►
From YouTube: kubeadm office hours 2019-07-24
A
A
So
basically,
the
change
here
is
that
added
some
minor
master
control
main
fixes,
but
the
main
change
here
is
that
now
I
I
thought
that
it's
a
good
idea
to
instruct
the
users
to
not
use
API
server
advertised
address
and
instead
use
a
control,
plane
import,
and
to
achieve
that,
you
can
use
the
work
of
8dc
hosts
on
all
the
nodes
and
eventually
you
can
also
use.
You
know
I
record
since
cname
records,
if
you
prefer,
if
you
have
a
DNS
server
installed.
A
A
B
Hi
guys
just
a
couple
of
quick
updates
from
me
last
night,
I
pushed
out
this
new
PR,
which
basically
propagates
the
dual
stack
feature
gate.
So
if,
in
the
cluster
configuration
you
specify
a
dual
stack
future
grade,
it
gets
pushed
out
to
all
the
kubernetes
components.
I
think
this
was
requested.
I've
also
added
this
PR
to
the
to-do
list.
For
the
the
dual
stack
feature,
please
have
a
look,
I
think
it's
reasonably
straightforward.
The
code
is
pretty
independent.
We
could,
you
know
once
feature
once
dual
stack
is
in
GA.
We
can
take
this
stuff
out.
D
Just
a
quick
comment
with
regards
to
your
tests
that
I've
seen
so
far
I,
don't
currently
see
it
I'm
just
doing
a
quick
scan,
but
if
the
test
is
dependent
upon
dual
stack
and
it
is
feature
gated,
there
is
semantics
of
tagging
that
we
need
to
enforce
inside
of
the
tests.
So
it
has
to
have
a
bracket
feature
:
name.
The
feature
in
bracket
are.
B
B
The
second
PR
is,
is
for
the
end
of
end
tests,
and
that,
if
you
remember
was
that
when
I
looked
at
the
the
current
set
of
tests,
I
had
noticed
that
we
didn't
have
any
tests
for
network
configuration
values,
so
I
added
these
per
two
tests
here,
which
are
single
stack
only.
This
is
just
to
have
this
completeness.
This
is
just
pure
single
stock.
Nothing
to
do
with
dual
stack
necessarily,
but
after
this
PR
assuming
these
tests
going,
I
can
follow
up
with
the
actual
dual
stack
tests,
which
will
follow
a
similar
format.
B
D
A
E
Thank
you.
This
is
really
cool
and
I've
been
are
just
curious.
Do
you
have
any
guidance
on
how
the
do
testing
with
dual
stack,
because
it's
been
really
I've
more
ipv6,
because
finding
environments
that
support
that
has
been
challenging
for
me
personally,
even
just
starting
a
dhcpv6
server,
I,
don't
know
what
segments
I
should
use
since
they're,
so
broad,
so
new
recommendations
from
when
you're
doing
ipv6
testing
so.
B
Yeah
I
mean
that
is
definitely
one
challenging
aspect
of
it.
So
when
I
started
working
on
this,
so
my
kind
of
environment
is
basically
I
create
a
clustered
on
my
laptop
I
use,
big
nerd
for
it
I
have
my
own
set
of
scripts.
It
worked
by
the
way.
That's
a
public
project
if
you
wanted
to
play
around
with
it.
I
could
share
that
link
with
you,
so
it's
my
own
development
environment
with
ipv6
on
I,
guess,
you're,
asking
more
about
like
a
live
like
more
like
a
production
ready
cluster,
no.
E
You
know
I
don't
want
to
bike
on
this.
I
will
send
the
note
to
the
group
and
maybe
start
a
thread
on
discussing
it.
I
just
think
as
dual
step
becomes
more
prevalent
and
there
needs
to
be
more
ipv6
testing.
It
would
be
really
nice
to
figure
out
a
real
simple
way
to
do
that
and
I
would
also
encourage
you.
E
I
know,
there's
some
work
going
on
with
docker
and
I've
thought
about
doing
one
for
VMware,
fusion
or
VirtualBox,
but
maybe
we
could
talk
offline
about
implementing
a
cluster
epi
provider
for
vagrant,
which
would
get
us
all
of
the
above,
just
something
that
would
kind
of
tie
back
into
cluster
lifecycle
and
enable
this
testing
and
be
valuable
for
others.
Yeah.
