►
From YouTube: 20190327 kubeadm office hours
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
A
B
C
Pretty
sure
it
does
it
does
a
direct
reach
through
yeah
I
think
we
should
well,
it
depends
on.
It
depends
on
the
version
of
SCD
and
even
more
so
it
depends
on
the
version
of
the
sed
client.
Whether
or
not
you
have
it
available
directly
through
the
client,
where
you
have
to
go
through
the
rest
interface.
A
I'm
about
I,
don't
know
if
I'll
go
that
high,
maybe
like
60%
certain
ish,
that
the
status
reached
through
for
sed
directly
talks,
the
endpoint
for
the
API
server
directly
reaches
through
it
talks
to
the
sed
members
against
this
against
the
client.
The
correct
client
version,
so
I
will
I'll.
Take
a
look
at
this
one
today
and
I'll
verify
that
how's.
That
sound
sounds.
B
B
B
B
A
B
So
this
is
something
we
are
currently
discussing
with
Fabrizio
and
it's
a
length,
lengthy
topic.
The
document
is
still
private
because
we
don't
want
to
bother
everyone
with
our
ramblings.
There's
a
lot
of
noise
in
the
document
we
don't
have
a
solid
proposal,
but
so
we
like
peace
PSA
here,
is
that
we
started
already
planning
on
how
to
test
stuff
like
that,
and
hopefully
like
my
idea
of
a
finished
plan
by
the
end
of
the
first
month
of
the
cycle,
something
like
that
optimistically,
possibly
with
some
implementations
ready
to
use.
B
Okay
I
wanted
to.
We
also
have
a
separate
discussion
with
kind
versus
kinder
and
the
Monday
meeting
with
respects
to
kind.
We
basically
discussed
that
with
possibly
the
phases
are
not
sufficient
in
kind
for
us,
because
we
we
won't
be
able
to
test
things
like
that.
Upgrades
are
completely
out
of
scoping
kind
for
the
time
being.
If.
A
A
So
the
the
current
strategy,
or
the
current
thought
process
that
I
had
was
phases,
gives
you
not
all
the
way
there,
but
you
can
do
what
you
need
to
do
for
an
upgrade
by
stopping
before
you
actually
do
the
deployment
or
before
you
do
the
upgrade
process
and
do
it
manually
now.
If
we
want
to
wrap
that
behavior
in
in
kinder,
that
seems
totally
fine.
What?
What
more
are
you
thinking
there
because
I
know
that
phases
wouldn't
solve
the
problem
like
entirely
for
upgrades,
especially
so.
B
B
So
my
idea
was
to
pretty
much
execute
a
bar
script
from
brow
job,
which
is
going
to
act
like
an
entry
points
directly
into
kinda.
As
a
front-end
and
from
there
we
mean
Fabrizio
are
currently
working
on
a
solution
that
is
going
to
support
some
sort
of
a
config.
The
config
can
have
multiple
tests
in
Stata
that
you
want
to
run
like
possibly
upgrades,
then
tests,
then
the
cube
ATM
specific
intent
is
that
we
have
in
the
separate
package
fuel
that.
A
Seems
fine
to
me
I
would
definitely
not
recommend
getting
entangled
within
their
auto
builder
stuff
on
a
builder
test
stuff.
It
you
know,
if
you
even
need
to
you,
can
even
you
can
even
support
federated
testing
right
and
just
kick
it
be.
A
totally
separate
means
it's
up
to
you
or
whoever
wants
to
work
on
the
test
in
four
pieces,
but
that
that's
a
conflating
constraint
right.
That's
that's
like
they
they've.
They
buried
the
bodies
in
the
specific
area
and
you
need
to
like
need
to
get
around
it,
but
that's
a
tested
for
a
problem.
A
We
could
also
there's
another
strategy.
We
could
employ
there
too,
and
we
could
push
hard
and
getting
cluster
API
automation
is
blocking
release
signal.
I
think
we
might
conflate
a
bunch
of
problems
for
things
like
Kappa.
If
we
were
to,
you
know,
make
it
automatically
test
the
tip
of
everything
right.
A
A
A
There's
there's
no
reason,
especially
if
you
give
a
sháá
to
the
latest
versions
and
so
EE
with
114
requirements
on
there
for
not
to
it
can
just
give
you
signal
as
the
output.
