►
From YouTube: kubeadm office hours 2020-01-29
A
A
A
Thanks
all
right,
I
don't
see
anybody
else.
I
knew
I
wanted
to
go
through
some
PSAs
quickly,
so
KK
is
quite
flaky.
So
if
have
peers
just
about
that,
merging
them
currently
is
almost
impossible,
because
we
have
very
consistent
flakes
in
some
of
our
six
storage
and
API
machinery
jobs,
and
this
is
because
we,
or
rather
someone
disabled,
a
certain
logic
to
candle
flake
attempt.
So
currently
we
actually
catch
real
problems,
and
previously
we
have
this
logic
to
retry
certain
things.
A
So
this
is
a
PSA.
People
are
working
on
that,
but
it
might
take
a
lot
of
time
to
fix
the
Flex.
Another
PSA
is
that
client
goal
I
announced
this
yesterday
during
the
sequester
lifecycle.
Meeting
that
client
go
is
adding
a
new
API
interface.
That's
going
to
support
goal
contexts.
So
currently,
if
you
do
an
API
call
such
as
you
know,
config
Maps
get
you
don't
have
a
way
to
timeout
this,
but
with
the
context,
support
you're
going
to
have
primary
arguments.
That
is
a
goal
context.
A
You
can
supply
timeout
to
that
and
you
can
timeout
the
call,
but
this
is
a
breaking
change.
So
what
they
decided
to
do
is
to
introduce
a
new
interface,
basically
a
new
client,
pretty
much
and
they're
going
to
have
the
two
clients
in
parallel
for
a
while,
but
I,
don't
know
how
much
time
cubed
in
being
inside
the
cake
a
is
going
to
benefit
from
that
because
they
have
automation
to
already
convert
to
the
new
client.
There's
a
big
PR
where
I
linked
here
you
can
see.
A
A
C
C
A
A
I
see
I
see
that
people
already
approved
it.
So
maybe
tomato
said
it
because
I
was
a
dpi
machine
recall
when
they
discuss
this,
so
I
think
they
are
going
for
the
way
that
they
already
explained.
But
it's
not
clear
to
me
if
they're
actually
going
to
execute
the
way
they
discussed
it
and
the
way
it's
described
in
the
cap.
C
E
If
this
fits
here
on
the
PSAs,
but
just
for
letting
you
know,
I
I
opened
this
like
minutes
ago,
so
we
have
this
early
on
the
cycle
as
soon
as
possible.
Actually
so,
if
you
have
some
time,
please
have
a
look
at
that.
It's
implementing
the
custody
status
dependency
removal,
but
we
still
are
maintaining
that
from
qadian,
just
deprecating
that
so
yeah.
Please
have
a
look.
If
you
have
time.
A
A
Just
a
second,
so
rushkoff
updates
the
PR
for
CDSA
support,
and
tomorrow
saying
he
added
this
as
a
feature
gate
now,
and
we
have
a
very
clear
notice
that
this
is
supposed
to
be
used
experimental
as
a
future
gate
from
Cuba
da
minute
upgrades
are
not
supported
or
switching
between
different
algorithms
as
well.
If
anybody
has
time
this
is
the
the
PR,
it's
eighty,
six,
nine
five
three.
C
A
Part,
switching
between
the
organisms-
she's,
not
supportive,
but
anyone
who
has
a
time.
Please
take
a
look
at
this.
This
is
not
critical
for
us
too
much
in
118,
because
we
haven't
seen
that
much
demand
for
it.
That's
the
main
reason
I
guess,
but
if
you
have
time,
if
we
have
reviewer
bandwidth,
we
can
proceed
with
margin.
It
I.
A
G
G
Just
I'm
imagining
if
the
API
server
is
not
configured
to
use
that
crypto,
then
maybe
that's
what
prevents
the
algorithm
switch.
Otherwise,
I
don't
really
see
why
we
couldn't
do
an
upgrade
with
downtime
and
switch
the
algorithm,
or
is
it
that
couplets
will
lose
communication
and
that
they
have
to
all
be
upgraded
to.
