►
From YouTube: kubeadm office hours 2020-03-11
A
A
A
A
B
Yes,
I
mean,
if
there's
nothing
else
to
talk
about,
we
could
probably
chat
about
it
a
little
bit.
That
was
basically
some
ideas
about
in
including
some
some
aspects
of
virtual
IP
handling
and
and
load
balancing
for
for
the
for
the
AP
server
into
cube.
Edm
set
up
I,
think
that
was
excluded
at
earlier
stages,
but
could
have
got
interesting
again
now.
I.
A
B
I
mean
I
I
started
off
with
the
usual
suspects,
which
are
keepalive,
t44,
the
virtual
IP
and
and
nginx
for
load
balancing.
But
there
was
somebody
who
has
developed
some
component,
which
is
actually
meant
for
use
in
kubernetes,
for
this
cube
the
IP,
which
does
both
of
them
and
can
be
run
as
static
ports,
which
actually
would
make
setting
up
a
CH
a
class
using
ship
ADM,
rather
simple,
I've
played
around
with
it,
and
it
looks
very
good
I
think
he
won't
actually
join
himself
but
doesn't
seem
to
have
made
it.
A
A
So
normally,
if
we
have
big
proposals
not
only
for
Kapadia
but
kubernetes
in
general,
we
nowadays
employ
the
so-called
kept
process
at
the
key
processes.
Cobre,
this
enhancement
proposal,
I
guess
I-
can
make
use
that
to
just
take
it
enhancements
tips.
Maybe
we
have
dogs
on
here.
Ok,
so
I
want
to
share
this.
A
So
basically,
we
have
a
template,
so
the
idea
is
to
create
a
file
that
follows
the
template
and
provides
detailed
explanation
of
what
you
want
to
add
to
a
certain
part
of
the
project,
whether
Canadian
Coupland
API
server.
The
idea
of
the
depth
I
owed
the
idea.
The
idea
of
the
extra
detail
can
is
so
that
we
can
catch
any
potential
issues
with
the
idea
princess.
You
know
I
immediately
asked
like
what
do
we
do
with
external
networks
and
the
kept
process
is
exactly
for
this
purpose.
B
A
Basically,
in
this
particular
location,
Quincy
has
MIT's
Kip's.
We
have
a
folder
for
sequester
cycle
that
all
medium
and
inside
here
we
have
the
cube
ADM
folder.
So,
basically,
when
you
create
a
proposal,
you
have
to
create
the
PR.
The
PR
template
is
going
to
be
fueled.
For
you,
sorry,
it's
not
gonna
be
fueled
for
you,
but
you
basically
inside
the
PR.
You
have
to
create
a
file
that
is
going
to
reside
in
this
folder,
but
before
that,
what
we
do,
because
we
usually
have
a
lot
of
discussion
around
proposals
anyway.
A
What
we
do
is
we
create
a
Google
Doc.
So
if
you
create
a
Google
Doc,
we
can
go,
everybody
can
rally
there
and
we
can
start
discussing
on
the
Google
Doc
your
ideas.
If
the
idea
is
leaning
towards
implementable,
the
idea
is
leaning
towards
becoming
a
good
idea.
We
can
move
it
to
this
official
process
of
a
cake.
Okay,.
B
D
A
A
A
A
A
A
E
Thank
you
marking
for
being
back
so
thing
is.
That
is
not
abandoning,
but
it
is
something
that
that
we
can
investigate,
because
it
will,
if
we
can
have
this
option
to
offer,
which
is
a
self-hosted
load
balancer.
This
will
make
easier
for
people,
starting
with
could
mean
playing
with
a
a
because
they
don't
have
an
external
dependency,
and
so
this
would
be
an
option
that
we
can
offer
and
support.
E
C
C
Imagine
it
intersects
in
a
very
complicated
way,
but
but
in
some
sense
in
some
sense
we
should
think
about
trying
to
use
those
mechanisms
for
not
hard
coding.
You
know
hard
coding.
