►
From YouTube: kubeadm office hours 2020-01-08
A
A
A
A
All
right,
I
have
some
pieces
myself
for
the
chair.
Pic
deadline
for
one
17.1
is
this
Friday
and
next
week
Tuesday
is
going
to
be
the
actual
release
of
this
batch
version.
I
saw
some
notes
about
the
release
schedule
for
1:18,
but
it's
the
work
in
progress
and
the
release
date
is
planned
for
the
end
of
March
I.
A
So
this
is
a
real
PC.
We
found
a
couple
of
bugs
during
the
holidays.
I
was
searching,
I
was
watching
the
issue
tracker
from
time
to
time,
and
one
of
them
was
a
public
checkpoint,
panicking
the
CPU
manager-
and
it
actually
was
fixed
today,
the
pr
merged
for
this,
and
it
was
blocking
couplet
upgrades
from
117
to
118.
So
if
you
have
a
couplet
woke
early,
that
is
117
and
you
change
the
binary
to
be
version
118,
the
118
binary
is
going
to
panic
because
it
cannot
recognize
the
old
checkpoint
file
properly.
A
So
one
17.0
is
broken.
If
you
see
this
in
the
wild
just
so
that
you
know,
and
the
fix
is
going
to
be
117
actually
so
I'm
sorry
I
make
a
mistake
here.
This
is
only
master,
so
it's
already
fixed
yeah.
The
other
bug
is
yeah
I
mixed
it
to
develop.
The
other
bug
is
slightly
before
Siders
are
broken
one
17.0.
A
We
we
saw
some
error
messages,
basically
that'd
be
a
locator,
cannot
allocate
the
proper
right
piece
for
you,
and
we
said
the
fix
for
this
Jordan
sent
a
fix
for
this
in
for
117
dots,
1
and
also
the
fix
is
already
master.
We
have
to
do
some
digging
where
the
problem
is,
but
we
found
this
like
very
much
on
Christmas
Day
and
we
fixed
it.
C
Yeah
I
worked
with
this
on
this
with
Jordan
I.
Think
I
have
an
idea
where,
where
the
issue
is
I'm,
not
sure,
though,
if
it's
totally
fix
it
everywhere,
so
I
need
to
take
a
look
again.
A
So
Jordan
didn't
backbones,
you're
your
PR.
Basically,
we
didn't
backported
completely.
He
took
portions
of
it
through
fixed
core.
The
rest
we
fixed
in
a
separate
cube,
a
DMP
are
because
ok,
you
also
touched
the
validation
cube
a
DM
to
check,
but
we
had
a
separate
peer
because
it's
not
critical.
We
check
very
small
ranges
in
there.
Ok.
A
A
A
Much
should
share
more
details
here,
but
yeah.
Basically
it
writes
a
file
to
disk.
This
is
the
the
state
file
of
CPU
manager,
so
on
disk
117
had
one
format
and
118
has
a
different
format.
118
cannot
convert
properly
from
the
old
format,
so
we
had
to
apply
a
fix
to
handle
this
manual
combustion,
because
this
is
not
using
API
machinery.
This
is
just
no
manual
parsing
of
three
fields
in
a
JSON
structure:
okay,.
A
All
right
so
I
translated
the
my
document
for
cube
Adi
moving
out
of
kaykai
into
a
camp.
The
KB's
here
so
I
extracted
some
of
the
decisions
that
we
made
in
the
dock
as
well,
but
I
still
pretty
much
copy/paste
big
portions
of
it.
So
this
cap
I'm
not
exactly
happy
how
big
here
ended
up
being
because
it's
a
lot
of
information.
A
So
this
might,
you
know,
pull
people
away
from
it
because
it's
just
too
large
but
in
any
case
I'm
hoping
that
Morais
can
see.
This
I
see
that
so
I
already
spoke
with
sick
dogs.
They
are
fine
with
the
changes
it's
pretty
simple,
but
the
sig
release
stuff
is
really
complicated
and
I
said.
If
I
already
reviewed
some
requests
here
from
from
Stephen.
A
A
This
is
just
mentioning
that
I
did
some
work
on
Windows
worker
notes
during
the
holidays
as
well.
This
pretty
much
support
us
adding
support
for
upgrade
of
workers,
because
Michael
from
cig
Windows
was
asking
me
multiple
times
already.
So
I
did
some
investigation
if
it
even
works,
and
so
it
works
fine.
You
can
upgrade
windows
workers
using
this
very
simple,
powershell
script.
You
know
as
much
as
I
understand
PowerShell.