Then
you
take
that
signal
and
you
use
federated
testing
to
just
put
the
results
in
the
GCS
bucket.
It's
up
to
us.
There's
many
ways
to
slice
this
one
but
I
would
I
would
not
kill
yourself
over
trying
to
make
I,
don't
know
being
something
so
really
try
to
bend
in
ways
that
was
never
designed.
Yeah.
B
Also,
possibly
the
the
discussion
we'd
been
with
a
burrito
from
Monday
like
gravitated
around
the
fact
that
we
share
the
same
back
end
kind
in
kinder,
and
if,
if
we
decide
to
push
on
a
new
feature,
we
can
always
get
it
upstream.
So
I
don't
see
the
fragmentation
problem
in
terms
of
back
end
in
terms
of
front
end
kind
is
very
limited
for
us
and
for
a
bit
Co
turrent
kinder
into
a
Swiss
Army
knife,
so
I
think
in
115
we
can
try
kinder
and
see
how
it
goes.
A
That
seems
fair
to
me
and
you
know
we
should
constantly
pushing
on
Ben
over
time.
The
problem
I
currently
see
with
upgrades
are
a
weird
thing
in
kind.
Not
gonna
lie
it's
super
useful
for
us,
but
from
an
end
user
perspective,
it's
a
little
weird
right
from
from
a
from
a
Kubb
Adm
developer's
perspective.
It's
like
oh
yeah.
We
want
this
yesterday
because
it
makes
our
test
automation
that
much
easier,
so
I
think
we're
gonna,
get
there
I
think
it's
just
gonna
be
social
crew
for
a
while.
A
If
we
get
phases,
that's
one
step
in
the
right
direction:
they're,
they're,
moving
in
that
direction
and
I
think
as
we
as
we
start
to
decrease
the
level
of
macro
commands
that
we
have
a
top
of
the
phases
that
we
need
to
do
to
execute
an
upgrade.
We
can
eventually
push
that
itself
into
maybe
an
experimental
things.
You
know,
there's
the
benefit
of
having
COO
medium
as
a
precedent,
and
this
was
my
thought
process
as
well.
The
benefit
of
having
Kumi
as
a
precedent
is
that
we
have
things
like
alpha
phases
right.
A
B
Yes,
it
makes
a
lot
of
sense
to
get
this
upstream
definitely
I
agree.
Ben
is
also
a
constraint
on
how
much
he
can
work
on
kind
and
basically
is
going
to
take
a
while.
But
given
the
the
process,
we
we
plan
for
115.
We
had
a
time
constraint
and
I
I
think
that
we
mere
four
Blitzer
should
just
push
to
get
something
out
and
think
later
of
cultural,
get
things
back
into
kind.
A
So
we
can
sync
on
that
one.
This
is
that
this
was
this
issue
rate
I
would
to
like
this
it
after
the
meeting
pretty
much
just
with
a
problem
is
if,
if
you
eventually
get
to
the
point,
where
you
say
forget
it
all
just
do
federated
testing,
that's
a
totally
legitimate
thing
to
so,
don't
feel
constrained
by
the
current
environment.
If
you
need
to
get
around
it,
do
whatever
you
need
to
do,
but
I
wouldn't
devote
time
and
energy
towards
making
fixing
tests
infra
itself
right.
A
D
B
B
The
next
topic
is
again
from
me:
is
we
so
long
story
short?
We
pushed
the
cni
upgrade
to
become
zero,
seven
five.
This
is
the
kubernetes
CNI
package
and
it's
broke
not
the
latest
tip
of
the
any
release
branch
packages,
but
the
other
bracket
just
like,
for
instance,
one
13.2
is
now
broken
because
of
this,
that
we
are
entering
the
land
of
packages
of
packaged
marriage
and
semantics
like
up
aptitude
candles.
This
is
one
way
a
young
Candace
is
in
a
different
way.
B
A
B
Broken
so
the
exact
output
is
that
it's
it
tries
to
install
zero
seven
five,
but
the
package
is
pinned
to
zero
six
zero
I
think
it's!
You
can
see
it
later
in
the
output.
B
But
we
have
to
lead
push
new
revisions
of
the
packages
in
that
way.
