►
From YouTube: SIG Cluster Lifecycle - Cluster Addons 20190709
A
B
B
If
that
is
helpful,
but
otherwise
know
that
progress
should
be
being
made
and,
and
it
is,
it
is
even
easier
now
to
create
one
of
these
atom
operators,
but
always
hold
two
pulls
that
until
later,
everyone
do
a
demo
later,
but
it's
there.
If
we
need
it,
it's
not
free.
We
just
passed
pattern,
equals
add-on
and
magically
be
add-on
appears
or
they
have
an
operator
appears
and
then
I
think
well,
actually,
Lee
do
you
want.
Do
you
want
to
go
next?
B
C
Yesterday,
when
I
started
working
on
this,
but
anyway,
I'm
a
building
the
code
base
right
now
for
just
a
little
Installer
POC,
that's
mostly
going
to
be
shelling
out
and
exacting
stuff
and
gluing
it
together
with
a
cooper,
90's
api
that
has
a
list
of
packaged
graphs.
So
that's
that's
what
I'm
working
on
right
now
that
should
be
PR
to
the
repo
either
like,
maybe
probably
midday
tomorrow,
be
my
guess.
I'm
gonna
be
out
for
the
next
three
weeks
after
that.
Just
reachable
intermittently
on
vacation
and
I
would
love
once
that.
B
B
I'll
just
say
before
this
doesn't
compel
us
to
do
kits
that
IO
api's,
but
it
does
mean
that
if
we
follow
the
patterns
that
they
want,
that,
anyone
that
wants
to
create
a
case,
study
or
add-on
should
then
have
a
much
easier
time
of
doing
so.
So
hopefully,
like
the
dream
scenario
with
is
we
create
a
an
add-on
that
basically
is
pre-approved
and
then
the
any
only
the
additional
fields
need
to
be
approved.
B
B
I
think,
two
weeks
ago
we
discussed
you
know
that
it's
a
possible
security
issue
to
reflect
all
these
things,
that
we
don't
necessary
have
a
clear
use
case,
and
so
probably
we
should
pare
down
the
automatics
or
automatic
status
generation
to
either
just
remove
it
entirely
or
to
use
a
single
condition
for
health,
ie,
healthy,
I.
Think
one
of
the
nice
things
about
conditions
is
it's
also
extensible.
B
So
if
an
atom
operator
wanted
to
put
other
things
in
there
wanted
to
surface
other
things,
there
is
a
mechanism
for
them
to
do
so,
at
least
from
the
API,
and
we
can
build
a
mechanism
into
the
example
controllers,
but
yeah
I
think
that
feedback
is
pretty
actionable
and
reasonable.
So
I
am
currently
quite
happy
with
the
API
review
process
and
yeah.
It
feels
feels
great,
but
that's
that's
what
that
that's?
B
So
like
cops,
for
example,
we
started
using
cops
that
case
that
I,
oh
and
that
is
a
no-no
without
API
review,
so
anything
in
Kate's
that
I
do
and
subdomains
thereof,
yes,
is
subject
to
currently
it's
a
there's,
an
ongoing
discussion,
but
I
actually
thought
it
was
a
nice
thing
to
get
a
clear
review
anyway.
It
feels
like
a
nice
thing
to
have
and
to
like
it,
and
particularly
this
early
stage,
if
we
can
follow
the
patterns
that
the
API
team
believes
to
be
the
coop.
You
know
these
patterns
that
feels
like
a
win.
D
C
Something
that
I
ran
into
regarding
just
domains
is
that
the
official
recommendation
from
the
recommendation
from
component
standard
is
to
use
config,
qetsiyah
and
so
like
I
wasn't
for
for
any
component
configs,
which
I
wasn't
exactly
it
kind
of
feels
like
the
add-on.
Installer
configuration
is
a
component
config
right.
So
would
that
be
that
kind
inside
of
config
Kate's
I/o.
B
And
like
well
be
I,
don't
I,
don't
know
I
thought,
but
that's
certainly
a
good
idea.
I,
don't
know
whether
to
what
extent
the
API
group
should
riff
to
the
group
that
eventually
ends
up
owning
it.