A
That's
part
of
the
problem
is
you
have
to
generate
all
certificates
in
keys,
even
if
your
certificate
authority
is
LSA,
you
have
to
generate
everything
to
be
now
at
CDSA.
But,
yes,
you
have
to
manually,
stop
the
couplets
and
yeah.
You
have
to
pass
some
flag,
I,
think
somewhere,
but
I
I'm.
Sorry
I
forgot
where
the
API
server
itself
is
I.
Think
it's
sufficient
to
pass
the
clip
to
switch
streets
having
HCDA
say
in
the
list.
Yeah.
G
A
So
a
Roscoe
fees
are
in
this,
but
I,
don't
think
he's
thinking
about
what
you
are
discovering
here
and
I
told
you
I
would
have
the
bandwidth
to
start
investigating
what
is
supported
here.
So
if
you
want
to
comment
on
the
PR
itself
to
Roscoe
I
think
this
is
going
to
be
appreciated
and
and
then
we
can
start
the
discussion
of
the
PR
look
or
you
also
have
a
ticket
for
it
by
the
way
yeah
yeah.
C
I
am
fine
too
ever
something
which
is
feature
gated
that
is
incomplete
as
if
it
is
documented
and
then,
of
course,
if
part
of
is
part
of
being
alpha
and
feature
gate,
also
not
being
completed.
So
for
me,
for
me,
it
is
acceptable
even
because
I
guess
that
the
problem
for
upgrade
is
to
regenerate
cells
to
ensure
a
proper
certificate
rotation
across
the
cluster,
which
is
something
that
no
one
has
documented
properly.
Yet
so
I
don't
want
to
block
the
contributor
on
this
task,
which
is
huge.
A
C
A
A
All
right,
I
want
to
give
a
quick
update
on
the
qadi
moving
out.
Keep
this
kept
sort
of
discussion.
It
hasn't
like
200
comments
already
and
the
final
comments
came
from
Ben
the
other
and
suggesting
they
have.
Basically,
he
proposed
that
we
meant
certain
sections
to
clarify
the
testing
scenarios.
I
agree
with
that
I'm
going
to
update
some
of
these
parts.
Some
of
them
are
not
clear
to
me.
Some
of
these
questions
are
not
clear
to
me.
He
has
to
respond
for
Britt
syrup.
A
F
A
Right,
let's
I
wanted
to
mention
something
else
about
the
Cape,
so
I
went
to
sick
tests
in
yesterday
and
we
discussed
like
spent
20
minutes,
discussing
the
Cape
parts
of
the
proposal
that
are
related
to
them.
So
the
cap,
if
you
remember,
contains
a
certain
section
explaining
that
we're
going
to
use
a
post
submit
job
for
automating.
A
They
need
to
have
high
privilege,
for
instance,
we
have
something
that
synchronizes
llaman
files
in
cradle
cap
teams,
but
to
do
this
they
have
to
give
this
application
a
high
privilege
token
for
the
kubernetes
organization
and
to
be
able
to
modify
tags
and
branches.
We
also
have
to
give
this
pasta
meat
to
job
the
same
access
and
it
has
to
reside
in
something
that
is
called
the
secure
cluster
of
prowl.
A
So
progress
has
multiple
clusters
and
the
secure
cursor
is
maintained
only
by
testing
for
folks,
so
immediately
we're
going
to
use
us
lose
our
access
to
this.
We
are
not
going
to
be
able
to
make
any
changes
without
them,
approving
it.
It's
a
very
tight
process
and
I
said
okay,
but
this
is
the
security
aspect,
the
blocker
here,
and
they
said
no,
we
think
instead,
the
targeting
branches
should
be
done
in
the
signal
is
tooling,
so
I
also
went
to
seek
release
some
time
ago.
