►
From YouTube: kubeadm office hours 2019-10-30
A
A
C
A
Thank
you
for
yourself
happy
to
have.
You
here
feel
free
to
ask
question
at
your
point
to
the
agenda
and
we
are
looking
forward
to
your
contributions,
all
your
issues.
So,
let's
do
on
travel,
I
have
to
point
in
agenda
the
first
one.
Is
that
I
think
that
I
want
to
discuss
with
you,
so
the
it
is
the
result
of
some
tests
in
cluster
API,
where
basically,
we
discovered
that
could
mean
join
Muse
means
control.
Plane
is
fearing
due
to
the
due
to
a
timeout
that
happens
when
why,
basically,
it
is
a
mist.
A
Kwame
is
waiting
for
the
UTC
remember
to
be
ready
investigated
the
problem.
What
I
discovered
that
cupid
mean
is
waiting
for
cold
ET,
s'en
member
to
report
a
status.
So
if
we
assume
we
have
a
cluster
with
two
nodes,
we
join
a
new
one
and
cupid
mean
before
telling
that
the
tornado
is
ready.
Just
wait
for
the
status
of
all
the
three
nodes
and.
A
A
No
problem
so
when
you
join
a
new,
a
new
member,
a
new
control
plane
node
at
the
end,
we
have
a
weighted
await
loop
that
wait
for
all
the
etcd
members
to
report
a
status
and
we
find
us
on
some
properties
whenever
we
are
doing
joining
parallel
or
or
putting
pressure
on
to
ethnicity.
So
my
idea
is
that
why
we
have
to
wait
for
all
the
distal
member
and
instead
we
wait
only
for
the
new
member
to
the
born
status.
D
Well,
a
part
of
this
could
be
solved
by
treating
it
like
a
stateful
set,
which
I,
don't
necessarily
think,
is
a
bad
idea
like
a
sed
deployment
or
the
API
server.
The
control
plane
itself.
If
you
treat
it
or
think
of
it
like
a
staple
set,
it
has
to
go
0
1
2.
It
cannot
go
highly
in
parallel,
because
I,
don't
even
think
sed
can
really
even
support.
It.
I
think
there's
a
bunch
of
race
conditions
that
live
Amir
found
some
of
them
and
we
found
some
other
ones.
A
A
E
E
A
Our
tester
are
getting
better,
so
we
are
tornado
green.
Also,
the
tests
that
are
marketers
with
flaky
is
just
a
delay
on
the
on
the
refresher
on
the
logic
behind
this
code
is
bottom,
so
our
test
grade
is
is
turning
in
in
good
shape,
but
what
I
would
like
to
discuss-
and
we
suffer
at
some
problem
recently
due
to
external
reasons,
so
one
reason
was
accordion
page
for
core
DNS
and
including
some
problems.
Another
reason
that
was
a
Adelina
variability
on
the
artifact
seen
in
DCI
CD
buckets
for
both
of
them.
A
We,
we
have
implemented
some
workaround
in
order
to
get
to
make
out
test
more
these
resilience
such
problems
for
cárdenas.
We,
we
have
not
put
keeping
anymore
the
the
manifest
from
from
Internet
map,
but
we
have
a
core
a
copy
of
the
manifest
that
we
know
it
works
and
for
the
the
artifacts
we
have
some
summer.
Try
waiting
that
the
machinery
published
the
they
the
binary
we
need
for
for
our
test.
The
problem
was
that,
while
stabilizing
for.
A
D
So
we've
went
through
this
a
couple
times
and
I
believe
Chris
Rousey
even
submitted
a
PR
to
make
it
easier
to
filter
issues
when
they
come
through.
As
you
know,
this
is
from
test
grid.
So
the
answer
to
this
is
like,
if
you're
on
the
mailing
list,
that
means
that
you
care
about
these
things
and
that
you
can
easily
filter
them.
So
maybe
an
answer
to
this
would
be
here
as
a
filter.
You
can
copy
that
you
can
apply
if
you
don't
care.
D
B
A
A
Okay,
thank
you,
Tim,
okay.
That
last
point
with
regard
to
the
to
the
test,
and
is
that
the
kind
of
need
which
is
our
test
tools
eat
some
attention.
First
of
all,
we
have
to
upgrade
to
the
latest
to
the
latest
kinda
release
and
wreak
the
center
a
PR
some
time
ago.
It
is
like
being
the
biscuit
painting
if
you
can
keep
up
and
finish
this
work,
it
will
be
appreciated.
Otherwise,
I
will
step
in
and.
E
Yeah
I
got
this
to
a
point
where
it
was
working
and
I
think
we
were
just
struggling
with
node
conformance
if
I'm
not
mistaken.
I
just
saw
you
posted
I
just
checked
this
because
of
the
agenda
item.
So
you'd
left
the
comment
six
hours
ago.
Thank
you,
but
I
can
I
can
come
back
to
this
and
get
it
working.
E
A
A
E
B
E
A
E
E
E
A
D
E
D
A
A
A
Okay,
this
is
a
cleanup
that
I
remember
now.
It
is
a
cleanup
I
discusses
with
at
the
bar
and
basically,
as
part
of
the
work
is
doing
the
for,
for
the
machine.
Readable
out
of
I
asked
him
to
try
to
remove
all
the
the
print
statement
that
we
have
deep
in
our
country
and
have
and
at
all
the
print
statements
only
in
the
first
layer
of
our
product,
so
the
depakene
will
be
clean
and
he
created
an
issue,
and
I
see
that
a
new
contributor.
A
A
Okay,
just
only
one
commenting
in
from
me
about
this,
and
one
interesting
things
about
the
proposed
now-
is
that
it
it
works
in
the
initial
stage
of
of
the
upgrades.
So
if,
if
it
fails,
it
fails
fast
without
screwing
up
your
cluster
and
the
second
is
that
it
works
on
all
the
nodes
and
then
this
it
is
really
important
when
you
are
standing
in
a
locally
built
images,
so
those
two
behavior
should
should
be
preserved.
A
F
A
A
A
A
A
B
A
A
A
So
we
also
agreed
to
to
to
adopt
a
new
convention
for
the
package
from
the
package
name
and
let's
see
it
was
already
done.
Okay,
we
are
switching.
We
are
removing
the
beta
bear
shown
from
the
package
name.
So
from
now
on
changing
the
whenever
we
will
switch
our
release.
This
will
be
easier
and
the
number
of
change
would
be
much
lower
than
this
PR.
So
this
one
was
easy.