►
From YouTube: kubeadm office hours 2019-11-13
A
Hello,
everybody-
this
is
the
sequel
of
psycho
Cuban
mafia
showers
today
is
the
13th
of
November,
and
we
are
one
day
before
cold.
Freeze
of
117
I
want
to
share
that
general
document
in
chat.
Please
have
your
names
to
the
list
of
participants
and
also
your
potential
agenda
topics
all
right.
Let
me
share
my
screen.
C
A
No
problem:
alright,
let's
proceed
with
the
PSCs.
We
have
a
bunch
of
them
as
well.
So
this
is
the
release.
117
timeline
I'm
quickly
going
to
go
over
this,
so
it
helps
me
as
a
husband
freeze,
past
cold
freeze
is
tomorrow.
The
14th
and
cold
fish
is
going
to
end
when
cold
towel
begins,
and
this
is
a
22nd
of
November
and
then
we
have
like
a
one-week
for
cherry
pick.
So
this
is
like
one
week
after
code
freeze,
we
do
code
Tao
when
big.
A
When
is
the
period
when
we
have
to
start
cherry-picking
fixes
to
117
branch,
because
117
branch
splits
from
master
at
that
point,
they
will
have
a
cherry-pick
deadline
for
that
and
the
release
is
on
the
9th
of
December,
pretty
tight
schedule.
I'm,
not
sure.
If
you
have
any
dogs
pending
I,
don't
think
we
have
anything
in
terms
of
dogs,
I
already
coordinate
with
enhancements
people,
so
we
drop
the
we.
A
D
A
Yes,
if
something
is,
if
you
want
to
also
document
the
extension
you
are
adding,
you
basically
have
to
send
appear
for
the
page
with
the
cube,
ADM
self
management.
Ok
have
another
look,
but
you
have
to
send
it
for
deaf
won
7-1
17
branch
that
don't
send
it
for
master.
Ok,
thank
you.
Alright,
this
is
the
timeline.
A
A
A
Another
PSA
is
that,
which
is
a
very
good
news
for
us
is
the
cubed.
I
am
no
longer
depends
on
KK
I
just
found
the
depth
check
yesterday
and
we
are
completely
decoupled.
At
this
point.
I
notified
the
the
code
organization
projects
of
cigarette
texture
that
they
were
also
very
happy
about
it
now
I'm
going
to
close
this
issue.
A
This
is
the
import
boss,
TK
that
we
had,
because
we
completed
all
the
tasks
that
we
had
in
this
document
and
at
that
point
we
should
create
you
issued
that
is
4
cubed
I'm
moving
out
and
we
have
a
list
of
tasks
there,
because
the
picture
is
completely
unclear.
From
my
point
of
view,
we
don't
have
a
clear
picture:
how
to
tack
test,
build
and
release
cube
ATM
at
this
point
now,
when
we
move
out,
so
this
is
going
to
be
an
issue
that
we
should
discuss
also
with
sig
release
and
internally.
A
We
have
to
also
discuss
it
like
with
this
group.
How
are
we
going
to
manage
some
of
this
stuff
because
it's
complicated
another
PSA
is
that
the
white
listen
of
me
reports
that
sig
off
working
on
is
punted
to
118.
The
reason
for
that
is
that
the
feature
became
very
complicated
in
terms
of
like
decisions
where
to
exactly
enable
the
feature.
If
you
want
to
check
more
details,
click
this
link,
but
they're,
going
to
partially
enable
it,
but
it's
not
going
to
be
alpha
in
170
right.
This
is
these:
are
the
PSAs
any
questions.
A
D
Is
a
quick
one
because
we
already
discussed
it
on
Monday,
and
so
basically
we
are
going
to
add
a
note
on
the
page
that
explained
how
to
set
up
an
Instamatic
city
cluster
using
phases,
and
this
note
basically
makes
clear
that
this
is
a
an
example.
But
the
wave
for
or
the
Forex
penalty
city
is
a
year.
D
D
A
C
A
D
A
F
A
Adopt
this
document
so
fix
the
things
that
we
need
add
static,
pod
support-
maybe
it's
the
first
of
all,
but
also
there
is
a
problem
that
it
doesn't
have
a
config.
Yet
so
it
doesn't
have
an
API.
That's
a
much
much
larger
topic.
The
idea
has
to
be
clean
for
us
to
basically
the
configuration
there's,
no
configuration
file
for
a
CDI
DM
at
the
moment,
and
maybe
later
the
same
configuration
has
to
be
embedded
not
embedded
but
used
from
cube,
ATM,
okay,.
D
G
All
right,
Lee
you
coming
I,
don't
do
you
want
me
to
delete
that
the
calendar,
if
we
haven't
done
so
already,
didn't
you
and
Jason
delete
it
already.
I
may.
A
F
F
The
major
thing
that
I
think
would
make
this
a
no-go
without
changes
would
be.
