using
the
add-on
operator
and
then
when
they
you
want
to
do
more
advanced
customization
or
they
can.
H
We
already
had
mean
and
configure
we
already
have
splitted
in
several
document.
What
is
specific
for
cubed
mean
and
what
what
is
the
Cooper
proxy
component
configure?
So
it
is
already
isolated
from
from
the
rest,
and
the
point
is
okay.
I
have
to
recognize.
This
is
a
proxy
component.
I
have
to
pass
today
at
dawn
for
the
Cooper
proxy,
but
but
my
point
is
I
can
pass
to
the
Tonys
toddler
and
then
that
oh
nice
tada,
let
me
say,
take
responsibility
of
math
this
component
config
with
the
bunch
of
llamó
that
exists
inside
our
Cooper.
H
G
H
H
Yes,
but
I
I
think
that
we
can.
We
can
agree
in
some
more
generic.
Let
me
say
way
to
past
configuration
to
the
dots,
so
the
consumer
can
should
not
be
responsible
for
the
done,
since
instance,
tell
interns
the
rod,
rod
four
to
two
main
programs.
The
first
money
is
the
the
configuration
and
the
second
one
is
the
darshan
further
upgrades.
H
D
So
for
something
like
cárdenas,
the
the
machinery
that
we've
prototyped
is
that
there
is
a
custom
resource
definition
called
accordion
s
and
that
custom
resource
definition
is
capable
of
defaulting
the
config
and
specifying
like
what
service
IP
should
be
used
in
the
cluster.
And
that
kind
of
thing,
but
also
the
operator
is
able
to
deduce
those
things
from
the
cluster.
D
The
so,
for
the
coordinates
add
on
an
upgrade
would
look
like
using
the
add-on
installer
to
install
a
newer
version
of
the
potentially
a
newer
version
of
the
operator
and
absolutely
like
a
version
bump
on
the
CRT,
so
there's
an
existing
CID
in
the
cluster
and
then
like
okay,
we
want
to
move
it
to
you
know
core
DNS
fee
whatever,
like
we
can
change
the
field
value
in
the
CRD
but
custom
resource.
This.
C
H
C
D
H
Now
I
understand
I
understand
the
general
direction,
I'm
only
a
little
bit
concerned
about
the
usability
of
these.
So
if
a
soon
the
equipment
mean
capable
twist
on
any
of
dawns,
the
user
sure
I
at
thorns,
the
user
choice,
the
choice
they
start
I
do
expect
that
there
should
be,
let
me
say,
common
common
pattern
for
studying
those
are
done:
configuring
them
and
upgrading
10,
because
if
each
add-on
is
caught,
the
NSS
its
own
story,
if
could
be
proxy,
has
its
own
story
and
so
on
basically
I.
C
Right
and
we're
defining
we're
using
a
duck
typing,
so
we're
defining
standard
fields,
at
least
for
the
version
and
for
the
versions
that
conversion
status
so
you'll
be
able
to
like
the
version
that
you
want
to
apply
will
always
be
inspect
up
version
as
I
think
the
current
field
references
to
configuration.
Yes,
we
probably
should
have
a
reference
to
the
component
conflict
I.
Think
that's
a
good
idea.
I
I,
don't
know
if
it's
as
important
to
centralize
that
but
I'm
very
open
to
to
that
idea,
and
then.
D
So
something
like
the
config
map,
forked
proxy
could
literally
just
be
called
config
inside
of
the
the
base
customized
base
and
that
that
keeps
all
of
your
references
in
code
and
stuff
like
very
like
minimally
coupled
and
so
that
there
aren't
all
of
these
other
like
decorating
things
that
you
have
to
take
into
consideration
and
that
should
be
per
add-on
right,
yeah
and
then,
if
people
want
to
compose
these
things
into
like
higher
aggregations,
like
somebody
might
have
an
add-on
set
right.
That's
a
customized
er!
Nothing
prevents
people
from
doing
that.
Okay,.
H
D
Does
yeah
I
think
the
the
minimum,
like
kind
of
overlap
that
we're
seeing
so
far
on?
Unlike
what
does
it
look
like
to
work
with
versioning
and
with
installing
and
upgrading,
is
exactly
what
the
add-on
installer
is
about
because
currently
could
cuddle
apply
like
isn't
an
F
for
people?
They
like
a
lot
of
people
who
want
to
install
an
add-on,
have
to
go
manually
edit
files
and
all
that
kind
of
thing
right
now
and
so
upgrading
to
the
customize
there's
allows
people
all
of
that.
Ux.
H
And
the
last
ten
just
to
discuss
the
field
become
the
PC.
Then
the
last
comment
that
I
remember,
which
is
importance
about
our
carpet
environment
and
the
story
about
how
the
usual
repair
ISM
is
environment
for
cupboard,
meaning
in
a
if
it
is
in
an
urban
environment
and
the
and
packaging
manifest
into
image
is
an.