Yes,
and
basically
came
it
yesterday,
was
suggesting
that
we
probably
should
overwrite
the
old
packages,
because
we
we
are
essentially
pushing
a
security
fix
because
of
the
CVE
fix.
You
know
in
seven
five
in
pepper,
disagreed
with
that,
and
he
said
that
we
should
probably
go
in
the
revisions
instead,
but
then
you're
gonna
have
a
broken,
be
vision
forever.
Yes,
one
of
the
reason,
though,
the
primary
revision
is
going
to
be
broken.
A
This
is
part
of
the
reason
why
this
is
one
of
the
many
reasons
why
an
ass's
proposal
exists
and
why
both
hanok
and
Marik's
proposals
together
rp0
for
this
release.
It
should
be
super
clear
now
to
everybody
that
this
needs
to
be
fixed
for
obvious
reasons.
We
would
not
have
this
problem
had
those
two,
those
two
proposals
been
implemented.
There's.
B
Also
another
problem
like
if
we
again
my
discussion,
we
turned
it
from
yesterday
if
we
pinned
the
the
version
of
the
cni
in
the
couplet
spec,
for
instance,
to
more
than
equals
the
version.
This
means
that
we
are
going
to
always
pick
the
latest
CNI
package
and
we
might
face
a
scenario
in
the
future
where
a
new
CNI
package
is
not
compatible
with
the
old
version
of
coordinate,
ease.
A
That's
entirely
possible,
but
the
the
problem
is,
we
version
all
of
the
packages
together
as
a
thunk
and
if
a
person,
somehow
pre
installs
a
different
version
of
a
CNI
ahead
of
doing
all
the
other
updates,
that's
kind
of
a
pathological
condition
right,
it's
possible
that
they
could
do
this,
but
part
of
the
disabling
auto-updates,
which
is
part
of
our
installation.
Documentation,
is
to
prevent
people
from
blasting
their
own
foot
off.
A
So
they
have
to
turn
on
the
repo
explicitly
in
order
for
them
to
do
the
update,
and
we
tell
them
to
turn
on
an
update
all
we
don't
tell
them
to
turn
on
an
update
one.
So
I
hear
what
you're
saying,
but
the
instructions
would
I
don't
see
how
a
user
could
do
that
unless
they
went
off
the
beaten
path.
Yes,.
B
There's
another
problem,
which
is
a
pretty
big
one
related
to
our
dependency
chain.
We
currently
the
cupid
en
package,
depends
on
the
couplet
and
many
people
yesterday
expressed
the
concern
that
this
is
a
problem.
We
should
stop
depending
like
that,
because
this,
by
the
way
breaks
up
grades
because
you
have
to
hold
you
have
to
hold
the
corporate
version
when
you
upgrade
the
Canadian
package,
because.
A
Of
that's
bad,
that's
been
along
stats.
A
separate
issue.
I
do
know
that,
but
that's
a
different
thing.
That's
been
a
long-standing
issue
for
a
long
time.
That
was
the
one
that
Craig
Tracy
filed
a
long
time
ago
and
there's
another
one
around
that
too,
as
well
as
should
we
be
fixing
this
as
well
and
their
efforts
in
this
cycle.
Yes,
we
should
I.
B
B
B
This
was
one
of
a
we
shipped
multiple
images
for
fedora
for
Ubuntu.
This
was
more
of
a
generic
idea,
like
your
proposal
is
close
to
the
one
that
ended
it,
and
he
said
that
we
should
have
this
metal
container
that
installs
the
binaries
locally.
But
then
we
faced
the
question
like
how
about
all
the
the
couplet
dependencies
that
depend
on
jealousy.
How
are
we
going
to
install
them
on
the
host?
We
have
to
execute
the
local
package
manager
and
yeah.
That
was
the
other
proposal.
B
A
Too
high
of
a
bar
I
mean
I
I,
hear
what
you're
saying,
but
the
bar
is
too
high
for
average
people.
The
whole
reason
why
cube
ATM
got
traction
is
because
we
lowered
the
bar
and
made
it
easy
for
everyone,
and
that's
like
the
fundamental
premise
behind
a
lot
of
the
things
into
video.
Is
that
make
it
easy
for
everyone
democratized
and
can
monetize
the
installation
process.
I
think
we
need
to
find
a
balance
in
an
off
road
here.
