►
From YouTube: SIG Cluster Lifecycle - Cluster Addons 20191001
A
B
Did
not,
but
I
did
manage
to
start
playing
with
the
masterbuilt
of
Kubota,
but
it
does
seem
to
work.
So
that's
it
he's
one
good
thing
which,
although
that
was
your
action
item
so
I'm,
sorry,
but
yes,
I
am
working
towards
doing
in
cuffs,
but
I
have
not
actually
funded
together,
but
we're
getting
closer.
We
got
a.
We
got
a
controller
in
two
cops,
so
we
never
had
a
controller
before
so
now
we
have
our
first
controller
on
the
master
branch
so
as
like
a
long-running
kubernetes
crew
builder
controller
than
that
runs.
So
we
have.
C
D
A
A
Had
a
brief
look
at
what
the
you
know,
the
the
scaffolding
looks
like
and
how
big
the
changes
are
in
total,
it's
about
2,000
lines
of
changes
and
I
think
that
would
need
an
expert
to
you
know
teases
of
calm
and
figure
out
like
how
much
of
it
is
stuff
we
can
further,
simplify
and
how
much
of
it
is
actual
code
of
the
coordinates
operator.
I
think
we
can
take
a
look
at
that
online
at
an
offline
I'm,
not
sure
how.
A
A
A
So
I
started
with
lean
I
were
talking
about
this
and
we
saw
that
there's
different
user
roles,
there's
the
user
operator
maintainer
and
there
are
all
kinds
of
different
activities
and
yeah
I'm.
Not
quite
sure
this
is
the
best
format
for
it.
What
I
try
to
get
towards
is
that
we
sort
of
have
a
little
script
like
okay.
This
is
what
the
user
wants
to
do.
This
is
the
expected
outcome.
A
This
is
the
interaction
I
just
used
cube
EDM
as
a
placeholder
to
get
us
started
and
to
make
it
a
bit
easier
to
understand
for
somebody
from
the
outside.
Looking
at
at
our
effort
and
listed
issues
related
to
this
stuff,
that
needs
to
be
done,
I'm
sure
there
are
better
tools
for
this
I
don't
know
if
it's.
A
If
you
all
feel
like
this
is
a
useful
exercise
in
general,
I
just
felt
like
at
some
stage
we're
going
to
reach
out
two
more
people
who
either
want
to
write
operators
or
who
might
want
to
integrate
the
add-on
installer
in
into
their
distribution
of
kubernetes
or
they
installer,
and
it
might
make
sense
to
track
the
stuff
we're
doing
and
what
still
needs
to
be
done
and
I.
Take
it
from
a
user's
perspective.
I,
don't
know.
F
Valuable
exercise-
and
it
will
make
sort
of
talking
about
this-
a
lot
easier
to
folks.
They
think
the
first
question
a
lot
of
people
will
have
with
this
is
like
well.
Why
would
you
introduce
this
overhead
I
mean-
and
this
is
the
start
of
all
those
reasons-
I
would
say
mm-hmm
in
terms
of
format.
You
know,
there's
no
easy
way
to
these
are
hard.
The
the
getting
the
phrasing
in
the
words
right
is
the
really
hard
problem.
Mm-Hmm.
D
A
B
A
C
C
C
We
need
the
cap
right,
so
that's
a
P
1.5
and
then
the
actual
implementation
merging
as
an
alpha
feature,
but
we
don't
have
any
qualms
about
that
being
feature
gated
and
all
that,
but
it's
just
parked
as
a
PT
plane,
five
for
planning
and
so
I
did
begin
working
on
that
as
well.
No
that's
where
we're
at
and
then
I'll
be
joining
the
coop
idiom
office
hours
again,
I
missed
it
last
week,
but
that
is
tomorrow
at
the
same
time
block.
If
you
want
to
hear
some
more
updates.
C
After,
like
figure
out
how
to
use
my
window
manager,
can
everyone
can
see
my
screen
now
play?
Thank
you.
So
it's
basically,
this
is
a
pretty
small
diff
here,
I've
linked
it
in
the
meeting
notes,
kind
of
non-traditionally:
it's
not
a
PR!
Yet
maybe
just
like
go
to
this,
then
you
can.
You
can
see
the
commits
here.
This
is
like
a
thousand
lines
of
vendor,
but
the
actual
patch
itself
is
quite
small.
So
what
is
it?
It's
like
I'm,
sorry,
stat.
C
So
we're
looking
at
41,
+,
23,
+,
155,
plus
155
minus
4,
so
pretty
pretty
small
patches.
This
is
a
clean.
Build
of
this
commit
right
here.
I
am
in
a
kind
cluster
that
it's
just
a
single
node
instance
that
looks
like
this.
