►
From YouTube: kubeadm office hours 2019-09-04
A
A
B
A
A
All
right
so
I
wanted
to
start
with
some
PSA
PSA,
so
I'm
going
to
go
through
these
pretty
quickly.
First,
one
is
that
there
seems
to
be
a
booking
test
grid
where
some
of
the
tests
we
run
appear
to
be
stale.
I
have
not
run
today,
but
the
reality
is
that
they
are
running
is
just.
There
seems
to
be
some
sort
of
a
buck
intense
bit
and
I
already
notified
suggesting
about
this.
A
A
Another
PSA
is
that
so
we
a
bumpy
coordination,
it's
adidas
cycle,
Kapadia
McMaster,
is
already
using
core
DNA's
162
and
takes
or
a
Johnson
tip
for
this
a
particular
work
who
helped
with
an
integration
of
the
migration
of
core
files
incubate
them
during
upgrades.
We
have.
We
first
started
using
151
5.0,
but
then
we
decided
this
banner
to
move
to
this
new
version,
which
has
some
bug
fixes.
B
A
Yeah,
something
we
have
to
discuss
probably
with
Chris,
is
that
eventually
Cardenas
has
to
move
to
another,
maybe
add-on
operator
we're
not
sure
yet
to
my
knowledge.
Chris
already
started
working
on
this,
but
we
maybe
we
should
book
a
separate
session
about
this
and
talk
like
what
is
the
future
of
the
code
for
Cody,
honest
that
is
currently
to
medium.
A
C
C
A
A
A
A
A
Okay,
a
lot
a
quick
PSA
with
me
for
bridge
to
sell
a
couple
of
beers,
so
hopefully
the
flake
our
kind
thing
kinda,
taste,
jobs.
I
said
one
for
testing
for
Brazil
well
for
kinder,
because
the
release
team,
but
constantly
being
honest
about
our
flake
in
it,
the
flakiness
of
our
upgrade
signal
and
also
I.
Guess
we
have
a
flaking
the
kind
jobs
as
well
for
I
had
a
quick
question
here.
Are
you
on
the
call?
Yes,
yes,.
E
A
E
Is
a
workaround
for
because
we
saw
flakiness
so
so,
basically
how
them
creative
works.
It
works
that
we
do
the
upgrades
in
order.
First,
the
console,
the
initial
control
plane
did
the
secondary
control
planes
and,
and
then
workers.
It
is
a
sequential
now
in
kinda
what
happened
that
when
I
upgrade
or
the
second
or
the
secondary
control
plane
or
the
workers
I
access,
the
I
try
to
assess
the
cluster,
and
sometimes
we
get
announced
well
from
the
IBI
server.
It
basically
says
that
the
the
nodal
has
no
grains
back
against
for
reading
the
config
maps.
E
E
Don't
expect
that
could
be
that
the
API
server
should
give
this
answer,
because
they
are
back
rules
are
there.
Probably,
it
is
something
that
happens
while
the
did
not
controller
or
or
I,
don't
know
something
in
the
API
server
is
starting,
and
so
we
we
have
this
fast
answer
or
in
the
API
server,
and
this
problem
basically
solves
by
itself
after
seconds.
E
D
D
E
C
E
We
don't
take
a
report
from
the
user
about
this
problem.
This
problem
happened
only
in
kind
kinder
when
we
are
so
when
you
are
using
docker
in
Dhaka,
which
is
lower
than
other
environment,
and
when
we
are
doing
the
operation,
let
me
say
automatic
in
a
tight
sequence,
so
at
this
stage
and
I'm
fine
in
having
the
workaround
on
driving
to
improving
the
workaround,
but
as
soon
as
we
have
a
report
from
the
user,
we
have
to
investigate
properly.
D
A
Right
I
have
also
not
seen
this
when
I
test
walk
early.
It's
a
very
weird.
You
know
it.
It
only
happens.
Sometimes,
okay,
maybe
we
can
get
back
to
this
system
at
some
point,
alright,
so
another
PSA
that
all
of
our
dogs
fear
surpassed
technical.
If
you
take
a
footbridge
so
for
reviewing
my
changes,
we
are
now
waiting
for
the
signal
scheme
to
apply
approval.
A
We
have
another
flake,
and
this
is
you
remember
the
PR
that
I
sent
what
concurrent
join
for
control
play
notes
with
regards
to
you
know
having
an
exponential
back-off
for
EDD
when
we
are
the
new
members
I
yesterday,
I
did
something
like
10
at
least
10
runs
of
creating
a
coaster
in
kind
with
three
control
plane
nodes,
which
means
that
the
after
the
first
one
is
created.
That's
the
secondary
to
control.
Plane
knows
try
to
join
concurrently
and
I
saw
one
flake
out
of
10
attempts,
but
I
didn't
get
back.
A
C
A
C
One
of
the
things
that's
coming
for
things
like
cluster
API
in
v1,
alpha
3
is
the
notion
of
control,
plane,
set
and
ordinal
numbers.