A
E
There
are,
there
are
just
some
occasions
where
having
a
virtual
machine
over
kind
is
useful
for
testing,
specifically
around
cluster
api,
being
able
to
test
variations
of
the
hardware
and
different
boot
scenarios
so
as
much
as
I
love
kind
than
I
do
there
are
still
occasions
we're
having
a
actual
fully
operational
guest
operating
system
is
useful,
not
to
recommend
me
pretty
of
cases,
but
for
the
when
you
do
encounter
it,
it's
nice
to
have
it.
There
is
an
option.
B
B
F
F
Yes,
so
quick
update
on
on
the
announcement
so
for
the
customized
proposal
and
all
the
comments
are
addressed,
except
one
that
we
can
discuss
after
plate
and
also
the
we
are
in
good
shape.
Also
for
for
the
machine,
readable
output
proposal
and
from
Andrew
and
aid,
there
is
a
general
agreement
that
there
are
no
blocking
comments
on
on
this
proposal,
but
that
there
is
a
the
open
comments
are
related
to
implementation
detail.
So
it
is
fine
at
least
word
for
follow
all
the
people
that
aspirin
in
slack.
F
F
A
A
A
Is
a
pattern
that
is
occurring
in
kubernetes
and
kind
as
well
as
a
good
example,
originally
in
kind
I
proposed
that
we
should
allow
the
users
to
specify
the
hockey
Batum
config,
and
then
you
can
have
basically
defaults
that
if
you
diverge
from
the
defaults,
the
coaster
is
going
on
break.
But
Ben
was
opposing
to
that.
He
said:
hey
we
should
get.
Customized
is
the
standard
pattern
in
the
community,
and
now
we
are
going
to
do
the
same
incubate
iam.
A
D
That's
a
fair
statement,
I'm
concerned
about
being
able
to
detect
this.
When
user
reports
err
read
like
the.
If
a
person
taints
their
environment,
if
I'm
customizing
it
I
think
we
should
officially
say
that
you're
you're
going
off
the
beaten
path.
We
will
do
our
best
effort,
but
there's
no
guarantees
and
if
you
find
a
a
weird
bug
in
your
customization
that
we
will
try
to
fix
it
because
there's
so
many
incantations
that
are
humanly
possible,
there's
no
possible
way.
We
can.
We
can
support
all
the
different
ways
that
you
could
break
something.
D
Maybe
if
we
had
some
tank
record
that
denotes
that
they've
tainted
it.
The
reason
for
customize
is
that
we
could
never
possibly
plumb
all
the
user
configurations
through
the
comedian
config.
It's
it's
literally
impossible
when
we
do
that
the
configuration
explodes
and
we
before
we
went
to
beta,
we
runner
one
of
the
earlier
alpha
phases.
D
We
have
so
many
nods
as
a
result,
because
we
wanted
to
be
able
to
have
a
nice,
some
customization
for
user
experience
for
the
twelve
thousand
knobs
that
exists
inside
of
kubernetes,
so
I
think
there's
a
weird
balance
that
we
need
to
figure
out.
I,
don't
think
we
are
actually
captured,
but
that
balance
is
yet.
F
F
And/Or
component:
let's
call
it
component
config,
while
we
forgot
to
the
sandbox
I,
don't
think
that
the
samples
they
say
I
understand
at
the
points
from
the
Pamir,
but
I,
don't
think
that
the
sandbox
is
bad
for
all
the
users.
We
have
different
kind
of
users,
someone
who
want
simplicity,
they
just
that
they
are
good
with
few
flex.
F
D
One
thing
that
we
did
with
things
like
pivot
is
we
left
it
every
half
of
phases
right
or
we?
We
denote
it,
there's
something
some
way
of
denoting
that
once
you
do
this,
you
are
off
the
beaten
path
and
we
can
no
longer
there's
so
many
ways
that
you
could
break
something
that
we
can
no
longer
guarantee
that
we
can
support
you.
We
need
to
do
something
like
that.
B
F
D
But
it's
a
question
of
how
we
surface
the
UX
for
this
knob
right.
We
were
very
careful
to
surface
the
UX
for
certain
knobs
through
the
alpha
phases.
Sub-Command
right,
this
is
gonna,
be
mainline
through
in
it.
So
it's
different
right
and
I.
Don't
I
haven't
really
given
that
much
thought
lemierre's
point
makes
me
think
about
it
more,
but
we
should
be
careful
about
how
we're
gonna
expose
the
user
experience
for
the
snob
through
the
main
init
command,
because
that's
a
well
supported,
first
class
command.