A
A
We
also
have
some
security
problems
in
the
way,
so
they
wanted
to
stop
running
proxy
and
the
CNAs
windows
services
and
instead
use
daemon
sets,
but
there
is
a
very
nasty
hack
that
they
use
with
a
certain
tool
that
gives
high
privilege
to
you
know
a
binary
that
is
from
the
Windows
host
to
tree
some,
basically,
some
containers
that
are
not
privileged
as
privileged,
so
it's
but
it
it
exposes
a
security
hope
and
they
have
no
solutions
for
that.
So
I
don't
think
this
should
graduate
to
beat
in
that
set
and
is.
A
It's
handled
by
Ben,
moss,
Ben
moss
is
from
pivot.
Oh,
he
people
don't
just
join
VMware
like
on
New
Year's
Eve
spend
pretty
much
but
I.
Don't
even
have
all
the
details
right.
So
the
person
to
ask
is
Ben
and
I
think
he's
tracked
somewhere,
but
I
don't
even
know
I,
don't
think
it's
even
tracked
as
an
issue.
Yet
it's
pretty
much
slack
discussions
to
this
point,
but
yeah
Ben
is
the
contact
for
that.
A
So
yeah
also
the
developer.
This
end-to-end
tests,
I'm,
not
sure
who
see
windows,
are
going
to
a
scientist.
I
guess:
Ben
moss
can
also
help
with
that
I
think
Adelina
I,
don't
know
what
what
company
she
works
at,
but
we
have
some
people
that
can
help
and
I
can
also
help
with
coordination
there,
because
I
totally
know
stuff
like
Asia,
I
guess.
Ben
Ben
also
mentioned
that
we
can
somehow.
A
A
Which
is
a
really
nice
escape
route
for
this
instead
of
you
know
creating
a
new
deployer
and
using
Asia,
but
this
is
only
slag
discussion
at
this
point.
I
don't
even
know
so
we
have
to
pull
Ben
potentially,
if
you,
if
this
group
wants
to
know
more
about
Windows,
we
have
to
pull
him
into
this
meeting,
so
he
can
explain
because
everything
is
slack
at
this
point.
B
A
E
A
B
A
Yes,
so
this
is
interesting
because
so
I
have
criticism
for
the
work
that
was
done
here.
I,
don't
think
these
scripts
are
very
well
written
and
I.
Don't
think
that
costal
API
should
use
them.
These
are
PowerShell
scripts
I
mean
maybe
cost
repair
windows
nodes
can
execute
these
PowerShell
scripts,
but
this
this
work
needs
people
who
understand
power
show
because
it's
not
it's
not
like
five
lines
of
code.
This
is
kind
of
complicated
and.
A
C
B
B
Multi
master
was
a
part
of
whatever
we
support
upgrade
and
as
soon
as
they,
it
is
clear
what
supported
schools
could
be
meaning
it
is
up
to
the
provider
to
choose
if,
if
to
to
go
through
this
path
or
not
I
open
today,
today
for
the
provider,
this
will
be
implemented
by
the
cluster
BIA
WS
provider.
But
but
let
sorry
that
bide
by
the
azure
provider.
B
A
B
A
B
B
F
B
B
B
A
A
A
You
can
do
that,
but
also
I
believe
what
Fabrice
was
trying
to
say.
That
is
that
you
can
no
longer
use
a
game
in
it
as
an
action
you
have
to
and
also
comedian
join.
Potentially,
you
have
to
remove
these
commands
and
add
your
own
commands
to
go
on
all
the
containers
and
call
cube
ad
a
minute
or
cube
a
DM
join
with
your
custom.
Configure
is
already
in
the
containers.
A
A
B
It
is
still
there
because
we
don't
have
a
good
alternative
for
the
user
yet,
but
what
we
are
saying
that
is
that
when
you
change
the
conflict
before
a
trade,
basically
there
are
no
guarantee
of
the
result,
because
user
who
want
to
do
crazy,
stuff
or
by
changing
the
config
and
coopered
mean
upgrade
the
currently
support.
Only
two
or
three
type
of
changes
so.
B
But-But-But
I
really
can
sync
offline.
Basically,
the
point
that
that
I
was
trying
to
make
is
that,
if
you
use
Cooper
the
mean
do
is
like
a
macro,
and
this
macro
has
some
assumption
behind
I,
don't
know
exactly
what
kind
of
change
are
you
trying
to
test
and
depending
on
of
the
the
kind
of
change
you're
trying
to
test?
Maybe
the
macro
works
for
you
or
you
have
to
do
something
different
yeah.
G
F
G
F
B
F
Okay,
yeah,
but
that
would
be
perfect,
but
yeah
I
didn't
even
know
about
the
get
artifacts
command.
So
that's
that's.
Useful
I
can
use
that
already.
Just
fetch
my
build
from
CI
so
and
then
does
the
the
tasks
in
the
workflow.
Does
it
actually
upgrade
the
couplet
as
well.
B
B
F
F
A
B
Okay:
let's
try
to
maybe
open
an
issue
and
try
to
count
the
issue.