D
C
C
D
H
H
D
I
agree
with
what
you're
saying
I
think
the
minimum,
or
the
first
thing
we
should
point
people
to
assuming
that
customize
accepts
the
patches
for
the
OCI
support
is
to
use
the
images.
But
the
great
thing
about
the
UX
is
that
it
is
flexible
to.
However,
the
the
operator
wants
to
to
put
things
together
so
in
the
air-gap
environment,
there's
probably
other
things
that
people
would
want
to
use
as
well.
That
aren't
images
since,
since
these
happen
to
be
manifests
in
the
cluster,
they
probably
want
to
check
them
in
with
their
other
manifests.
D
H
It
is
the
same
we
are
already
been
during.
We
are
starting
to
vend
or
customize
inside
cupboard
mean,
and
if
it
is
an
earlier
matter
of
of
call
customize
apply,
we
can
even
skip
the
donee
in
its
current
form,
but
but
really
it
depends
by
how
this
contact
with
this
touch
point
for
per
second
finger
and
for
passing
the
effort
we
have
upgrade.
We
will
take
force.
D
D
C
D
C
That
would
be
cool
all
right.
I
will
yeah
I.
Definitely
I'll.
Take
an
action
item
to
like
look
at
that.
I
think
it's
it's
a
good
one.
Yet
because
a
lot
of
these
concerns
I'm
sure
they're
valid
for
cubanÃa,
but
they
don't
necessarily
speak
to
me
as
much
as
I'm
sure
cups
will
have
its
own
challenges,
but
I
think
it
sounds
like
there
are
different
challenges.
I'm.
D
C
So
we
have
a
cups
configuration,
we
have
a
cluster
configuration
which
currently
has
sort
of.
We
saw
working
three
or
four
years
ago
when
we
started
I.
Remember
when,
like
component
conflict
was
coming
any
day,
so
we
basically
embedded
in
a
version
of
component
config
which
took
like
where
it
was
going
and
they're
currently
all
like
in
the
cluster
object.
That
was
a
mistake
which
will
address
over
time,
but
the
the
cop
CLI
tool,
when
you
run
it
it
generates,
manifests
which
are
then
applied
like
literally
applied.
So
today
that
has
some
pretty
big
shortcomings.
C
The
other
challenge
is
suppose,
there's
a
new
version
of
can
we
have
to
the
manifests,
are
currently
baked
into
the
cups
binary,
so
we
have
to
update
cops
and
then
users
have
to
run
cups
update
with
a
new
version
of
cops.
So
that's
nice
in
that
you
know
you
have
stability
of
what
the
add-ons
are,
but
it's
not
nice
and
that
that
is
a
terrible
workflow
for
a
add-on
update
all
around
like
both
for
developers
and
for
comes
users.
Yes,.
H
C
Yes,
I
do
hope.
We
we
complete
our
conversions.
You
know
like
I,
think
sed
manager
is
a
big
part
of
that
I.
Think
add-on.
Operators
are
a
big
part
of
that
I.
Think
cluster
API
is
gonna,
be
a
big
part
of
that
and
then
then
there
will
be
well.
My
hope
is
that
actually
both
cops
and
crew
medium
end
up
is
very,
very
thin.
Shims
over,
like
these
standardized
components
and
yes,
then
there
will
be
the
difference,
will
purely
be
like
the
flavor
of
shame.
C
E
D
Cool
what,
while
you
still
have
your
screen,
shared
I,
have
a
quick
question.
Yeah
one
thing
that's
really
interesting
to
me
about
this-
is
that
I
think
this
is
the
first
time
that
it's
gonna
be
easy
to
pull
an
add-on
that
someone
has
just
published
exterior
to
communities
like
previously
it's
yes,
then
you
bend
it
in
yeah.
So
that's
the
huge
thing
there,
which
is:
what
is
it
opening
this
up
enables?
D
Could
a
team
to
execute
different
opinions
of
clustered
bootstrap
right
so,
like
vendors,
could
do
right
right
exactly
there,
there's
just
some
stuff
that
I
didn't
see
in
this
PR.
That
I
think
are
really
important.
That
I
think
things
like
if
it's
an
operator
it's
talking
to
the
kubernetes
api,
so
it
has
a
requirement
on
the
set
of
API
is
that
are
provided
and
then,
during
an
upgrade
of
either
the
component
or
the
underlying
cluster.
You
have
to
ensure
that,
after
an
update,
it
still
has
access
to
all
those
things.
D
H
D
So
if
I
deploy
an
operator
that
depends
on
you
know
like
absolutely
one
alpha
1
and
then
that
gets
deprecated
and
you
know
coop
117,
then
that
operator
is
now
going
to
be
incapable
of
interacting
with
the
API,
perhaps
yeah.