A
I'm
in
this
bureaucracy,
administration,
rabbit
hole,
trying
to
figure
out
who,
once
this,
my
conclusion
is
that
nobody
wants
this
or
what
we're
going
to
do
is
we
are
going
to
have
to
manually
tag
in
branch,
the
upholstery
with
a
tool,
the
tool
you
are
executed
to,
and
it's
going
to
do
it
for
you
is
going
to
result
in
the
comedian
repo
and
this
I'm
going
to
change
the
cable.
To
basically
do
this
to
explain
this
any
comments.
H
J
H
A
A
H
A
A
A
So
when
you
apply
the
cube,
dns
adult
for
some
reason,
instead
of
Cardenas
you're
going
to
see
a
warming
saying
that
Cuba
DNS
deprecated
is
going
to
be
removed
in
a
future
release,
I
said
a
message
to
the
mailing
list:
KDF
and
I
didn't
see
any
replies
from
people
potential
consumers
of
Cuba
DNS
and
the
coordinates
folks
have
expressed
that
this
is
a
maintenance
problem
for
us.
Therefore,
after
a
graceful
period
of
time,
I
decided
to
do
this.
I
didn't
see
any
objections,
but
if
anybody
has
comments
now,
please
say
we
can
see
report
PRS.
A
A
With
the
coop
EDM
upgrade
apply
common
in
the
future.
If
we
are
able
to
skip
the
DNS
phase
completely,
the
users
will
be
able
to
apply
the
custom,
DNS
solution
and
ultimately
others
such
as
cube
DNS,
just
the
Yama
file.
So
they
can
maintain
this
separate
process
if
they
want
and
I
don't
think
you
medium
should
handle
multiple
options
for
this
but
roasty.
This
means
that
we
have
to
change
the
API
to
maybe
stop
supporting
the
DNS
option.
A
The
next
item,
I
edit,
is
for
that
we
in
kata,
if
we
started
discussing
the
potential
for
removing
CID
epsilon
P
ins
for
KK.
This
is
something
the
signal
is
want.
Another
team
made
a
proposal
some
time
ago
that
everything
should
be
in
KK,
although
signal
is
have
something
else
in
mind
and
they
want
all
the
specs
all
the
tools
to
reside
in
KK
release
and
they
want
to
branch
intact.
The
repository
there's,
let
me
fetch
the
peers
in
discussions
for
this
actually,
so
this
is
the
name
by
Stephen
augustus.
A
A
We
built
depths
in
our
ps4,
the
API
server
a
control
plane
and
we
then
inject
these
depths
inside
docker
images,
and
this
is
done
as
a
workaround,
because
buzzer
does
not
support
setting
permissions
for
binaries
and
I
saw
some
very
peculiar
stuff
going
on
in
the
puzzle
code
as
much
as
I
can
read
buzzer,
but
we
got
banned.
The
other
eye.
Xd
I
commented
on
this
and
this
whole
code
mystery
factor,
but
it
seems
that
these
folks
are
walking.
They
don't
want
us
to
remove
the
the
generation
of
depths
and
rpms
from
there.
A
J
I,
don't
care
as
long
as
somebody
owns
them.
The
problem
that
we've
had
is
that
we
have
two
locations
and
no
one
really
owns
them,
either
right
and
so
long
as
we
have
a
single
owner
or
we
I
don't
want
to
sing
over
I
want
to
set
it
winners.
I
want
a
group
that
is
responsible
for
the
ownership
of
these
things
and
they
should
take
ownership
rights.
But
technically
that's
a
release.
Artifact
thing,
so
I
would
be
fine
if
serially.
If
cigarette
this
has
been
owned,
this
they
own
it
own
it.
A
Completely
agree:
in
the
past
few
cycles,
people
like
Stephen
has
stepped
up
to
maintain
CNI
updates
and
also
also
updates
in
the
specs
for
the
couplet
for
for
the
other
packages.
So
it's
already
things
are
already
leaning
towards
seek
release
owning
them.
We
can.