Yeah.
F
F
B
F
F
Thank
you
for
doing
that.
That's
that's
very
useful,
so
I
guess
basically,
I
can
use
the
workflow
to
pass
a
config
and
then
I
want
to
make
a
note
variant
image
with
the
additional
artifacts
directory
fetch
from
CI
and
then
I
boot
up
a
cluster
with
that
and
run
the
workflow
and
then
that
that
should
do
the
couplet
upgrade
and
the
coop
idiom
upgraded
with
the
proper
config
and
I
can
use
a
volume
mount
to
use
the
build.
F
F
B
B
A
A
Okay
thanks
the
next
one
is
such
the
job
is:
I
walked
in
2060.
This
is
the
this
basically
proposal
to
integrate
the
coaster.
Basarab,
slash,
TOS,
bas,
trap,
sludge
discovery
into
cork
or
eighties,
and
I
think,
like
parts
of
it
are
already
there
I'm,
not
sure
what
is
like
what
is
spending
here.
I,
don't
know
the
context
in
any
case,
if
somebody
wants
to
take
a
look
at
this,
please
do.
D
I
think
it's
so
old,
and
it's
so
early
days
from
where
it
was
today.
I
would
recommend
just
closing
it
and
say
I'm
currently
deprecated,
because
it's
not
some
of
these
issues
that
are
so
old
that
we
can't
can't
make
sense
out
of
it
anymore.
I
think
we
should
just
close
because
it
just
alleviates
the
craft.
A
A
Ability
to
create
dynamic,
HP
questions
for
this
will
have
to
decide
like
what
what
else
do
we
have
to
do
to
graduate
to
stable?
It
has
been
two
releases
since
we
went
beta
for
aging,
so
basically,
we
have
to
create
a
ticket
in
Cuba
in
the
cube,
ATM,
repo
and
few
a
list
of
tasks
that
remain
for.
If
we
need
to
to
do
something
else
for
I
guess,
the
coastal
status
is
something
related
did.
B
Okay,
so
I
I'm
told
to
update
the
ticker
and
telling
that
we
need
two
things
to
make
it
GA,
which
are
the
learner
and
and
the
status
and
then
fix
the
problem
on
the
management
of
the
status,
and
there
is
a
caper
for
this
one.
So
we
are
not.
We
are
not
going
to
ad
cycle,
but
we
are
fixing
one
of
the
two
problems.
A
A
Companies
are
sorry.
This
is
not.
This
is
component
status,
so
I'm
in
favor
of
deprecating
component
status,
because,
for
instance,
if
how
I
HEA
into
the
question
is
still
going
to
print
only
the
first
member
of
the
question
and
I,
occasionally
we
get
a
ticket
in
the
qualities,
qualities
repository
people
asking
a
hey
why's.
Why
are
my
entity,
members
missing
so
to
clarify
corporate
status
is
basically
exposed
it
first
of
all,
it's
a
structure,
but
it's
exposed
with
this
command
cube,
corrugate
CS
and
it's
a
API
that
I
think
it's
reasonable
to
deprecated.
F
A
F
H
A
F
I
mean
the
repo:
is
there
I,
wait,
let's
setup
and
staging
it,
there's
there's
a
lot
of
active
work
that
needs
to
occur
to
pull
stuff
into
this
repo.
It's
really
very
it's
a
long-term
continual
effort,
so
I
don't
know
really
like
what
the
like,
how
to
track
the
enhancement
like
there's,
multiple
things
happening
in
this
repository
that
will
be
with
varying
levels
of
quality.
A
Think
the
purpose
of
this
issue
is
because
it
had
a
kept,
and
we
also
have
a
cape
for
kept
in
an
issue
for
the
cube
ADM
moving
out
but
I'm,
hoping
that
we
are
going
to
the
cause.
The
cube
a
day,
moving
out
tickets
fairly.
Soon,
though,
the
purpose
of
these
issues
against
too
much
the
kept,
because
it's
a
requirement
in
the
Cape
to
have
an
issue
in
here.
Yeah.
F
Well,
the
issue
lets
you
track.
Multiple
changes
to
the
cap
as
it
graduates
your
faces
and
things
like
that,
so
yeah,
they're
they're,
just
maybe
bullet
points.
You
know
that
should
be
irrelevant
here,
but
yeah
I
mean
yeah.
It
is
the
only
confusing
thing
here,
but
this
this
issue
has
probably
got
useful
information
in
it.
I
would
imagine
yeah.
A
F
To
exist,
I.
Imagine
it's
important
for
their
tooling
for
determining
what
happened
in
a
release
or,
if
not
to
Lane,
just
workflow,
because
I
do
get
comments
on
kept
issues
where
people
paying
like
hey
like
what's
happening.
Did
we
make
this?
You
know?