So
it's
going
to
then
fall
into
a
degraded
state,
hopefully
so,
like
all
of
that
stuff
is
really
ugly.
So
what
you're
suggesting
is
kind
of
like
with
OLM
you
were
able
to
map
the
api
dependencies
right,
yeah,
so
that's--
getting.
D
Think
the
ideal
scenario
is
that
you
are
prevented
from
from
doing
that
and
like
override
it
sure
if
you,
if
you
want
to
but
like
we
know
that
if
an
operator
uses
an
API
that
no
longer
exists,
it's
not
gonna
work
properly
right,
yeah
I
mean
you'll,
get
there
you'll
either
get
the
error
from
the
Installer
or
you'll
get
the
error
from
the
component.
That's
so
great!
D
H
D
G
G
And
one
of
the
challenge
is
always:
how
do
you
even
know
which
API
changes
are
going
to
come
down?
The
second
that
you
update
the
cluster
yep
you'd
have
to
have
something
formalized
right
to
say
this
update,
say
this
work,
you
change.
It's
gonna
affect
these
things
to
be
deprecated
these
things
through
your
move,
these
new
things
coming
I.
G
D
D
I,
don't
like
how
soon
doesn't
like
highlight
the
speaker,
presentation,
yeah
I
mean
I,
guess
the
the
best
you
have
right
now
is
like
the
release.
Notes
right
for
the
kubernetes
version
that
you're
targeting
like
if
you're
upgrading,
API
server,
suppose
yeah
I
mean
you
can
pull
the
swagger
doc
I
guess
from
each
release
and
get
a
good
idea.
That's
pretty
good
I,
don't
know
if
there's
scene,
if
that
captures
everything
I
think
it
does,
but
so
a
if
I
had
to
guess
like
really.
D
The
only
safety
that
we
have
currently
in
the
community
is
that
people
during
the
release
are
testing
these
things
like
with
kubja
DM
and
with
cops,
and
we
kind
of
just
canonically,
know
yeah,
because
we're
involved
in
a
community
like
okay.
Well,
these
are
going
to
be
the
blessed
release.
Paths,
I.
D
D
Installer
I
I
would
assume
right
that
a
tool
would
provide
a
default
add-on,
installer
configuration
and
that
you
could
use
the
tools,
generation,
command
line,
experiences
to
get
the
default,
yeah,
yeah,
and
so
in
someone
familiar
with
the
the
right
versions,
puts
a
list
somewhere
and
github,
or
something
like
that
right,
so
yeah,
yeah
I
mean
I
mean
the
tool.
Will
you
know
like
for
that?
Kubernetes
version
is
going
to
have
a
default
right,
great
yeah,
I,
guess
I.
D
D
We're
the
people
that
wrote
flat
car
Linux
kin,
Volk
like
if
you're,
if
you're
kin,
Volk
or
if
you're
the
open
source,
typhoon
project
right
and
you
want
to
use,
cops
or
ETS,
cuddle
or
Covidien
right
as
part
of
your
workflow,
but
then
distribute
all
these
extra
things.
Then
you
might
choose
to
use
the
add-on
installer
configuration.
D
H
D
Used
to
used
and
make
that
so
that,
when
the
controller
is
packaged,
that
there
is
some
bit
of
metadata,
it
doesn't
sound
like
with
the
current
use
of
customize
and
COO
builder,
that
we
we
couldn't
have
some
object
or
some
fields
in
the
customization
to
suggest.
That
sounds
like
that
would
probably
go
over
well
I
like
that,
because
yeah,
as
things
become
more
sprawled,
and
we
have
individual
components
that
people
are
selecting
or
not
selecting
it's
more
important
to
know
what
api's
are
in
use
by
each
component.
D
Yeah
yeah,
especially
if
you
think,
like
the
typhoon
use
case
I
think,
is
an
interesting
one
like
you
want
to
support
that
use
case,
but
then
I'm
sure
there's
lots
of
people
who
want
to
do
it
like
typhoon,
but
with
their
own
set
of
things
and
like
it
would
be
nice.
If
we
could
help
yeah
the
the
more
shared
bits
people
can
use
the
easier
it
is
to
fork
right
if,
like
I,
could
not.
D
You
know
for
typhoon
in
two
hours
right
now,
because
it
wouldn't
be
possible
to
like
know
the
directory
structure
and
how
things
are
being
applied.
Oh,
no,
that
great,
maybe
it's
a
little
easier
to
search
for
it
out
on
the
Installer
configuration
okay,
well
yeah,
maybe
at
some
point
we'll
we'll
get
to
sync
with
Ilya
as
well
and
ask
him
what
his
opinions
are.
Ets
cuddle
is
taking
an
interesting
direction
right
now,
with
integrating
with
a
git
repo.