A
I
do
think
that
long
term
maintenance
of
these
packages
it's
become
a
problem
because
we
did
everything
wrong
and
we
never
actually
paid
down
the
technical
debt
if
it
was
not
a
problem,
you
know
if
it
just
worked,
seamlessly
was
turnkey
and
was
relatively
simple
and
we
actually
built
the
same
artifacts
as
part
of
our
build
process.
As
we
do
this
release
I
doubt
we'd
be
having
this
conversation.
A
E
We
can
combine.
We
continue,
rised
approach,
who
is
like
new,
can
move
all
the
packages
so
like
I
might
have
script,
which
will
do
exactly
what
chorus
guys
are
doing
scripted
and
if
we
need
to
execute
like
package
manager
to
install
an
additional
dependencies.
It's
not
what
hard
to
detect
it's
like,
like
MDNA,
a
fat
zipper
Parkman.
What's
it
like?
What's
the
most
important
ones,
yeah
I.
A
Don't
think
we
can't
do
both
we
should
do
both
for
the
time
being,
I
I
loathe
the
maintenance
of
packaging
for
a
long
time.
I
have
since
the
beginning
of
this
particular
project,
because
it's
one
of
those
things
were
like
who
wants
to
volunteer
for
packaging,
and
everyone
looks
around
no
one
stands
up,
but
the
the
consumption
bottle,
though,
is,
is
key
to
our
success,
as
always
has
been
because
people
have
pre-existing
ways
of
managing
these
things.
A
A
F
B
Yeah,
basically,
the
problem
is
also
that
seek
release
are
not
really
owning
this
problem,
we
as
a
sequencer
lifecycle.
We
have
some
sort
of
a
responsibility
to
distribute
to
medium,
but
the
packaging
stuff.
It's
really
forced
on
the
sick
release
and
they
don't
have
engineers.
That's
a
very
dense
organizational
problem.
A
Yes,
I've
been
saying
this
for
a
long
time.
I
think
you're
you're
coming
full
circle
to
my
realization
of
about
like
two
years
ago,
so
they
you
come
to
the
understanding
of
why
I
broke
out
things
that
testify,
and
you
can
also
come
to
the
full
understanding
of
why
we
kind
of
do
things
of
packaging
implement
every
second
of
it.
So
they
welcome
to
my
world.
B
A
G
B
B
So
the
the
signal
is
not
super
complete
yeah,
we
cover,
we
have
a
couple
of
jobs.
One
of
them
is
running
that
Lucas
created
its
checks
for
the
existence
of
the
packages
and
the
second
one
that
I
created
is
to
install
all
the
packages
and
then
uninstall
them.
We
don't
have
a
job
that
tests
upgrades
with
holes
and
stuff,
like
that.
We
don't
have
a
such
a
job
and
also
everything
like
I
said
earlier
is
Debian.
We
don't
have
rpm
testing.
B
B
D
B
B
A
B
B
B
E
B
B
A
H
B
G
Think
that
there
is
an
interesting
part
topic
to
consider
that
are
popping
out
requests
about
having
in
the
conflict
the
possibility
to
configure
components
on
secondary
contraband
node,
so
configurate
etcd
on
the
secondary
control,
plane,
configure
controller
manager,
API
server
and
the
secondary
control
plane.
I.
Think
that
as
I
see,
we
have
to
define
the
bar
for
this
flexibility.
G
G
B
H
B
A
H
A
A
B
I
wanted
to
see
if
we
have
some
like
something
a
critical
event
or
important,
so
perhaps
to
escalates
stuff
too
much
in
master.
We
also
have
some
a
bunch
of
PRS
that
are
like
staying
there
for
a
long
time.
We
can
look
at
a
couple
of
peers
for
met
butters.
He
has
some
extensions
to
the
qadian
pocket
come
on.
Basically,
we
have
stuff
to
discuss
if
you
want
to
during
the
meeting.
A
B
A
A
A
Everything
or
nothing
so,
let's
just
walk
backwards
for
the
bottom,
because
a
lot
of
these
are
mill
tools
full
sakes.
We
could
even
do
it,
so
anything
wrong
signal
usually
has
been
like
encompassing
so
no
no,
no,
no
definitely.
No.
This
is
an
API
changing
itself.
No,
no,
not
really!
No!
No!
No!
No!