Were
there
meaningful
changes
that
we
need
to
document
yeah.
A
F
F
A
A
A
A
A
A
F
A
A
A
He
changed
our
internal
config
in
P,
1,
beta
2,
and
also
they
won't
be
the
one
to
have.
This
I
said
that
these
are
locked.
We
shouldn't
change
the
old
config,
but
I
think
4
V
1
beta
3.
We
can
have
this.
We
can
provide
an
optional
field
to
specify
the
the
type
of
encryption
the
user
wants
to
use
any
say
or
HCD
si.
A
The
alternative
is
a
feature
gate
which
I'm
not
in
favor
of
because
future
gate
I
think
our
or
a
complication
for
such
things,
and
also
the
other
components
in
communities
are
already
support.
Elliptic
curves,
so
we're
just
enabling
something
that
is
already
supported,
but
I
think
if
we
enable
this
in
our
v1
v3,
we
have
to
have
end-to-end
tests
for
sure.
A
A
It's
just
a
question:
if
this
is
enough,
wait
for
the
next
person
and
I
think
we
have
to
pretty
much
decide
in
a
separate
meeting.
Yeah.
A
Yeah
I'm
also
considering
the
fact
that
the
person
who
wanted
to
contribute
this
code
came
out
of
his
other
work
to
help
us
here
and
they
have
also
you
know
they
have
a
schedule,
so
they
can
contribute
this
now
and
if
we
say
okay,
we
are
not
going
to
add
this
in
the
next
couple
cycles.
They
might,
they
might
bail
out,
so
we
have
to
take
over
their
work.
So
this
source
of
fact
on
here.
F
B
I'm
I
am
splitting
the
two
problems.
If,
if
they
want
to
contribute
to
the
feature,
they
can't
contribute
the
feature
and
put
behind
a
feature
gate.
If
we
want
to
expose
the
feature
in
in
the
main
API,
this
required
an
iteration
of
the
War
API
and
to
do
an
iteration
we
are,
then
then
we
have
to
make
several
other
changes.
B
A
With
supporting
this
separate
algorithm,
because
it's
not
clear
to
me,
how
are
we
going
to
support
user
transitioning
away
from
NSA
to
ecliptic?
How
do
we
support
this
transition?
Because
when
people
see
this
new
support,
they
are
going
to
want
to
migrate
their
existing
clusters
to
the
new
functionality
and
the
PR
does
not
answer
this
question.
How
are
we
going
to
do
that.
A
D
D
No,
no,
like
you
know
it's
harder
to
decode,
even
with
a
quantum
computer
elliptic
curve
cryptography,
so
that
people
will
have
restrictions
about.
You
know
this
capabilities
that
especially
given
the
current
climate
of
things,
so
I,
don't
I,
don't
know
how
you
deal
with
this
enough.
The
source
project
just
being
honest,
we've.
Never
it's
only
concern
I,
think
kind
of
lately,
given
the
current
political
climate.
A
A
A
A
D
This
is
just
open.
We
need
to
just
remove
the
code
from
it
touches
cluster
lifecycle,
but
we
I
think
we
just
have
to
figure
out
all
the
touch
points
you
originally
had
in
this
and
just
remove
it
because
there's
the
original
code
for
this
is,
it
was
always
feature
gated
yeah.
We
decided
not
to
go
down
this
road
anyway,
so.
D
So
kind
of
yes
and
no
so
the
boot
Kubb
is
a
very
highly
opinionated
version
of
the
world
like
and
there's
a
bunch
of
caveats,
the
more
you
dig,
especially
with
regards
to
security
implications,
so
the
as
we
dug
into
it
and
started
to
sort
of
migrate
features
from
boot,
cube
into
proper
into
cake,
a
prompt
are
especially
into
the
people
it
like.
We,
we
just
came
into
a
hornet's
nest
of
problems
that
we
said.
D
I,
don't
necessarily
know
if
this
is
worth
the
effort
not
to
mention
that
the
the
implications
for
doing
the
encoding
on
a
local
machine,
or
just
it
was
just
onerous.
So
we
just
punted
and
there's
a
limited
set
of
people
who
really
want
to
do
self
hosting
in
the
non
aj
fashion,
because,
if
you're
in
a
chasin
Aereo
across
spanning
Easy's,
this
is
a
not
an
issue.
D
This
is
basically
if
I
have
an
H
a
cluster
with
its
given
DC
there's
a
whole
DC
outage
like
how
do
I
recover,
and
for
that
scenario
we
just
recommend
static
configurations
and
because
we
are
working
on
things
like
the
comedian
operator
and
cluster
API.
Those
are
the
sort
of
the
plan
plans
for
managing
the
lifecycle
of
the
control
plane
and
not
having
to
deal
with
the
the
finicky
details
of
security
implications
are
to
each
employee.