►
From YouTube: 20181205 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
Hello
today
is
December
5th
2018.
This
is
the
standard
cluster,
a
life
cycle,
meaning
we
don't
actually
have
an
agenda
for
today.
Are
there
topics
that
folks
want
to
discuss.
I
know
that
we're
gonna
do
a
lot
of
work
out
at
cube.
Con
I'm
spending
a
lot
my
time
right
now,
kind
of
prepping
for
that
I
can
walk
through
some
broader
sig
details
as
well
as
some
other
stuff
and
folks,
if
you
don't
have
any
other
topics.
A
B
A
C
A
A
C
A
C
A
You
can
actually
still
get
the
gold
docs
for
an
older
version.
How
let
me
think
about
this?
First
again,
I
have
to
take
a
look
to
see
if
you
can
actually
link
to
older
versions
of
go
ducks.
This
won't
be
a
problem
going
forwards.
This
is
a
point-in-time
problem
for
sure
for,
like
anything,
that's
V,
111,
so
I
we
have
a
temporary
problem
with
dock
linkage.
For
now,.
A
C
C
B
A
A
Well,
like
because
of
the
version
windowing
of
how
how
the
support
lifecycle
works
for
beta
plus
versions,
we
actually
don't
have
this
problem,
because
we
will
always
have
a
v1
beta
one
for
the
entire
lifecycle
support
that
exists
for
the
support
cycle
for
kubernetes.
So
it's
just
an
overlapping
problem
so
because
we're
a
V,
1
beta
1,
now
that
one
will
not
be
a
problem
in
the
future
right,
because
we
have
to
maintain
support
review
one
beta
one
for
at
least
two
plus
cycles
after
the
initial
release.
A
That's
three
windows,
so
that
means
that
you
have
support
now
for
the
entire
lifecycle
for
supported
versions
of
kubernetes.
So
as
soon
as
we
drop
the
next
version
of
kubernetes,
we
drop
view
1,
beta
1
and
now
we're
on
V
1,
beta
2
right,
so
so
for
beta.
Once
we
eventually
get
the
beta
checkpoint,
it's
no
longer
a
problem.
It
is
a
problem
currently
for
the
next
six
months
for
be
111
support
and
B
112
support.
B
D
B
Cube
a
gaming
it
we
have
some
outer
content
in
here
and
we
have
a
section
about
the
config,
and
here
this
is
currently
what
is
it
master
of
the
dogs
willing
to
this
page,
and
this
this
basically
exists
in
the
old,
the
old
websites,
like
the
old
version,
the
112?
Oh,
no
one
want
to
open
113
in
the
113
Docs.
This
is
going
to
be
hard-coded.
Okay,
and
this
is
a
hard
color
link.
If
I
open
this,
it
opens
go
dogs,
it's
not.
B
B
A
Technically,
here's
the
reason
why
I
don't
care
once
we
get
to
beta
one
is
because
kubernetes
only
supports
three
versions
right
so
like
as
as
a
as
a
district
as
upstream
community
communities
will
only
support
111,
112
113
for
any
window
time,
there's
only
three
versions
that
are
ever
supported
and
that's
that's
been
policies
for
four
years
now.
So
if
if
a
person
is
stuck
on
an
older
version,
they're
literally
out
of
the
support
window
right,
so
you
know
they
should
have
updated
when
they
were
at
the
edge
of
support.
A
And
if
they
didn't
update
at
the
edge
of
support,
then
that's
a
provider
problem,
not
necessarily
a
community
problem
right.
So
if
your
provider
is
somehow
sticking
you
if
you're
still
stuck
on
like
111
in
a
year
from
now,
that's
a
problem
with
the
provider
and
not
necessarily
the
community,
is
that
our
problem
to
solve
right,
but.
C
C
A
Also,
you
know,
like
it's
a
point
like
I
mentioned
before
it's
a
point
in
time.
Problem
I
also,
don't
think
we
need
to
deprecated
or
remove
v1
alpha
3
like
it's,
not
a
requirement
for
us
to
remove
it
at
this
point.
In
fact,
I
kind
of
wants
to
have
a
quiescent
period.
I
do
not
want
to
do
a
V
1
beta
2,
unless
it
is
absolutely
positively
necessary
to
try
and
match
up
as
much
configuration
Delta
as
faithful
as
possible.
C
A
C
A
Iii,
don't
think
I
don't
see
a
need
to
remove
so
there's
a
couple
things
we
can
do
to
mitigate
the
problem
in
the
future.