So
if
Cordy
and
s
ends
up
owning
their
operator,
like
does
that
require
it
to
be
either
Cordina
so
Kate's
at
I/o
or
accordionist
at
I/o.
Or
can
it
be
config
duck
a
study
of
/
core
DNS
I?
Don't
see,
I,
don't
or
add-on
sockets
of
a
sh
t,
NS
or
add-ons
a
config
Kate's
that
iota
but
I?
B
B
C
It's
like
almost
like
it'd,
be
helpful
to
have
a
map
to
see
what
ap
eyes
are
used.
That's.
B
B
Mean
the
neighbors:
don't
you
know
they
weren't
actually
used
for
anything.
So
as
far
as
I
know
and
I,
don't
think
it
even
implies.
Like
a
versioning
thing,
you
can,
you
can
evolve
the
versions
of
different
kinds
independently
in
theory,
so
I
don't
think,
there's
any
technical
limitation.
So
we
just
need
a
some
guidelines
to
establish
like
sanity,
like
a
map,
yeah
Oh,
basically,
owners.
B
C
Yeah
and
it
kind
of
affects
our
back.
It
doesn't
he's
the
star
yeah
right,
yeah,
yeah,
okay!
Well,
we
can
talk
more
about
it.
That's
why
we're
an
alpha
alright,
so
exactly
yeah,
I,
didn't
I
didn't
have
much
to
add.
I
did
kind
of
have
like
a
which
was
today.
We
have
any
concern
with
like
a
network
interaction
that
the
Installer
is
going
to
be
doing
and
how
flexible
it
can
be.
If
we're
like,
mixing
refs
between
get
HTTP
and
OCI
images.
C
B
So
making
it
relatively
easy
to
mirror
something
is
good,
is
a
good
idea
and
then
I
think
the
thing
the
other.
So
in
other
words,
if
there's
an
official
hates
repo
of
manifests,
it
should
be
relatively
easy
to
copy
it
to
a
private
web
server,
where's,
three
bucket
or
GCS
bucket,
ideally,
and
we
get
repo
exactly
and
and
then
the
other
one
is
I,
think
I'd
be
careful
about
like
crossing
switching
contacts.
B
So
if
you're
pulling
from
a
HTTP
like
hub
or
an
HTTP
a
public
website,
and
then
you
like,
pull
from
a
local
git
repo,
that's
a
little
weird
right.
So
it's
sort
of
like
the
browser
trust
model
where
it's
like.
You
can't
get
from
up
to
a
file
Wow
anyway
yeah,
as
we
were
discussing
earlier
in
a
different
context.
Yes,.
C
B
C
B
C
C
D
C
B
C
B
C
C
All
the
package
manager
stuff.
It.
B
Is
perhaps
like
you
could
put
it
in
the
bucket?
If
don't
do
that
right
like
if
I
am
installing
a
package,
and
the
package
makes
a
reference
to
my
local
repo,
like
once
I
install
that
package
I'm
already
implicitly
trusting
it
to
a
high
degree
anyway,
on
my
cluster,
like
that,
it
can
already
do
mean
things
to
my
cluster
like.
Is
it
that
much
worse
to
do
mean
things
to
my
local
repository?
B
C
C
B
B
C
C
E
C
B
B
See
share
you
see
me
say
hello,
yes,
okay,
break!
That's
right!
Screen
I
only
got
a
little
bigger,
so
we
have
nothing
up.
My
sleeve
I'm
in
an
empty
directory.
I
do
coop
builder
in
it
as
normal
I'm
using
the
main
example.com.
So
we
don't
have
to
do
the
if
you
I
review,
but
that
is
super
fast
and
create
API,
so
very
much
the
same.
We're
creating
a
dashboard
dashboard
except
we
are
using
the
pattern,
equals
add-on
flag
here.
B
B
B
So
it's
a
Koopa
dirt
is
using
go
modules
as
well,
so
no
more
vendor
directory
yay,
but
that
and
now
so
the
controller
is
running
locally
on
my
machine
and
I
should
now
be
able
to
apply
F
the
example
dashboard
I
need
to
place
system,
because
that
is
not
like
elsewhere
right
now,
and
there
goes
the
the
operator,
so
it
says
successfully
reconciled
it
applied
it,
and
if
we
have
a
look,
ok
get
appointment
and
use
this,
you
can
see.