This
one's
been
sitting
around
looks
like
it
needs
a
rebase.
B
A
A
You
can
just
let,
if
they're,
if
they're,
already
marked
as
life
cycles
run,
you
can
let
it
auto
close
it
that
way,
you
won't
get
any
sort
of
flack
from
the
community
I.
Probably
let
that
happen
to
be
honest,
because
sometimes
you
don't
want
to
have
the
you,
don't
want
to
be
the
closer
of
issues
and
PRS
yeah
sure.
A
Iii
base
means
rebase
has
got
LG,
TM
use
your
base,
multiple
cigs,
multiple
cigs,
multiple
cigs,
multiple
saves
needs
rebase.
A
lot
of
these
are
just
basically
stale
and
old.
Because
of
the
common
problem
is
you
need
it?
It
was
probably
an
API
change
of
some
kind
or
something
that
affects
multiple
perverse
and
it
was
never
prioritized.
D
B
A
E
F
B
Honestly,
with
the
group
here,
I
wanna,
be
honest:
I,
don't
even
know
why
we
have
the
fetching
of
a
version
as
the
first
option.
Ii-I've
never
seen
a
software
that
do
this
does.
Does
this
like
that,
we
should
be
for
to
the
whatever
the
client
version
is
walkin,
II
and
optionally.
We
should
have
the
user
power
from
the
internet
if
they
want
I.
E
B
Yeah,
it's
okay
to
know
what
the
latest
parsed
version,
because
we
guarantee
that
we
are
not
going
to
break
this
particular
minor,
but
at
the
same
time
it's
it's
I,
don't
even
know
why
why
this
is
done
like
that?
We,
so
is
it
not
be
a
factor
in
the
future,
because
these,
like
all
the
logic
that
ESP
are
surrounding
that
fight
with
rice
and
stuff,
we
can
completely
avoid
this
I
think
if
we
get
a
factory
in
a
better
way,
I.
B
B
A
A
B
H
B
H
B
B
A
I
guess
I
don't
see
a
problem
with
it.
Do
you
want
to
finish
to
review
this
Ross
I.
B
H
A
I
D
A
B
A
B
So
this
this
PR
at
Bardo
she's,
trying
to
his
on
the
call
he's
trying
to
add
an
option
to
list
more
than
one
token
I
mean
not
so
this
one,
that
more
than
one
talking
about
at
the
possibility
to
list
the
tokens
in
machine,
readable
format
and
I.
Think
that
there
agrees
that
this
is
a
good
option,
except
that
we
have
figuring
out
how
to
properly
process
the
outputs.
And
he.
I
I'm,
just
thinking
guys
that
maybe
it's
not
the
best
place
this
list
tokens,
maybe
it
should
be
moved
somewhere
else
like
as
a
part
of
bigger
thing.
I
mean
like
to
output
as
much
information
about
the
queue
bottom
and
the
cluster
is
possible
or
something
like
that.
But
I
don't
know,
I
didn't
find
it.
B
So
my
idea
about
the
machine,
readable
output,
support
in
cube
ATM,
is
that
we,
we
might
end
up
with
some
sort
of
common
mechanism.
Yes
implement
this,
but
at
the
same
time
I
think
we
should
extend
the
existing
commands
with
like
the
flag.
Outputs
in
Yama.
A
supplier
creates
new
commands
for
the
sake
of
machine,
readable,
I.
G
Am
a
Peter
of
two
problems.
One
is
how
we
define
the
output
if
we
have
to
define
a
new
API
group
in
order
to
get
the
output
version
as
and
so
on,
and
second
is
how
we
apply.
The
machine.
Readable
are
good
because
for
common
delight
talking,
it
is
easy,
but
for
common
alike
in
it,
where
we
have
all
the
printf
function
spread
across
the
code
is
much
more
difficult.
So.
A
But
let's
take
this
to
our
issue
because
this
is
I,
don't
know
if
I
actually
closed
the
root
issue
that
this
was
originally
stemmed
from
a
long
time
ago,
which
I
probably
did
is
that
we
wanted
this
behavior
for
a
very,
very
long
time.
I
don't
see
an
issue
referenced
here,
but
why
don't
we
take
this
one
to
an
issue
because
we're
actually
out
of
time
and
then
we
can
discuss
that
for
next
time.