So
we
have
a
three-month
window
of
v1
alone,
Docs
being
potentially
slightly
off
I
think
we
can
leave
v1
alpha
3
and
not
there's
no
there's
no
forcing
function
that
forces
us
to
remove
it.
Usually
you
always
have
to
straddling
versions
in
the
code
right
and
yes,
ideally,
we
would
clean
it
up,
but
because
of
this
issue
I,
don't
there's
no
real
need
for
us
to
remove
it.
A
E
B
A
Well,
yeah
yeah,
we
were,
we
were
doing
a
lot
of
API
turn
to
try
and
get
us
into
this.
This
steady
state-
and
we
didn't
want
to
maintain
the
potential
version
apparatus.
No,
the
goo
I
call
it
goo
it's
a
lot
of
it's
a
lot
of
as
you
call
it.
Boilerplate
I
call
it
goo,
because
it's
more
accurate
description,
it's
basically
a
bunch
of
auto-generated
gobbledygook
that
that
does
the
conversion
for
us
and
we
fill
in
a
couple
of
templated
functions.
A
A
It's
all
documented
in
in
the
link
it's
just
not
as
pretty
as
go
doc
right.
It
just
doesn't
have
the
formalized
HTML
version
of
it
and
we
might
find
a
way
for
us
to
clean
this
up.
I
think
it's
worthwhile
to
log
an
issue.
If
someone
wants
to
address
it
in
the
Covidien
repo
but
I,
don't
it's
not
I,
don't
see
this
as
like.
There's
a
wood,
there's
workarounds
that
are
fine.
A
A
In
the
meantime,
I
know
that,
like
there's
a
there
is
a
fairly
reasonable
sized
backlog
and
I
don't
want
folks
to
feel
like
they're
blocked
on
execution
feel
free
to
do
PRS
against
the
backlog.
It's
still
relatively
sorted.
We
haven't
gone
through
the
full
prioritization
list
for
this
cycle,
but
I
think
it's
in
good
shape
from
a
sink
perspective
like
not
just
qadian
perspective.
A
One
of
the
things
I
wanted
to
make
sure
folks
are
aware
of-
and
we
have
a
conversation
later
on
today-
is
that
we
really
really
really
want
to
beef
up
kind
to
be
the
tool.
That's
the
default
PR
blocker,
as
we
saw
in
last
cycle,
we
got
broken
once
we
hit.
What
was
it
alpha?
Beta
clayton
did
a
PR
push
and
there
was
no
PR
blocker
for
goobidy
m
style
installations
and
it
broke
us
immediately.
A
It
was
a
Saturday
morning
and
you
know
making
breakfast
for
my
kids
and
I'm
trying
to
fix
this
so
stuff,
sorry
be
recording
fix
this
stuff,
so
I
think
getting
us
on
a
fast
PR
blocking
job
that
uses
kind
and
and
remove
some
of
the
other
ones.
We
talked
about
this
with
this
league
level
and
remove
the
other
things
like
cops
and
some
of
the
other
GCP
jobs
to
be
more
periodic.
It
makes
the
PR
blocking
jobs
much
more
consistent
and
less
flaky,
because
it's
localized
it's
a
lot
faster
to
run.
C
Team
I
want
one
question
about
this.
This
part,
because
I
was
I,
already
started,
working
with
been
unkind,
initially
aiming
at
replacing
the
cupid
mini
playground,
so
aiming
at
using
kind
as
a
way
to
quickly
test
new
cupboard,
min
deployment
and-
and
so
I
started
with
this
in
mind.
But
I
got
your
point
that
we
can
use
kind
as
a
CI
tool.
C
What
is
is
not
really
clear
to
me
is
which
kind
of
use
cases
we
are
going?
We
want
to
cover
using
kind
I'm,
making
an
example.
We,
when
I,
do
testing
for
the
for
the
new
release,
I
have
a
more
or
less
a
matrix
with
some
variation
of
the
cluster
that
I
can
create
so
I
test
external
tcd
I
test
as
external
CA
and
forest
an
LCA.
Now
we
have
at
least
three
variation,
astonishia
eastern,
a
front
port
from
proxy
Esther.
C
Not
so
the
text
matrix
is
is
complex
not
only
in
term,
of
course,
a
variation,
but
also
in
term
of
sequence.
You
can
test
so
in
it
join
another
master,
then
upgrade
or
in
it
upgrade
and
enjoy
master.
So
what?
What
is
that?