We
can,
of
course,
advise
if
something
is
not
appropriate
in
a
spec
file.
We
can
always
do
that
or
if
we
demand
changes,
but
they
are
the
signal
ISA
of
the
right
people
for
us.
J
Well
then,
we
should
make
it
explicit
like
they
should.
There
should
be
an
owner's
file
inside
of
their
thing.
That
says
this
is
the
canonical
source
of
truth
and
that
you
know
all
other
things.
It
sounds
weird.
It's
almost
like
religion.
This
is
the
canonical
source
of
truth
and
everything
else
is
a
false
TD.
J
A
Yeah,
that's
absolutely
true
yeah,
they
started
the
refactoring
I,
don't
even
know
what's
going
on
here
anymore,
but
there's
now
a
pic
eg
father.
We
have
cake
eg,
which
is
I,
think
a
new
tool
and
this
tool
in
the
future
is
supposed
to
be
building
the
artifacts
from
specs
I
would
be
happy
if
we
stopped
doing
this
in
KK.
But
the
conversation
is
a
bit
bored
here
and
what
to
monitor
this
issue
where
it
goes
in
the
future,
but
yeah
the
bus'll
specs
are
very
complicated,
very
entangled.
A
A
Yeah,
so
Mike
too
often
is
a
parent.
Sorry
look
Mike,
toughen
Mike
Denise
is
one
of
the
original
authors
of
this
code
next
to
Jeff,
but
he
made
a
comment
that
it
was
very
peculiar
to
me
and
ie
I
responded
that
my
major
concern
about
Basel
is
that
we
don't
have
enough
people
that
even
understand
this.
You
know
mutation
of
Python
that
it's
using.
A
So
if
you
don't
have
enough
people
that
understand
a
language,
why
do
we
even
have
this
language
in
the
source
code
and
the
people
that
understand
the
language
are
very
busy?
Which
means
that
if
you
send
a
PR,
you
have
a
problem
with
your
PR.
We
don't
understand
how
to
fix
it
now
all
nurse
if
they
are
busy.
They
are
not
going
to
respond
to
you.
J
I
think
I
have
strong
opinions
in
that
we
kept
on
saying
that
basil
basil
is
going
to
be
like
this
magical
Wonderland
of
greatness
and
I
have
yet
to
be
convinced
that
it
is
anything
but
a
yet
another
build
tool
which
is
cause
which
has
given
me
false
promises.
You
know
I'm
old
enough
to
remember
every
build
tool
that
has
ever
existed
for
a
C++
and
Java
to
know
that
they
keep
on
saying
all
these
things
are
gonna,
be
magical
and
great
only
to
realize
that
it
was
a
false
set
of
promises
that
you
know.
A
A
A
H
H
A
E
H
Yeah
this
is
trying
to
set
us
up
so
that
we
can
make
the
faces.
There
were
a
couple
functions
and
what
this
one
is
doing
is
there
was
a
function
called
enforced
requirements
that
basically
did
everything.
It
got
the
objects
for
upgrading
and
returned
those
as
well
as
ran
some
of
not
all
of
the
pre-flight
checks
and
other
things,
and
so
I
broke
that
apart,
so
that
you're,
basically
this
one
just
does
two
steps.
H
H
H
Right
now,
it's
all
very
integrated
in
yeah
intertwined,
so
anyways
I'm
very
open.
If
you
guys
want
to
have
a
zoom
meeting
or
whatnot.
If
you're
looking
over
this
and
our
little
confused
as
to
why
I
did
some
things,
I
might
have
done
something
wrong:
I,
might've
added
a
reason
to
do
it
so
anyways
yeah.
Just
let
me
know
what
your
thoughts
are
I'm
trying
to
get
this
out
as
fast
as
I.
Can
thanks.
A
H
G
I
found
it
difficult
to
plumb
dependencies
through
this
code
as
well.
I'm
gonna
need
a
rebase
if
we
merge
this.