There's
the
dashboard
and
I
can
also.
B
B
Back
thing
is:
remains
a
pain
so
obviously,
by
running
it
locally.
I
didn't
have
to
do
that
because
I'm
admin
on
my
cluster
locally
we
do
have
to
so
by
Schroeder's
I
guess
we
could
have
a
like
a
something
we
have
do.
We
have
the
place,
but
it
doesn't
have
the
our
back
the
additional
our
back
rules.
We
need
I'm
wondering
whether
we
can
like
look
at
the
manifest
we
creating
somehow
like
extrapolate
the
correct
our
back
policies
or
something
like
that.
Maybe
it's
a
said:
yeah.
B
I
think
I
think
in
the
example
we
in
the
example
in
the
add-on
operators
there's
a
collaborative
atom
patterns
repo.
We
do
like
separate
them
into
three
groups.
We
have
the
ones
for
the
like
for
the
operator
itself.
We
have
the
ones
for
applying
it
and
we
have
the
ones
which
we
need,
because
we
create
an
our
back
policy
because
we
create
a
role
and
therefore
we
need
the
the
permissions
of
that
role.
Yeah.
C
B
True
and
I
don't
know
whether
we
should
do
it
in
the
make
file
or
whether
we
should
yeah
a
lot
play
around
with
that
I
guess
this
this
two-week
period,
while
you
are
happily
on
the
beach
wherever
you're
going
I
will
I
will
be
grappling
with
our
back
yeah,
oh
yeah,
that
you're
right,
our
back
is
that
is
the
big
one.
At
this
point
and
I
think
I
guess.
B
B
Yeah,
it
might
be
that
the
better
it
might
be
that
the
better
workflow
is
to
spit
them
out,
and
you
just
copy
and
paste
them
into
your
go
file
right,
yeah
or
we
could
all
of
it.
We
could
have
coop
builder
if
I
want
to
get
really
fancy.
We
could
have
coop
builder
like
accept
the
manifest
that
you
want
to
like,
create
us
the
first
one,
instead
of
having
an
empty
one
and
then
do
all
that
for
you,
but
I
just
feels
a
little
too
magic
like
yeah.
C
C
C
B
C
D
C
B
C
B
C
D
Honestly
may
want
to
use
just
the
controller
tools
project
directly.
In
that
case
my
light
understanding
of
queue
builders.
It's
wrapping
a
lot
of
the
controller
tools
stuff.
Even
so
not
runtime
controller
tools-
I,
don't
know
about
this.
So
that's
like
the
generator
code
code
generation,
stuff,
you're,.
D
C
A
B
Hey
for
right
now
he's
my
change
in.
Oh,
it
might
neva
choir
a
main
file
as
well
now
into
L,
because
it
like
now
inserts
additional
things
into
the
main
go
he's
using
like
a
marker,
so
it
it
will
now.
It
will
probably
be
more
fussy
in
Kubota
to
l,
and
so
I
would
I
would
scaffold
it
and
then
delete
or
extract
or
copy
or
something
that.
B
C
B
C
But
I
guess
I'm:
it's
just
like
it's
hard
to
get
context
on
what
some
of
these
things
do,
so
I
might
be
misconstruing
and
feeling
like
they're
bigger
than
they
actually
are,
but
I'm
just
like
wondering.
What's
setting
up
all
and
all
of
the
informers
and
all
that
stuff
and
to
me
it
sounds
like
that
would
be
the
manager's
job
right.
I
think
the
manager
holds
them
yep.
B
B
Ducks
updates
yeah
and
or
that,
if
it
is
genuinely
heavy
that
we
like
that,
we
reduce
the
weight
because
I
think
they
would
much,
rather
that
there
was
a
pattern,
come
I'm
putting
words
in
their
mouth,
I'm
I
suspect
they
would
much.
Rather
there
was
a
pattern
component
convict
and
then
he
went
and
built
off
built
up
some
new
like
similar,
but
different
crew
builder
for
components,
thing
yeah.