What
would
be
really
useful
for
me
is
to
try
to
fix
what
kind
of
test
we
want
to
execute
automatically,
because
one
of
the
pain
point
of
kubernetes
anywhere
anywhere
was
that
it
it
was.
It
is
capable
to
spin
up
a
cluster,
but
all
the
cluster
is
up.
Basically,
you
are
stuck.
A
A
It
falls
over
underneath
certain
use
case
scenarios
and
if
we,
if
we
can
actually
set
up
and
provision
a
set
of
test
scenarios
that
covers
a
good
swath
of
our
configuration
that
we
want
to
do
I
one
of
the
part
one
of
the
pieces
we
absolutely
need
for
CI
for
for
PR
blocking
is
just
basic
basic
kuben
men,
cluster
single
node
in
it
multi
node,
join
up
to
three
by
default,
typically
and
just
verifies
end-to-end
test
behavior.
That
alone
is
it
is
a
eighty
percent
use
case
for
what
we
need
it
for
for
development.
A
Then
there's
the
twenty
percent,
which
is
like
I,
need
an
H,
a
deploy
rate
of
some
some
incantation
and
I.
Think
that's
gonna
be
a
little
more
complicated.
We
could
get
a
base
h
a
deploy
using
sort
of
localized
load
balance.
Maybe
at
middle
l,
be
in
this
scenario
or
h,
a
proxy
or
some
other
tool
to
give
that
full
h,
a
deployment
style
from
a
kind
environment
right,
oh
I,.
C
A
I
think
that
that
user
story
is
super
useful
for
all
of
us,
and
that
is
almost
a
requirement
for
us
to
get
a
che
into
beta
right,
because
we
need
to
have
the
auto.
We
need
have
a
general
auto
deploys,
but
also
does
like
that
uses
Liz's
tests,
the
test
apparatus
for
cube
ATM
to
actually
verify
and
do
H
a
specific
configuration
verification
that
we
don't
like
implode
things
right,
because
the
intent
tests
currently
today
don't
actually
verify
some
deployment
configuration
stuff.
So
we
need
localized
specific
tests
for
cube
ATM
that
that
does
intend
verification.
C
A
B
Just
wanted
to
say
that
BIM
wants
to
create
cubed
ezdeploy
for
q4
kind.
We
can
technically
avoid
that
if
we,
if
we
basically
feed
specific
scripts,
I
mean
this
probably
discussion
for
later,
we
can
potentially
feed
specific
scripts
to
jobs.
Instead
of
having
to
maintain
this
extra
deploy,
but
I'm
sure
that
Ben
is
going
to
be
very,
like
he's
gonna
push
for
this.
B
A
A
This
will
be
highly
leveraged
by
multiple
people,
not
just
for
us,
the
sig
as
part
of
CI,
but
as
potentially
a
developer
desktop
solution
that
is
much
more
powerful
than
a
lot
of
the
other
developer,
desktop
solutions
that
exists
today,
right
the
kubernetes
or
the
docker
desktop
a
single
node
and
doesn't
support
system.
D
capabilities
can't
do
a
bunch
of
stuff.
The
the
mini
cube
is
a
weird
traduction
that
actually
has
limitations
on
multi
node
dined
is
really
close,
but
it's
it's
a
pile
of
bash.
A
So
if
you
want
to
integrate
it
with
automation
and
your
own
code,
it
actually
becomes
a
little
bit
funky
and
weird
I.
Think
the
kind
is
the
distillation
of
all
of
these
ideas
together
the
best
of
my
ideas,
together
to
the
point
where
you
know
we
can
actually
start
to
create
programmatic
style
interfaces,
which
she
already
has
done.
He's
got
he's
got
well
versioned
and
well.
Typed
configuration
files
for
his
thing,
so
he's
got
to
have
automatic
version
upgrades
when
he
does
configuration
file,
changes.
B
D
B
A
A
A
A
A
For
folks
that
want
to
join,
but
I
would
I
would
love
to
get
Cabernets
anywhere
out
of
out
of
the
picture
entirely
for
114
and
I.
Think
you
know,
because
we're
not
going
to
be
doing
a
lot
of
shuffling
and
in
this
cycle,
I
think
fixing
some
core
tech
debt
issues
that
we've
had
for
a
while.
Both
helps
us
move
faster
in
the
future
too.
A
B
This
a
couple
of
days
ago,
Jeff
broke
the
CI
CI
a--
defects
for
you
medium
because
of
basel
versioning,
and
currently
the
only
the
only
end-to-end
test
that
we
have
that
catches.