So,
if
you
think
about
staple
set,
stateful
sets
only
allow
you
to
Drew
to
join
one
at
a
time
in
a
staggered
fashion
and
they
will
support
concurrency
right
so
that
it
will
be
like
zero
one.
Two
and
we'll
do
the
same
thing
for
for
cluster
API.
So
I
don't
know
if
this
is
really
a
big
concern.
A
C
E
Plain
to
the
ins
way,
I
want
on
only
to
add
that
in
a
CCD
they
are
adding
to
the
layer
mode,
and
this
probably
will
solve
the
problem
and
as
soon
as
it
is
in
3.4
as
soon
as
we
have
these,
we
can
change
the
how
the
new
LCD
member
is
added.
So
we
add
a
new
TC
remember
as
and
any
to
Israel
switch
automatically
to
to
mustard,
no,
the
after
it
is
alignment.
This
probably
will
solve
these
issues.
E
It
is
not
enough
34
because
in
three
four
there
is
the
Lebanon
mode,
but
there
is
no
there
automatic
switch
to
master
after
the
emember
is
as
complete
its
alignment.
So
you
have
to
check
if
there
is
some
watch
that
we
can
rely
on.
Otherwise
we
are
to
expect
that
that
they
allow
the
automatic
trader
from
the
learner
to
to
remember.
A
E
E
E
A
A
G
A
Push
one
for
the
walk
around
I
already
said
that
peer,
that
is,
that
was
fairly
similar
to
this.
One.
I
also
tried
to
fix
this,
for
the
upgrades,
which
was
it
didn't
work,
but
for
join
I
am
definitely
seeing
way
too.
Many
requests
for
this
already.
We've
been
delaying
this
like
optional
to
proxy
for
way
too
long.
I
think
that
we
can
still
claim
that
this
supported
cluster,
because
updates
are
not
going
to
work,
but
showing
a
warning
instead
of
erroring
out
seems
reasonable
to
me
at
this
point.
E
G
G
E
E
E
A
E
E
A
A
A
E
Some
monitor
is
interested.
I
can
also
even
over
to
you
or
provide
detail
on
what
was
done.
The
the
to
initially
read
me,
it
was,
was
updated
and
that
I
information
about
how
it
works
inside
the
coolant,
the
cluster
API
workflow.
So
it
is
I
think
that
it
is
relevant
for
for
this
team,
because
Custer
API
is
becoming
one
or
our
main
customers.
E
E
Okay,
the
need
show
file
it
by
team
about
the
convenient
character
and
taking
the
opportunity
that
we
are
at
the
end
of
the
cycle,
and
there
are
no
negativity
to
complete
I
started
to
try
to
write
down
some
idea
about
the
the
what
can
be
the
quad
min,
operetta
and
and
also
implementing
I,
asked
my
Osmo
QC.
So
I'm
writing
down
this
dog.
These
notes,
in
the
form
of
pepper,
steel,
a
whip,
and
maybe
then
the
next
week
or
the
week
after
I,
will
try
to
demo
some
figures
into
the
team.
C
That's
great
I'll
try
to
review.
It
is
if
you
can
send
a
link
to
this
in
the
main
slack
channel
and
notify
the
cluster
API
folks
too,
as
well,
because
there's
there's
a
common
problem
that
exists
where
people
will
want
mutable
and
immutable
infrastructure.
Like
there's
a
weird
balancing
act
here
right
where
some
people
they
cannot
do
a
mutable
right,
maybe
they're
an
on-prem
scenario
and
in
a
pure
on-prem
scenario.
C
If
you
have
a
thousand
node
cluster
and
you
find
a
security
vulnerability,
the
notion
of
being
able
to
roll
that
sucker
out
on
an
immutable
infrastructure
is,
is
a
incredible
amount
of
churn
on
the
workloads.
So
an
operator
is
super
important,
so
notifying
the
cluster
API
group
that
you
have
this
cut
and
they
should
take
a
look
at
it
is
important
to
them
and
I
think
if
the
main
group
should
also
see
as
well.
Okay.
E
E
Let
me
say
controversial
is
that,
for
instance,
what
what
are
the
boundaries
for
for
the
theater?
If
you
think
of
a
great
scenario,
what
what
are
the
boundary
is
that
installing
a
new
API,
a
new
could
have
been
there
shown
in
scope,
is
Tanja
new
kubilay,
Kubrick
version
in
scope,
so
yeah
for
you?
Yes,
well,.
E
Because
because
the
missile
pens
to,
let
me
say,
wider
range
of
romance
yep
like,
for
instance,
different
system,
operative
systems
and
and
properly,
it
was
so
large
the
scope
of
the
privilege
that
that
we
had
to
assigned
to
the
to
the
agenda
that
our
role
that
that
will
run
on
on
every
on
every
note.
So.
E
This
is,
for
instance,
happy
issue,
and
maybe
that
we
can
get
there
step
of
step
by
step,
for
instance,
defining
that
at
the
beginning
the
scope
will
be
only
what
is
managed
by
that
mean,
but
this
open
for
discussion,
so
my
management
focuses,
is
in
trying
to
define
a
good
semantics
for
the
for
the
Pareto.