The
cluster
ions
work,
because
the
upgrade
flow
was
very
they're,
very
different
from
the
init
phases
in
terms
of
how
it
organizes
its
data
dependencies.
How
it
calculates
paths
really
fantastically
different,
so
I.
H
A
I
H
A
G
A
H
So
what
what
I
did
was,
in
theory,
when
we
merged
this
one,
the
other
one,
can
just
be
rebased
based
off
of
what
got
changed
whenever
you
know
per
your
guys's
comments,
and
then
we
can
move
to
the
next
one
or
we
could
literally
just
work
through
each
one
and
at
the
very
end,
just
merge
the
last
one
and
it
would
merge
all
of
the
stuff
I.
Don't
know
how
you
guys
do
that
I
was
just
trying
to
make
it
into
smaller
chunks,
because
it's
a
lot
of
updates.
A
I
would
prefer
if
we
have
disappear.
We
clear
the
whole
picture,
make
it
ready
for
the
phases
and
then
proceed
with
the
others.
To
not
have
the
other
PR
scene
does
the
cube,
but
whoa
knie.
You
can
make
your
changes,
but
I'm
not
gonna,
say
like
here.
Are
the
other
PS
also
clean
ups,
or
are
they
phases
already?
No.
G
H
G
And
the
question
is
just
whether
or
not
we
want
to
merge
the
last
one
or
merge
the
three
in
succession
and
to
lube
lemierre's
comment:
if
we
merge
the
three
in
succession,
you'll
have
to
wait
for
the
merge
queue
in
order
to
do
that.
So
it's
a
question
you're
asking:
how
do
we
want
to
review
and
merge
the
work?
I
mean.
H
If
you
want
me
to
do
it
differently,
I
can
do
it
differently.
I
can
close
these
these
and
do
it
differently.
I'm
up
for
anything.
This
is
just
how
I
tried
to
do
it,
because
it's
a
lot
of
code
change
and
a
lot
of
it
is
rearranging
the
code.
So
it's
not
actually
as
much
code
changes.
It
is
just
reorganizing,
but
it's
still
large.
D
G
I
can
do
that.
You
basically
could
take
a
while
okay
I
think
I
I
agree
with
lumière
suggestion.
Just
it's
going
to
be
easier
on
the
automation.
It'll
be
hard
to
synchronize.
Merging
these
three
part,
three
PRS,
so
just
in
one
as
units
would
be
excellent
and
I.
Think
dear
last
ones,
probably
structured
like
that
right.
G
H
H
Correct
and
then
I
tonight
and
tomorrow,
I
was
gonna
work
on
the
the
last
PR,
so
they'll
be
four
in
total,
so
before
commits
and
I
will
I
will
do
that.
I
will
close
these.
These
are
you
can
if
you
want
to
get
a
look
over
them.
This
is
the
start
of
this
stuff,
though
this
will
be
the
first
commit.
So
if
you
want
to
get
a
look
over
it,
that
would
be
great,
but
I
will
be
making
a
new
PR.
Then.
A
H
G
A
G
I
just
wanted
to
make
sure
not
to
go
too
long
without
informing
everyone
on
what
the
status
of
cluster
add-ons
is
and
especially,
hopefully,
rusty
is
here
yeah
thanks
for
joining,
so
this
component,
config
refactor,
that
is
in
kibale
iam,
now,
has
been
pretty
difficult
to
work
with.
Thank
you
for
doing
it,
but
yeah
just
the
regression
of
like
removing
the
API
machinery.
G
That's
the
status
of
where
the
rebase
is
at
and
that
demo
of
the
implementation
with
the
init
and
upgrade
with
this
work
coming
in
as
well
for
refactoring
upgrade
I'll
need
to
rebase.
On
top
of
that,
which
may
take
a
moment,
we
haven't
updated
the
cap
and
address
the
comments.
I
will
do
that
this
week.
I
am
interested
as
well
to
talk
about.
Oh
there's,
no
progress
on
the
coup
proxy
or
core
DNS
operator.