An
engine
X
set
up
inside
cube.
Adm
is
exactly
the
kind
of
thing
we
were
trying
to
get
away
from
with
with
the
generic
Adams
work.
E
B
Yes,
I
think
what
I
would
find
most
charming
in
this
respect
is
a
lightweight
solution
like
this
cube
VIP
that
was
proposed
on
the
the
slack
channel
yesterday
and
actually
works.
Quite
fine
might
not
be
what
everybody
wants
to
use
and
productive
environments,
but
anyway,
we'll
help
setting
up
H
a
cluster
incredibly,
and
since
it
has,
it
is
very
simple
to
configure.
B
We
could
probably
embrace
this
somehow,
and
apart
from
that,
of
course,
everybody
should
be
free
to
set
up
its
own
h.I
environment
I
mean
eventually
IP
handling
and
load
balancing,
as
we
have
been
doing
before,
so
they
I
don't
see
any
contradiction
on
that,
but
we
have
to
have
to
see
how
how
deep
we
have
to
integrate
changes
in
order
to
get
these
configurations
and
done.
The
rest
I
mean
the
integration
of
the
static
ports
is
something
that
we
already
have
for
the
etcd.
It's
the
same
mechanism.
A
A
A
Something
asserted
that
the
others
maybe
Brian
is
going
to
be
interested
in.
This
is
something
that
we
are
looking
at
from
the
adonis
tower.
That
is
currently
not
supported
this
air
gapped
environments.
There
is
no
way
for
us
to
pull
the
images
that
we
need
from
the
basically
the
others
that
the
user
was
faced
out.
A
C
F
E
Can
try
to
answer
so
today?
Basically,
we
have
process
for
spinning
up
each
a
cluster
which
could
admit
that
there's
a
requirement
which
is
brain
day
or
or
load
balancer
and
and
then
you
can
spin
up
a
cluster
using
command
means
in
it
and
cubed
me
enjoying
Minos
control
play
which
basically
create
a
second
node.
E
Your
own
load
balancer
is
some
figure
that
makes
difficult
form
for
the
user,
especially
for
the
new
ways
to
test
it,
and
so,
if
they
want
to
test
a
chain
today,
they
can
use
kind
whatever.
But
they
are,
let
me
say
proudly
that
gives
you
a
cast
are
out
of
the
box
and-
and
you
cannot
understand,
internals
out
your
should.
F
E
Yeah
today,
the
attachments
really
simple
give
me
a
stable
address,
a
stable
ID
address
and
that
then
shoot
a
lot
balance.
So
the
structuring
is
really
simple.
Is
there
what
we
are
discussing
it
from?
My
understanding
is
to
provide
the
user,
our
mechanics
so
a
process
to
to
get
a
working
cluster
done.
C
C
G
A
F
F
C
G
C
To
me,
that's
next
level
right,
cube,
ATM
is
supposed
to
be
the
component
for
installing
I,
actually,
the
whole
the
whole
load,
balancer
think
could
conceivably
be
a
cluster
API
discussion.
Yeah.
A
High
level
from
qadian,
even
it's,
the
problem
users
are
having.
Nowadays
they
don't
know
what
you
don't
wanna
start
to
pick
and
we
really
don't
have
a
good
recommendation
for
them.
I
think
HT
proxy
is
becoming
pretty
good
at
this
point,
but
then
again
we
like
we
should
do
even
what
we
were
using
in
kind
as
a
proxy
I.
Think
I.
E
But
this
does
not
prevent
us
to.
Let
me
say
to
give
an
easy
part
to
the
user
for
development
purposes
or
or
whatever.
So
the
principle
could
mean
is
being
simpler,
being
taking
care
of
the
bootstrap
and
being
are
not
agnostic
to
infrastructures,
ni,
si
si
and
CRI,
and
so
given
in
this
principle
with
which
we,
of
course,
we
are
not
forcing
everything
that
can
make
the
user
live.
Simpler,
is
well
ascertain
the.