Just
has
the
output
paths
of
my
kubernetes
binaries,
as
well
as
some
config
files
mounted
in
and
then
I
have
to
install
git
currently
in
order
to
get
manifests
from
anywhere.
This
is
a
definitely
a
very
notable
dependency
to
have.
C
The
Covidien
team
is
keen
on
using
OCI
images
for
distributing
these
customized
bundles
I.
Think
I'd
also
just
be
fine
using
HTTP
manifest
list
and
upgrading
it
into
whatever,
but
currently
that's
not
supported
in
this
straight
vendor
of
the
Ayana
installer
P
of
C,
so
I
haven't
made
any
modifications
such
as
like
a
support
for
coop
config
and
all
that,
but
this
is
the
pulling
finished
for
a
Reuben
M
config
file.
C
This
is
a
customized
reference
from
the
github
repository
and
sub
directory
at
a
specific
tag,
and
then
I'm
I
found
a
bug
in
the
dry
run,
support
that
I
didn't
realize,
which
is
that
if
you're
dry
running
an
entire
cluster
coming
up,
there
is
no
API
server,
so
actually
running
the
coop
CTL
command.
To
try
and
get
the
dry
run.
An
output
doesn't
make
any
sense
when
the
API
server
is
not
there.
C
So
that's
something
I'll
need
to
add,
but
that
was
a
cubed
cubed
Ian
init
command
that
just
ran
so
could
been
in
an
it
with
some
ignore
flags
for
the
system,
verification
checks,
and
then
you
pass
it.
The
config
file
o
their
component,
can
fix
and
dry
run
mode,
and
then
I
should
have
I
should
have
done
this
earlier
because
sometimes
pulling
the
component
components
takes
a
little
bit
but
yeah
this
totally.
It
does
work,
oops,
config,
Etsy,
good
pennies
admin,
cons.
C
C
C
C
C
C
B
Is
very
quiet,
so
it
looked
like
you
added
a
binding
to
get,
or
at
least
github
or
what
was
going
on
there
and
is
that
in
oh
yeah.
C
B
C
There
we
are
okay,
so
basically
there's
a
few
this,
this
phase
runner
inside
of
cubanÃa,
it's
just
esoteric
knowledge
about
how
kuba
DM
runs
and
how
we
split
up
the
code
and
basically
there's
an
all
phase
inside
of
this
subsection
called
add-ons
and
all
will
run
everything
inside
of
the
phase
list,
and
so
there's
run
add-on,
installer
and
there's
run
dns
add-on
and
run
coop
proxy
add-on
right.
So
there's
three
phases
and
what
I've
done
is
to
separate
these.
C
This
functionality
I've
installed
a
feature
gate
which
can
be
enabled
or
disabled
and
is
disabled
by
default.
So
when
the
user
runs
cubanelle
normally
the
feature
gate
is
disabled,
which
means
that
this
does
not
short-circuit
the
core
DNS
phase
and
the
KU
proxy
phase.
So
these
are
the
previous
hard-coded
implementations
and
then,
when
you
enable
the
feature
gate,
these
things
just
do
nothing.
And
then
this
phase,
which
is
also
always
running,
only
actually
does
something
when
the
feature
gate
is
enabled
all
right.
C
B
C
C
Yeah,
like
there's,
there's
nothing
there,
except
for
just
the
core
control
plane.
So
there's
no
there's
no
cluster
add-ons
that
there's
no
none
of
those
core
add-ons
that
are
installed
because
in
my
add-on,
installer
configuration
I
only
said
to
install
this
like
multi
base
example,
which
just
puts
a
pod
into
the
default
namespace
yeah.
C
But
if
you
were
to
change
this
list
to
say,
write,
install
kubrick,
see
manifest
list
and
install
core
dns
operator
right
then
now
we
can
put
dumb
things
and
smart
things
into
cluster.
Using
this
manifest
or
using
this
add-on
list
of
customized
directories,
that's
very
cool,
and
then
that
is
also
this
works
from
your
present
working
directory.
So
if
you
just
want
to
specify
files
on
your
file
system,
that's
totally
a
thing
and
then,
with
the
work
that
Evan
Cordell
did.
Hopefully
we
can
get
some
OCI
image
support
in
to
customize.
C
C
Yeah
I
think
like
what's
really
exciting.
Right
is
now
you
can
just
like
pass
this
config
file
to
a
coop
in
a
minute
and,
like
your
cluster,
will
come
up.
The
way
you
describe,
which
is
yeah
an
operator
doesn't
have
to
do
anything
anymore.
It's
probably
simpler
ways
that
this
could
be
accomplished,
but
this
I
think
this
is
about
the
simplest
way.