This
is
Cabrini's
anywhere,
it's
pretty
good
at
that
it
catches
all
the
CI
breakages
yeah.
So
if,
if
we
start
using
kind,
we
have
to
also
have
the
end-to-end
tests
that
verify
back
is
yes,
I
do
yeah.
A
A
Or
a
second
now
paste
it
inside
of
the
slide
chat
or
instead
of
that
yeah
there.
It
is
so
folks
want
to
take
a
look
at
it.
This
is
this
is
multi-part.
There
is
a
lot
of
moving
parts
to
this,
but
I
think
we
I
think
it's
totally
possible
to
do
and
you'll
find
that,
like
the
third
bullet
point
is,
is
decoupled
Basile
to
strictly
function
as
an
override,
ER
I?
A
A
That's
a
that's.
The
last
bullet
point
which
just
says
like
create
a
well
create
a
meta
container
right,
that's
been
with
a
thing
that
I've
wanted
opined
for
and
if
we
have
the
meta
container
that
builds
the
packages
into
the
meta
container.
That
would
give
us
the
kind
solution
for
doing
package
installs.
A
The
the
last
bonus
thing
you'll
see
in
that
checklist
is
like
create
a
meta
installer
container
which
which
can
host
local
debin
rpms
repos
to
allow
for
a
local
install.
So
you
basically
call
docker
run
I
called
a
COO
BTM
er
right,
and
it
would
just
do
a
local
install
for
any
standard,
distro
installation.
It
would
do
the
right
thing
right.
He
would
have
to
have
a
script
inside
of
there
that
basically
had
the
local
packages
installed
in
that
container.
That
just
basically
did
local
install
from
that
environment
through
this
works.
A
Fine
with
I
have
to
double
check.
How
apt
works
with
that
stuff.
A
But
there's
all
you
need
is
a
system
to
unit
file
and
some
binaries
right.
That's
pretty
much
what
it
is
right.
So,
at
the
end
of
the
day,
it's
about
installing
these
things
in
the
proper
locations,
the
the
the
one
of
the
things
that
was
listed
in
that
list
too,
is
to
get
us
out
of
the
specialized
installation
tooling
like
there's
we're
not
really
doing
fancy
stuff
in
rpms
or
Deb's.
In
fact,
it's
like
we
are
doing
brute
stupid,
simple
things,
and
we
are
also
doing
them
wrong
so
like.
A
B
A
F
B
A
That's
the
reason
why
I'm,
focusing
on
test
and
release
automation,
I
think
are
super
important
for
this
next
cycle,
as
well
as
some
of
the
other
supporting
work
that,
like
Lucas,
is
doing
around
getting
the
configuration
more
formalized
in
the
system
right,
there's,
there's
two
major
high
priority
items
and
I
think
those
are
the
two
things
that
from
a
planning
perspective,
I'm,
gonna
push
for
I,
think
and
what
in
Lucas
and
I
will
chat
about
that?
Oh.
A
B
A
I
think
we
need
to
slowly
modify
it
over
time
to
get
people
into
the
new
frame
of
things
so
still
produce
the
package
of
this
normal.
But
this
time,
instead
of
producing
the
packages
in
the
release
repository
and
having
weird
CI
bifurcation
like
D
dupit
put
it
in
something
like
n,
F,
F
and
F
p.m.
and
make
KK
be
the
canonical
source
of
truth.
The.
B
B
A
A
Which,
I
think
is
good
I,
think
I
think
this
will
also
simplify
the
released
repository
to
basically
say
here
are
the
artifacts
from
KK.
Let's
sign
and
publish
the
artifacts
right
and
it
gets
the
release
repository
out
of
the
Auto
Bild
generation
bits
that
it
doesn't
doesn't
make
any
sense
for
it
to
be.
There
cause
it's
not
versions
right,
yeah,.
A
We
can
discuss
it
here
because
we're
the
folks
that
are
gonna
be
executing
on
it,
or
here
I.
We
think
we
should
mention
it
to
bend
as
well.
It's
also
gonna
be
used
in
other
things
like
the
cluster
API.
Has
this
bootstrapping
problem,
we're
currently
uses
mini,
cube
and
kind
is
just
way
faster
and
cleaner
and
simpler
for
that
bootstrapping
scenario.
A
Part
of
this
will
be
like
we
need
to
get
sick
cluster
lifecycle,
folks,
who
are
actively
contributing
and
vetting
and
verifying
this
as
part
of
the
owners.
We
have
a
stake
in
this
fight,
a
big
one.