A
F
A
G
A
It's
something
that
we
should
move
to
the
operator,
and
hopefully
we
can
have
it
one
day.
We
could
also
move
this
logic.
That
is
correct,
incubate
them
to
the
operator,
so
both
Kousaka
and
operator,
sorry
to
Costa,
Rica
incubate
him
conclusion
the
operator
instead
of
importing
the
migration
library.
D
G
A
Takes
a
lot
for
working
with
this
I
think
is
the
right
thing
to
do
for
us,
because
so
originally
I
requested
for
the
coast
repair
that
they
should
just
execute
the
cube
idea
binary
with
a
certain
face
and
they
can
get
the
latest
update
of
the
core
file.
That
is
the
crucial
part
here,
but
apparently
there's
a
problem.
Constipated.
They
didn't
want
to
execute
the
Canadian
binder
in
a
container,
so
they
decided
to
use
the
library
from
source
code.
A
A
A
A
A
So
this
is
part
of
the
problem.
The
other
problem
here
is
that
basically,
Rafa
wants
to
be
able
to
feel
the
dash
dash
config
to
upgrade,
but
sousei
feel
a
Dutch
configured
upgrade
and
they
basically
want
to
modify
the
tomato
Senegal
Rossi.
Please
correct
me:
they
want
to
modify
the
CD
image
during
upgrade.
D
D
It
may
also
be
a
further
description
of
what
the
HDD
version
is
inside.
So,
for
example,
it
may
be
three
point
four
point:
three
alpha
dot
3
or
something
like
this,
in
which
case
it's
an
extension
of
the
HD
version.
So
the
case
is
a
little
bit
complex
here
and
we
need
to
probably
think
in
those
terms,
rather
than
just
try
to
interpret
the
image
stack
as
a
smarter
version.
D
So
I
think
that
what
we
should
do
is
instead
of
trying
to
hard
code,
the
image
stack
into
the
cursor
configuration
well,
not
hard,
hard
coded,
but
rather
like
hardwired
at
opponent
grade.
We
should
probably
try
to
make
things
a
little
bit
better.
We
will
cache
what
our
tech
expectations
for
the
HD
image
are,
and
we
should
allow
for
some
room
of
additional
specification
there,
so
you
can
basically
do
attack.
That
says,
for
example,
three
point
four
point:
three
rusty
or
something
like
this.
E
E
So
I
think
that
it
will
be
interesting
to
manage
these
suffix.
But
the
problem,
however,
is
that
how
do
you
manage
upgrades
and
change
the
suffix,
because,
as
soon
as
the
user
start,
let
me
say
deviating
from
the
version
number,
which
is
means
by
Cuba
the
mean
it
is
not
clear
how
you
you
can
you
can
manage
up
a
trave
and.
E
Would
like
to
see
the
problem,
and,
let
me
say
spec
at
the
out
and
my
probably
I
have
to
sync
with
Rafa
with
this
as
a
whole,
because
I
I
pretty
sure
this
not
only
a
problem
of
it.
Ethnicity
and
I
would
like
to
see
a
consistent
solution,
and
probably
this
will
be
something
that
impacted
BIA
as
well.
E
A
A
A
Sorry,
if
we
compare
the
sed
image
from
a
coastal
configuration
to
the
HDD
image
in
the
latest,
cube
ATM
binary
and
both
of
them
contain
revisions
such
as
0
1.
We
can
safely
perform
the
blade
between
revisions
correctly
cube,
a
the
end
results
for
that
and
I
think
we
should
support
at
least
the
case
where
we
have.
Basically,
we
track
these
revisions
from
GCR.
But
for
the
other
case,
however,
where
damage
can
be
completely
with
a
completely
custom
metadata
I
think
we
should
probably
just
not
upgrade
a
CD.
G
A
D
A
Mean
I
personally
am
fine
as
long
as
we
support
this
case,
which
is
if
we
have.
If
we
see
a
some
sort
of
a
critical
bug
in
LCD,
there
is
not
bound
to
the
HD
version.