We
can
do
it
using
kubernetes
api
machinery,
since
the
add-on
install
our
library
is
pretty
simple.
C
Yeah
my
next
to-do
item
here
that
I
wanted
to
get
done
yesterday
and
did
not
is
to
get
a
kept
up
since
that's
going
to
put
this
really
on
the
merged
roadmap
for
this
cycle
and
then
yeah,
probably
moving
on
from
the
demo.
If
there's
no
more
questions,
but
the
diff
is
up,
it's
a
clean
build,
so
you
can
play
with
it
yourself
if
you'd
like
I,
don't
have
like
these
development
helpers
pushed
anywhere,
so
this
config
ammo
and
dev
cluster
SH
or
that
I
used
for
kind.
Those
are
just
like
my
personal
tools.
C
Let
me
know
if
you
want
some
help
messing
around
with
this
at
all
or
want
to
look
into
see
on
integrating
the
install
our
library
into
other
other
cluster
managers
and
then
oh
yeah
and
the
config.
The
convenient
config
I
believe
is
supportive,
a
cluster
API
phenom
mistaken.
So
there
are
some
cluster
API
providers
that
should
work
in
bootstrapping.
This
kind
of
thing
now
yeah
Justin
smiling
like.
B
C
Cuz,
that's
like
was
like
Lucas's
whole
vision
with
component
config
right
is
you
could
just
all
these
abstractions
would
bubble
up
the
stack
to
the
highest
level
order
things
for
cluster
management,
get
ops
and
all
that
cool?
Well,
we
have
other
items
on
the
list.
I'm
gonna
make
sure
we
can
get
to
them
in
a
respectable
amount
of
time,
but
thanks
for
watching
my
little
demo.
A
B
C
Yeah
I
think
I'm
like
if
the
vision
for
packaging
continues
to
mature
around
customized
for
core
stuff
that
having
the
HTTP,
client,
OCI
client
and
get
client
available
in
things
that
want
to
use
those
packages
makes
sense.
So
shipping
an
operator,
we
could
have
a
base
image
with
some
of
those
dependencies.
C
B
Yeah
I
mean
I
always
think
we
should
at
least
add
the
support
such
that,
if
you
have
get
in
your
container,
that
we
can
use
it
as
the
source
in
the
same
way
that
you
specified
it
there
yeah.
C
So
I
I'm,
not
sure,
if,
like
just
shipping,
an
OC
I
image
that
contains
all
of
the
base
manifests
by
defaults
is
probably
a
good
way
to
start,
but
having
the
ability
to
point
the
operator
easily
what
they
can
flick.
Config
flag
to
some
other
repository,
sounds
like
something
really
attractive
for
like
a
disaster
scenario.
Right
like
as
a
mature,
an
operator
of
a
cluster
in
a
mature
environment.
I
don't
have
to
build
a
container
I
can
fork
the
repo
and
then
point
the
operator
to
it
or
something
good.
I.
F
C
B
B
C
Actually
thinking
as
well,
that
I
would
like
the
add-on
installer
to
support
that
not
just
a
customized
ref
did
because
you
can.
You
can
point
to
a
version
to
manifest
on
github
so
easily
explicit,
the
comitia
or
a
branch
name
and
and
there's
so
much
out
there
that
can
be
installed
by
a
manifest
list.
It's
kind
of
a
I
was
thinking
even.
C
The
name
is
used
for
things
like
pruning,
so
you
can
change
the
ref
of
things
and
like
know
that
the
add-on
needs
to
be
removed,
but
the
other
thing
I
was
thinking
is
that
we
should
in
soup
we
should
support
inline
patch
list
inside
of
the
add-on
installer
config,
so
that
it
doesn't
require
complex,
like
assembly
of
a
directory
to
describe
modifications
like
you
don't
have
to
fork
it.
You.
D
C
B
B
B
C
B
C
C
It
can
tell
the
difference
because
it
has
informers
for
both
and
then
it
can
see
oh
thing,
that
uprated
got
upgraded
to
my
kind
and
then
it
starts
ignoring
thing,
so
you
can
delete
it
without
destroying
this,
whatever
kind
of
sophisticated
orchestration
but
yeah
I
think
basically,
the
declaration
is
like.
Oh,
my
new
version
of
my
operator
understands
that
if
there's
a
mind
kind
of
the
same
thing
than
ignore
the
thing.
B
F
Another
way
to
maybe
service
that
is,
you
know
we
think
about
those
manifests.
At
this
point
we
don't
have
a
concept
of
this
manifest
is
too
new
for
this
operator,
but
we
may
be
able
to
expose
that
sort
of
mechanism
where
you
know
that
generic
thing
controller
goes.