F
A
A
So
maybe
we
should
stop
doing
that
and
then
users
can
also
use
the
phases,
because
we
have
to
evaluate
whether
applying
this
like
potential
fix
or
I,
don't
know.
If
it's
a
bug
fix
or
a
feature,
then
the
users
can
use
the
phases
apply.
The
changes
to
the
manifest
and
the
upgrade
process
is
going
to
respect
the
local
manifests.
F
I
think
this
is
the
previous
comment
and
I
don't
agree
that
this
can
be
approach
to
the
patches
and
because
we
need
patching
patches
not
only
on
upgrade
but
also
on
in
it
and
then
join
so
patches.
As
a
wider,
let
me
say:
scope
of
application,
then
and
they're
portrayed
logic
that
we
were
discussing
about
and.
A
A
This
is
going
to
reflect
on
the
UX,
so
I
for
the
customized.
Yes
for
the
castaways
I
have
concerns,
like
you
know
it
I,
don't
have
a
good
explanation
for
this,
but
I
think
we
have
to.
We
have
to
get
some
users
during
the
office.
Basically
ask
them.
Is
this
good
like
what
what
are
your
thoughts
about
this?
A
Yes,
yeah
for
the
beater,
we
have
to
evaluate
like
where
we're
going
to
store
the
patches
punches
for
multiple
notes
can
be
different
and
things
it's.
It's
requires
a
lot
of
thoughts
after
the
Alpha
I.
Just
wanted
to
point
out
like
this
is
sure
not
going
to
make
any
sense,
because
it
conflates
with
your
understanding
of
how
we
should
sandbox
the
config.
But
if
you
want
me,
I
can
write
the
little
document
and
promise
that
it's
not
going
to
be
wonder
than
four
pages
that
can
outline
an
alternative
universe,
keep
config
a
cube.
A
File
with
all
the
types
where
we
can
support,
basically
the
user
customizations,
but
to
summarize
basically
I,
was
thinking
something
like
exposing
a
pod
object,
but
not
only
having
a
coaster
configuration
but
also
having
a
control
plane
configuration
per
node
and
it's
it's
doable.
But
if
you
think
that
I'm
going
to
waste
my
time,
maybe
I
should
not
write
this
document.
I.
F
I
think
that
what
they're
proposing
can
sit
into
a
discussion
for
the
next
seven
days
of
API
of
the
component
config
and
so
I'm
definitely
open
to
to
collect
ideas.
What
my?
What
I
expect
in
the
long
term
is
that
having
customized
and
having
also
component
config
implemented
in
other
components,
I
expect
our
component
config
to
become
smaller,
smaller
and
the
relies
on
the
other
fork
when
you
say
to
be
advanced
or
configuration.
A
F
A
F
The
component
config,
is
the
target
the
target
solution.
Now
we
are
using
Astra
arcs
for
the
component.
That
does
not
have
a
component
config,
but
for
let
me
see
the
most
of
the
use
cases.
It
is
enough
to
customize
the
API
servers
and
today
we
they
use
Astra
arcs.
Tomorrow
they
will
use
API
server
component
comfy.
F
A
D
Think
that
part
of
this
conversation
eliminates
that
we
should
be
recommending
customized
for
only
a
certain
set
of
use.
Cases
which
is
like
tried
to
use
it
primarily
for
I'm
guarantee
people
are
going
to
use
this,
so
we
should
come
up
with
guidance
along
with
documentation.
That
PR
is
this
at
the
primary
use
case.
That
I
see,
in
my
mind,
is
the
capability
of
modifying
the
deployment
itself
like
web
authentication
as
a
sidecar,
with
all
the
other
details
associated
with
it
is
a
concrete
use
case
example
of
why
you
use
customize,
but
I
guarantee
you.
A
D
A
F
A
F
F
Just
clean
it
up
to
the
directory,
so
I
have
a
folder
where
I
have
some
pages.
I
have
one
page
for
a
tcd
one
for
the
API
server,
one
for
the
controller
manager
and
one
for
the
scheduler.
The
pages
are
also
very
simple:
they
are
applying
an
animal
Titian
and
the
name
of
the
pages
is
free.
There
are
no,
there
are
no
constraints
on
the
name
that
that
you
can
use.
F
The
important
thing
is
that,
as
defined
by
customizer
each
fish,
the
object
where
each
page
applies
is
defined
by
and
in
a
kind
and
the
name
and
the
namespace.
So
this
is
basically
the
set
up
foreign
private
patches.
I
can
do
the
syntax
will
be
really
simple
or
cubed,
meaning
it
mean
Nino's,
Nino's,
experimental,
customize
or
mean
escape
and
and
and
I
can
pass
the
folder.