C
Like
if,
if
we
have
a
COO
builder
pattern
that
like
imports,
things
from
component
base
right
and
then
like
get
some
legacy
flag
stuff
in
there
and
can
generate
an
API,
that's
that
conforms
with
config
gates
IO
and
passes
the
testing
framework
that
we're
writing.
Then
that
would
be
really
sweet
and
probably
also
good
for
any
controllers
that
we
are,
that
we're
generating
from
the
COO
builder
project,
yeah
I,.
B
They're
using
exact
I,
replace
it
with
K
log
I,
replace
it
with
K
logger
very
quickly.
Personally,
the
I
was
gonna
say
the
your
you're
right,
the
coop
owner
generated
controller
should
be
using
component
config.
So
maybe
maybe
there
is
actually
a
like
a
it
shouldn't.
Maybe
shouldn't
even
be
a
optional
like
pattern.
Maybe
it
should
just
be
like
you
should
add
support
for
component
config
or
so
much.
That's
for
for
come
out
a
conflict
to
the
output
yeah.
C
C
B
B
C
B
C
B
C
C
C
D
C
C
So
like
what
a
let's
think
of
something
to
configure,
so
what
about
the
informers,
bolts,
right,
the
polling
rate
or
the
rate
limits?
Oh,
that
stuff
should
be
configurable
in
in
some
area
and
should
probably
be
done
like
for
the
runtime
portion
right.
So
this
is
a
config
API
that
needs
to
configure
the
runtime
of
the
manager
and
I
think
the
manager
is
generated
on
an
it
right.
But
you
need
an
API,
that's
generated
to
configure
the
manager
and.
C
That's
like,
if,
like
in
thinking
in
the
operator
controller
use
case
right
things
like
off
and
metrics,
and
all
that
stuff
I
believe
right
now
they
are
controlled
via
flags
and
then
I
also
saw
like
there's
a
sidecar
thing
in
there.
That
does
like
an
auth
proxy,
so,
like
all
of
that
stuff
right
now
is
like
in
an
operational
concern
that
doesn't
have
an
API,
that's
versioned
or
anything
like
that,
which
is
why
you
need
component
config.
D
There's
a
good
to
see
I
was
thinking
as
you're
talking
about
like
the
security
models
like
one
configuration
might
be
like
a
whitelist
or
blacklist
of
where
you
can
pull
from
you
know.
Let's
say
I,
don't
want
you
to
point
anything
from
HTTPS
only
file
or
something
or
only
get
and
waitlisting
domains
and
I
could
see
that
being
just
like
a
generic
configuration
option
that
you
want
to
expose
that.
C
C
F
C
So,
basically,
it
seems
like
the
the
manager
command
itself
doesn't
actually
have
delegated
off
its
put
into
a
different
component.
That's
then,
like
this
sidecar
proxy,
that
everything
talks
to
the
manager
through
manager
talks
to
everything
through.
That
was
a
little
interesting
to
me
to
see
it
done
that
way.
I
feel
like
it
could
probably
be
done
like
in
the
binary
itself,
but
that
gets
into
a
bunch
of
like
code
generation.
That's
probably
harder
to
do,
and
so
it
just
doesn't
meet
the
immediate
needs
with
project
right
now.
C
Yeah
I
mean
those
are
all
areas
that
I'm
kind
of
curious
about
in,
like
how
we
can
help
and
unify
a
little
bit
more
with
component
config
component.
Config
is
definitely
like
a
cluster
life
cycle.
Api
machinery
joint-venture,
so
I
think
we
all
have
some
skin
in
the
game
there
with
making
it
work
better.
C
Shawn,
sorry
I,
don't
think
we've
ever
met
that
do
you.
You
run
the
ku
builder
meeting
kind
of
yeah,
not
it
yes
kind
of
I
I'm
going
to
edit
it
and
try
to
stay
on
top
of
it,
but
female
and
sorry
are
the
ones
that
are
really
active
and
like
the
children,
time
keeper
there
know
if
in
the
projects
themselves,
so
okay.
C
E
C
Silly
or
stuff
on
charmant
ski,
like
join
those
meetings
ever
ever
slowly
does
defined
if
not
okay,
cool.