So
once
we
are
active
owners
and
maintain
errs
we'll
work
with
them,
absolutely
we'll
try
to
prioritize
that
at
the
same
time,
it's
like
any
other
portion
of
the
communities
project.
No
one
person
owns
it.
It's
it's
a
community
right
so
and
it
should
never
be
a
single
ownership
thing,
because
that's
just
a
recipe
for
disaster.
A
We
should
just
do
standard
operating
procedure
that
we
do
for
everything
else
and
we
pass
we
function
as
a
sig.
We
always
come
up
with
a
plan.
You
log
the
issues
we
talk
about
things
openly
in
public
and
loop
in
the
stakeholders
whenever
we
need
and
we've
never
really
disenfranchised
or
explicitly
disenfranchised
folks
in
the
project.
So
I
do
know
that
sometimes
we've
ruffled
feathers
with
CI
automation,
stuff,
but
that's
not
necessarily
our
problem.
B
B
So
the
release
planning
is
not
today.
I
just
wanted
to
say
that
it
seems
that
in
the
we
had,
we
had
a
section
in
the
survey
where
people
where
people
can
write
comments
that
feature
a
request
directly
to
us
and
it
seems
like
the
the
automated
H
a
is
like
the
biggest
one.
I
just
wanted
to
mention
this
to
Fabrizio.
A
B
A
F
A
Totally
off
on
an
island,
so
it's
part
of
the
buff
and,
as
part
of
you
know,
just
being
standard
Singh
leads.
We
should
always
make
sure
that
folks
who
are
not
present,
you
know
we
don't
make
decisions
for
the
folks
are
not
present.
We
can
make
plans,
but
then
we're
going
to
follow
up
afterwards.
I.
B
A
B
A
Sounds
good:
are
there
any
while
I
still
have
two
minutes
11
minutes?
Are
there
any
call-outs
that
folks
would
like
to
make
explicitly
for
the
talks?
There's
I
have
a
sink
cluster
life-cycle
talked
during
the
contributor
summit,
where
cuvee
teams
a
part
of
it,
but
that's
there's
a
bunch
of
other
beats
there,
but
we
do
have
a
comedian
deep
dive
talk.
Are
there
specific
items
that
folks
want
me
to
pay
extra
special
attention
to
to
highlight
it
either
now
or
in
the
future?
Obviously
hav,
one
of
them
so.
B
The
Tim,
pepper,
pepper,
did
a
LTS
talk
at
China
and
something
interesting
he
did
is
basically
he's
broke
the
meeting,
the
the
talking
to
a
couple
of
portions
and
the
second
portions.
He
turned
it
into.
He
made
it
into
a
Q&A
with
the
audience
by
gathering
feedback
from
users
what
they
think
about
a
certain
topic,
but
the
topic
in
this
case
was
healthy.
Is
it
just
a
proposal
if
you
want
to
interact
more
with
the
audience
like
what
do
these
think
about
Canadian?
That's,
like
only
for.
A
Me
particularly
I
think
I
prefer
that
I'm
much
more
colloquial
and
I
do
not
actually
like
talks
in
general
I
like
to
like
burn
through
here
the
highlight
reel
and
then
actually
have
a
conversation
but
Liz.
What
do
you
think
I'm?
Actually,
gonna
pull
people
on
stage?
What
do
you?
What
are
your
thoughts
there?
My.
A
F
What
sir,
if
our
venue
is
just
like,
if
our
venue
is
like
pretty
small
and
cozy
I,
think
that
could
work,
but
we'll
need
good
moderation
because
otherwise,
it'll
just
be
you
know
so
God
and
some
some
dudes
standing
up
being
like.
Let
me
explain
everything
I
know
about
caperna
needs
it,
the
form
of
a
question
for
which
is
I,
think
the
failure
mode
of
audience
questions
we're
all
familiar
with.
That
is
hilarious.
F
F
F
A
That's
not
the
way
it
goes
for
there.
There's
two
tracks
at
at
kubernetes:
koukin
one
is
the
developer
track
and
the
other
is
sort
of
the
talk
track
and
for
this
track,
it's
traditionally
since
epoch,
released,
has
been
mostly
attended
by
people
who
have
like
best
interests
are
doing
development
or
they're
integrators
or
they're
they're
people
who
are
directly
using
this
on
a
day
to
day
type
of
thing.
So
they're
there
they're
much
in
the
know
or
they're
people
who
are
initial
folks,
who
are
doing
that
initial
work
to
get
started
on
it.