F
I
recognized
that
there
is
only
one
page
that
apply
to
their
server
and
use
it
the
same
from
the
controller
manager,
the
same
for
scheduler
and
as
you
can
see,
it
is
not
considering
the
page
shown
on
a
CD,
because
it
is
big
I'm,
not
selecting
the
TCP
manifest,
so
it
is
already
faces
compliant
and
if
I
check
my
manifest
my
manifest
genetic,
manifest
I
can
see
that
the
the
page
are
applied.
So
again,
server
gets
the
right
of
the
right
page.
A
Yes,
sir,
something
interesting,
we
have
to
think
about
this.
How
are
we
going
to
share
common
patches
between
common
notes,
I'm,
hoping
that
we
are
not
going
to
end
up
in
a
scenario
where
we
have
to
create
the
new
feature
called
automatic
copy
of
patches,
because
we,
if
the
user,
wants
to
modify
something
on
a
another
control
playing
notes?
That
is
different
from
the
primary
control
playing
note?
F
F
F
There
are
patches
that
apply
to
all
the
nodes
or
to
group
of
nodes,
and
there
are
patching
that
which
are
specific
for
one
one
node,
and
this
is
a
program
that
probably
should
be
addressed
at
second,
is
that
in
future
might
be
that
we
use
a,
we
can
use
a
CFP
to
lower
the
patches
at
cluster,
and
so
each
joining
node
or
a
per
grade.
Node
can
read
from
the
CRP
on
the
page
that
that
applied
to
the
node
this
long-term
anew,
but
I
I
intentionally
keep
the
kept
simple
for
the
first
round.
A
F
F
F
It
is
applying
it
one
to
a
static
port
on
an
order
until
the
current
load
on
the
current
node.
So
if
you
imagine,
I
can
read
all
the
page
inferred
from
the
cluster
filter
according
to
the
label
of
my
node,
according
to
every
detail
that
I
can
think
of
and
then
apply
only
a
small
set
of
basis
that
I
set
up
that
apply
to
minor.
F
D
The
one
basic
question
I
have
here
is
what
happens
to
the
patch
nails
it
phase
the
genetics,
confident
manifesto
of
all
of
the
manifests,
so
like
your
your
an
init
control
plane.
So
if
one
patch
fails
like,
if
you
you've
already
patched
one
thing-
and
you
know
I'm
moving
on
to
the
next
thing,
what
happens
like
you've
dropped
a
file
from
the
phases
of
permit
rate,
so
the
control
plane
doesn't
doesn't
try
to
apply
them
all
before
it
drops
it.
Where
does
it?
Do
it
incrementally.
F
A
D
F
E
F
A
A
It
properly
I
can
fix
it.
Yes,
this
is
a.
This
is
bringing
something
else
that
I
brought
on
the
cap.
I
wanted
to
get
feeble
feedback
from
the
others
like
what
do
they
think
in
terms
of
shelling
out
to
keep
Cairo
K
compared
to
importing
customized
another
library
in
the
in
terms
of
photos,
we
agreed
that
we
should
share
out,
took
us
to
keep
grow
kind,
on
the
other
hand,
is
importing
customized,
the
library.
So
the
question
here
is
what
we
should
do.
Personally,
I
was
leaning
towards
shelling
out
to
control.
F
F
A
F
Okay
and
I
have
a
second
last
if
we
are
finally,
as
a
last
imitation
that
I
have
found
so,
which
is
a
related
to
JSON
page,
so
we've
customized
out
of
the
box,
you
can
use
a
page
in
JSON
and
and
basically
page
in
JSON
works
into
require
two
things.
The
first
one
is
that
in
the
customize
dot
Yammer
file,
you
have
to
define
the
target
and,
in
the
page
a
while
the
page
is
only
the
the
operation
that
that
change
your
source.
F
Yes,
the
program
proble
is
that
with
the
project
that
we
are
using,
if
ku-band
mean
the
user
is
not
defining
any
customization
at
Yammer
that
that
means
that
that
I
don't
have
a
target.
I
have
a
solution.
A
proposal
solution
for
this
is
that
our
JSON
page,
we
implement
a
little
bit
a
little
bra
raptor
dead
moves.
Basically
the
target
from
ear
here
and
the
page
is
nested
here
if
not
adhered,
is
little
difference
from
customize,
but
I
prefer
this
option
then
asking
the
user
to
define
customization
file
for
each
stud.