Maybe
something
is
bogus
in
the
image
itself
and
we
do
an
image
revision.
I
want
to
sing
qadian
to
be
able
to
upgrade,
like
the
rest
of
the
cases.
I
think
are
kind
of
less
important
for
me
at
the
moment.
Personally,.
D
D
G
A
D
A
D
Yeah,
that's
some
action
work
on
an
issue
that
I
had
created
some
time
ago
and
basically
I
haven't
actually
had
a
PR
since
I.
Don't
have
pretty
much
any
time
these
days
but
yeah.
The
idea
there
is
to
lay
down
some
work
to
enable
easier
machine,
readable,
put
for
the
print
available
upgrades
function.
A
A
A
A
E
A
A
A
A
A
A
A
A
E
E
You
studying
things
on
an
existing
cluster,
but
but
yeah
so
going
back.
I
think
that
the
here
from
the
main
concerns
of
team
is
that
in
the
community
that
is
not
so
clear,
cone
consensus
around
customized,
and
so
probably
we
instead
of
jumping
game,
meaning
customized
as
we
did
before.
Probably
we
have
to
be
open
and
become
an
agnostic
to
this.
E
A
F
A
E
So
the
distantly
so
I
think
that
the
main
difference
is
that
customized
letters
on
another
layers,
another
lace
and-
and
so
you
can
apply
transformation
on
existing
yam.
The
while
ytt
is
use
a
different
approach,
but
that
process
not
so
simple.
So
with
ytt
you
can
use.
Let
me
say:
template
like
a
go
template,
so
we
play
so
there
are
placement,
etcetera,
cetera.
You
can
have
loops
if
conditional.
So
it
is
a
full
language,
templating
engine,
but
it
also
support
overlays.
E
So
it
is,
let
me
say
something
in
the
me
in
the
middle
between
customizer
and
go
templating,
and
it
is
also
Yama
Yama
know
aware,
so
he
knows
the
symptoms
of
the
Yemen,
so
it
has
some
strength.
For
instance,
who
is
the
difference?
Big
difference
with
them?
Is
that
ytt
fully
understand
early
Yama
syntax?
So
you
don't
have
to
worry
about
space,
egg
and
indentation
whatever,
because
it's
fully
understand
the
syntax
but
yeah.
To
be
honest,
I
I'm
not
fans
of
any
templating
engine
that
they
have
ever
Pro
and
cons.
E
So
I
would
like
to
divide
the
discussion
into
layers.
Is
one
is
okay?
What
is
the
best
rate
engine,
and
and
in
this
field
everyone
can
opinion.
The
second
one
is
what
kind
of
integration
we
have.
It
could
mean
for
complete
engine,
and
this
is
the
one
that
I
most
care
about
so
properly.
It
is
interesting
too,
look
at
what
are
the
cupboard
mean
needs
in
thermo
from
a
customized
from
a
complete
engine
and
see,
then,
if
the
template
engine
out
there.
A
A
We
only
don't
support
them
for
mutiple
upgrades,
because
we
don't
have
phases
for
upgrades,
but
Tim
said
that
hey
if
the
v1
pod
spec,
that
we
have
incubate
them
changes,
this
is
going
to
break
the
punches
of
the
users.
I
said
okay,
but
if
we
include
templates
for
a
static
pot
like
first
of
all,
what
are
we
going
to
template?
I
have
no
idea,
because
the
pod
spec
has
so
many
fields
like
what
should
be
exposed.
E
A
A
E
A
E
D
I
see
that,
like
where
you're
going
for
p2,
we
can
actually
think
on
something
like
keeping
the
cluster
configuration
mostly
the
same,
but
then
we
can
also
have
an
instance
specific
configuration,
which
is
basically
something
that's
probably
stored
in
the
local
mo
file
from
each
node,
and
it's
probably
containing
stuff,
like
that's
missing
right
now,
for
example,
what
what
is
the
state
of
the
certificates
on
the
local
note
and
stuff
like
that?
Yeah.