Oh
no,
this
this
is
looking
for
a
specific
kind
operator,
I'm
just
going
to
keep
my
hands
off
for
now,
so
that's
sort
of
it
could
be
subscribing
to
the
latest
channel
or
even
explicitly
say,
I
want
this
kind
to
be
now.
You
know
version.
B
C
B
C
After
like
tracing
through
all
the
code
and
like
finally
like
figuring
out,
it's
like
you
feel
like
you
do
so
much
work
and
then
the
resulting
patch
is
like
you
know,
180
line
difference
which
is
really
like
a
success
story
about
the
add-on,
installer
library,
I
guess.
But
it's
yeah,
it's
kind
of
cool
to
see
how
thin
the
shaman's,
because
that
means
it
should
be
pretty
easy
to
implement
other
places.
C
C
But
I
promise
you
it's
not
pretty.
If
anyone
has
any
ideas,
I
just
don't
really
know
enough
about,
like
the
difference
between
writing
to
a
pseudo
terminal
and
not
like
some
of
those
UNIX
semantics
ku,
because
I
know
that
things
are
buffered
when
you
put
them
into
like
a
pipe
and
yeah
I.
Just
don't
know
like
if
I
write
a
goroutine,
that's
like
constantly
reading
from
a
writer
like
if
that's
the
right
way
to
do
it.
C
C
B
C
C
Just
there's
a
talk
from
last
year's
cop-con
that
we
did
for
beat
ports
in
the
coca-cola.
It's
a
encoding,
250,000
songs
a
day
with
that
should
be
one
jobs
and
then
the
title
at
the
very
end
of
the
talk
is
encoding.
A
hundred
and
forty
thousand
songs
a
day
with
that
or
with
core
v1
pods,
or
something
like
that.
So
there's
a
some
failure
stories
in
there,
but
I
have.
E
F
C
Like
we
should
we
could
there's
the
sick
cluster
lifecycle
roadmap,
talk,
which
I
think
we
should
try
to
snag
five
minutes
and
just
talk
about
add-ons
and
I
think
we
can
probably
just
sync
up
with
Tim
on
that
I.
Don't
know
what
he's
planning
for
that
session,
or
maybe
it's
more
like
two
minutes
thing,
but
think
that
the
impact
of
add-ons
potentially
for
life
cycling
ecosystem.
C
A
C
C
F
To
you,
Daniel
and
I
think
Sandeep,
who
dropped
but
I'm
sure
to
watch
the
rest.
The
I
was
going
through
to
create
a
cube
proxy
operator,
I'll,
admit,
I
started
doing
it
last
night
and
found
a
few
little
bugs
in
the
plugins
bit
and
I
think
I
have
sufficiently
fixed
up
the
plumbing
so
that
the
output
is
what
we
expect
it
to
be.
F
It's
got
a
PR
open
to
queue
builder
for
that
and
I
see
seeds
Justin
on
that
for
what
we
have
like
a
there's,
a
one
change
in
like
the
core
q
builder
part,
and
then
the
other
changes
are
just
in
that
pattern.
Cons,
one
and
so
I,
don't
know
what
the
review
process
is
for
Q
builder
to
get
that
through
er.
C
A
F
F
C
Do
you
guys
know
how
to
use
kinder,
because
it's
really
really
easy
to
use?
Thank.
F
C
C
Kind
of
like
a
it's
like
a
stripped-down
like
assembly
version
of
kind,
so
by
default,
when
you
create
a
kind
cluster
using
kind,
config
kubernetes
is
not
yet
running,
but
it
has
everything
and
then
like
you,
can
mount
whatever
you
want
into
it
or
build
a
different
base
image.
You
know
that
kind
of
thing
so
and
it
yeah
it
just
starts
up
in
a
little
docker
container.
C
So
you
can
create
like
a
nine
node
kind
cluster
with
a
bunch
of
you
know,
nodes
or
whatever,
and
then
there's
exec,
helpers
and
stuff
so
that
you
can
easily
use
your
your
built
artifacts
to
test
things.
We
use
it
to
perform
end-to-end
tests
for
the
release,
because
we
want
to
be
checking
that
the
release
built
binaries.
C
B
C
Yeah,
it's
basically
like
kind.
Does
that
same
thing,
but
gives
you
the
cluster
ready
to
go
whereas
like
kinder,
is
like
hey,
we're
not
like
really
going
at
all
yet,
but
here's
all
the
infra
it'll
even
start
like
a
load
balancer
container
for
you,
that's
like
not
running
yet,
and
that
kind
of
thing,
so
you
can
test
like
a
che
scenarios
using
some
symbol,
we're
close,
but
that's
a
way
to
like
get
a
cluster.
That's
like
not
running
yet
right.