My
thought
here
is
once
the
rebase
is
done.
A
F
G
If
you
were
expecting
conversions
to
her,
because
basically
we
would
need
to
do
two
conversions
in
the
client
instead
right.
So
if
I
accept
an
internal
type
in
the
library
as
an
input,
then
that
is
the
in-memory
agreed
upon.
Like
version
of
the
API,
that's
how
you
do
stuff
in
memory
with
a
struct
and
go
so
you
would
expect
that
kubja
diem
would
take
in
an
arbitrary
version.
G
C
K
C
G
A
We
should
go
to
the
core
of
the
problem.
The
core
of
the
problem
is
that
the
only
components
that
is
doing
conversion
is
the
component
itself,
not
the
add-on
installer.
Nothing
else
has
access
or
should
have
access
to
the
internal
tags.
Only
Q
proxy
should
be
able
to
convert
between
its
different
types
right.
A
If
Q
proxy,
if
API
machinery
are
telling
us
that
a
pair
machinery
is
not
supposed
to
convert
between
public
types,
then
the
corporate
itself,
in
this
case
Q
proxy
as
an
example.
Aha,
should
be
the
only
thing
that
performs
conversion.
Therefore,
if
the
if
the
user
is
passing
manifests
with
other
is
star,
we
don't
know
if
it's
going
to
work.
If
the
user
passes
some
version
of
the
to
proxy
config,
we
have
an
image
of
the
Q
proxy.
G
J
A
Ultimately,
for
so
these
corporates
that
we
cannot
expose
commercial
I
think
we
should
expose
the
conversion
as
a
service.
So
if
you
have
the
key
proxy
running
as
a
daemon
set,
if
you
have
an
endpoint
that
is
slash,
convert
whatever
a
pop
of
data,
you
pass
tweets,
you
should
be
able
to
return
to
you
some
sort
of
conversion,
like
maybe
another
API
type,
and
this
is
something
that
they
buy
Michelle.
You
don't
want
yeah
I
see.
This
is
the
way
to
go,
and
this
should
live
in
corporate
base
at
some
point.
But
you
see.
G
I
envisioned
a
binary
command-line
interface
for
the
conversions,
but
we
need
to
have
a
design
meeting
about
this.
That
I've
been
I've
been
looking
at
organizing
because
there
are
several
areas
of
component
config
now
that
meet
design
work
in
order
to
get
to
the
next
level
of
usability,
including
Docs,
and
that
kind
of
thing.
But.
G
G
K
H
G
You
don't
want
people
to
actually
use
the
struct.
You
just
want
it
importable,
so
that
the
code
can
perform
the
conversion
right.
So
it's
not.
It
considered
an
external
interface
it
it's
the
internal
face,
interface
of
the
code,
the
actual
component
uses
or
the
library
or
whatever
you
don't
want
people
actually
passing
around
and
creating.
You
know
constructing
internal
objects,
but
it
it
exists
as
a
necessity
for
the
code
to
have
one
version
of
the
API,
and
then
it
is
the
conversion
hub
as
well
for
conversions
and.
G
Also,
this
is
very
helpful
to
add
to
the
agenda
for
component
standard
discussions
on
Tuesday
but
yeah
as
far
as
it
goes
with
the
cluster.
Add
ons
thing
thinks
I
will.
It
sounds
like
I
will
just
update
the
library
to
accept
a
single
version
and
then
I
will
redress
these
PR
rusty.
Can
you
DM
me
your
the
PR
that
you're
working
on
yeah
sure?
Thank
you
very
much.
G
A
G
C
C
Then
point
my
point
is
the
following:
is
that
now
there
is
only
config
and
all
the
upgrade
and
when
you
do
a
great
using
this
config?
Ok,
if
you
are
going
to
support
different
type
of
corporate
config
when
doing
join,
join
us
to
take
action
in
order
that
the
next
upgrade
of
the
same
node
will
reapply
the
right
coffee,
I.