So
a
semantics
that
is
a
really
simple
for
the
parrot
or
for
the
user,
Akua
cattle
imply
operator.
Does
this
action
and
I
all
the
complexity
that
they
act?
Shona
can
ever
so
multiple
step
on
different
nodes
and
so
on.
A
C
E
Let
me
stay
to
to
consideration
the
first
one
is
that
we
have.
We
want
at
least
at
the
beginning.
This
dissipater
will
be
optional,
so
not
is
valid
by
default.
Why
could
mean
so?
The
user
has
to
do
optional
of
team
and
the
icon
is
that
at
least
four
cubed
mean
we
are
giving
to
the
pods
this
the
same
grain
that
the
user
has
to
to
whoever
for
using
Cooper
meaning
itself.
E
A
A
A
C
Well,
not
necessarily
the
operator
itself,
but
the
pods
that
the
operator
will
execute
on
its
behalf.
We'll
need
to
have
privileged
access.
The
operator
itself
does
not.
The
operator
is
only
an
API
construct
that
would
communicate
via
secure
DS,
but
the
the
actions
that
the
operator
takes
by
executing
and
creating
pods
will
need
system
level
access,
which
should
be
the
same
as
what
4bc
was
mentioning
as
an
administrator
in
require
for
execution,
and
you
could
do
all
kinds
of
security
policy
around
that
too.
So
mm-hmm.
E
E
A
D
Basically,
to
move
medium
out
of
tree,
the
only
player
pending
items
are
the
cubelet
Roxy
config,
plus
the
API,
a
locator
for
the
API
P
allocator.
A
PR
is
already
open
to
move
the
code
out
and
they're
Cuban
Eddie's
utils,
but
for
the
cubelet
and
proxy
config
I
wanted
to
know.
If
we
had
any
like
long-term
solution
for
this,
because
I'm
not
sure
if,
if.
C
G
No,
we
there
is
PR
and
I
think
that
it
still
exists
and
it's
still
open,
but
it
hasn't
been
merged
and
PR
simply
drops
the
internal
types
and
uses
only
external
ones.
But
this
isn't
of
like
very
much
long-term
solution,
so
I'm,
currently
working
on
a
like
also
for
a
long-term
solution
which
is
going
to
include
narrowing
down
of
the
scope
of
our
component
of
X.
G
C
G
C
The
needless
to
say,
this
is
I
want
to
make
sure
we
do.
This
is
not
a
rushed
effort
yeah
and
we
do
the
due
diligence,
so
moving
stuff
on
a
tree
is
important
to
get
it
right
in
the
long
term.
So
if
we
need
to
take
a
little
bit
longer
and
dot
all
the
I's
and
cross
all
the
t's,
then
let's
do
that
I'd
rather
I'd,
rather
rather
do
it
right
and
into
judiciously
than
to
watch.
This
watch
weird
mekka
Nations.
C
A
C
I
would
say
this
is
best
effort
right
like
which
basically
means
like
we
get
to
it.
We
try
to
resource
it.
We
spend
time
on
it
and
we
do
it
when
it's
we're
ready
and
we
feel
like
it's
proper
time
to
do
it,
but
there's
no
guarantees
the
biggest
problem
that
they
had
was
that
we
were
ven
during
bad
libraries
right
so
long
as
we've
reduced
the
state
space
of
that
def
graph,
which
we
have
already
and
we
have
import
boss
in
place
which
we
have
and
we're
still
moving
to
go
forwards.
That's
what
matters.
G
A
D
A
D
A
That's
so
yeah
the
MP
rotator
and
the
remaining
Cobra
t
/
PG
stuff
is
approaching.
The
couplet
I
just
wanted
to
make
sure
to
Tim
that
we
found
some
bugs
in
the
import
boss,
import
boss,
apparently
completely
gloss,
underscore
test
dot.
Gough
I
also
you
can
import.
The
dog
here
is
the
key
in
at
this
file
and
nobody
is
going
to
notice,
but
Jordan
said
that
we
should
not
merge
the
peer
because
of
so
in
direct
dependencies
like
the
test
file.
A
We're
pretty
good
on
the
issues
I
just
wanted
to
mention
that
I
already
moved
a
lot
of
stuff
from
the
160
to
170
miles
on
the
mouse.
Roll
for
now
looks
mostly
cleanups
documentation,
more
features
books.
What
we
have
here
is
this
is
apartment.
It's
a
very
complicated
one.
I
guess
this
kind
design
not
bug,
so
we
only
have
something
like
20
tickets,
I
guess.
A
The
windows
I
saw
again
some
movement
in
the
windows
work.
The
install
scripts
are
already
in
place.
They
are
currently
working
on
documentation
on
how
to
setup
windows
worker
loads,
using
comedian
I'm,
going
to
test
this
myself
if
they
are
able
to
complete
the
work,
for
you
know,
claim
alpha
for
this
release.
I
want
to
test
it
to
make
sure
that
it's
working,
good,
I.