You
mentioned
the
code
organization
and
the
dependencies
on
kaykai.
We
would
need
to
make
sure
there's
no
dependencies
on
KK
as
a
result
of
this
new
vendor,
but
this
implements
the
add-on
installer
configuration
along
with
some
feedback
with
regard
to
the
cap,
which
is
incomplete
and
provisional,
not
implementable,
so
it's
I.
F
F
There's
certainly
a
trade-off
if
it
doesn't
get
into
117,
then
we're
not
gonna
be
able
to
get
feedback
until
next
year.
So
I
would
really
appreciate.
If
anyone
has
the
cycles
to
do.
The
review
on
this
I
will
be
at
coop
Khan
and
the
contributor
summit
the
entire
week.
So
I'm
open
for
a
face-to-face
demos
and
conversations
regarding
this.
B
F
A
A
F
E
H
F
I
think
that
that
is
fair,
I,
I,
agree
and
it'll
give
more
time
for
the
add-ons
ecosystem
to
have
some
more
work,
because
this
feature
set
is
really
there
for
something
that
doesn't
fully
exist.
Yet
right,
like
there's
no
default
behavior
in
this
patch
here
to
replace
the
behavior
disables
right.
So
it
allows
you
to
like
install
things
like
a
CNI.
You
know
and
and
all
of
these
creative
things
with
customize.
But
there's
no
like
solution
right
now.
A
F
I
think
that
the
cap
is
not,
it
doesn't
enumerate
it
in
the
way
you're
saying
the
information
is
in
there
in
more
of
a
user
story,
descriptive
fashion,
because
it
did
get
pretty
detailed.
There's
feedback
on
that
cap,
though,
that
I
haven't
gotten
back
to
yet
I
just
haven't
had
the
cycles
to
push
this
all
the
way
through.
So.
A
Detail,
it's
really
since
better
to
come
for
the
next
cycle.
I
was
wondering
who
is
going
to
be
able
to
talk
with
you
at
cube
con,
so
we
have
Yahoo
and
Fabrizio
there.
Rusty
is
not
going
I'm.
Also
not
going
Tim
is
going,
so
maybe
you
guys
can
have
a
table
discussion
to
basically
talk
about
the
implementation
details
of
this
mm-hmm
yeah.
F
A
F
F
C
A
F
F
F
F
F
A
F
F
A
A
That's
sorry,
I
okay,
so
we
have
some
questions
related
to
like
how
often
are
people
upgraded,
upgrading
and
things
like
that,
and
we
were
wondering
whether
we
should
create
a
new
survey
in
a
copy
base.
The
old
one
potentially
changed.
Some
questions
may
be
introduced.
You
know
something
for
the
people
at
cube,
ADM
office,
sorry,
the
meeting
the
lecture
that
rich
and
young
were
going
to
do
and
for
people
to
scan
and
participate
on
this
in
the
survey.
E
It's
pretty
tight
to
get
on
the
logistics
nail
down,
I,
wouldn't
mind
going
out
at
the
blog
post
for
release,
I
think
to
get
feedback
from
a
survey
to
make
it
useful
I.
Think
we
need
to
think
about
some
of
the
key
questions,
because
one
of
the
key
feedback
items
we
got
from
last
time
was
that
hhha
we
already
kind
of
knew
that
right.
We
knew
that
a
chain
is
super
important,
but
it
was
super
clear
that
what
the
users
wanted
was
high
availability
and
we
knocked
that
on
that
one's
done.
E
I
Along
those
lines,
I
think
we
have
one
question
important
that
we
and
we
had
some
issues
with
certificates.
So
maybe
another
question
we
could
ask
is
for
how
long
have
you
been
running,
cube
ADM
in
our
living
cluster,
like
is
that
you
use
for
CI,
or
have
you
been
running
it
for
six
months
or
one
year
or
something
like
that
as
well?
Okay,.
A
It
looks
like
we
have
some
extra
questions,
so
this
list
of
question
has
to
be
amended.
I,
don't
think
we
have
the
time
to
prepare
this
for
Keep
Calm,
so
potentially
for
the
next
cycle.
We
are
going
to
work
on
the
new
questions.
Are
we
going
to
pack
a
new
survey
and
we
can
present
it
to
the
mailing
list
and
channels
and
other
areas
yeah.
E
G
A
E
A
A
A
So
add
workarounds
for
core
DNS,
so
this
is
a
back
port.
This
is
not
for
master,
so
we
shouldn't
care
about
this
much
Chris,
okay,
very
same
in
a
fix
where
this
is
mitigating
a
que,
walk
problem
and
it's
for
an
old
branch.
We
should
have
look
at
this
now
other
we
try
for
HCD
calls
for
visual.
It's
blocked
on
the
PR
BOTS
I
assume.
A
J
A
A
All
right,
basically,
this
pair
is
adding
retries
for
our
HDD
operations,
because
we
saw
failures
in
a
larger
HDD
Questers
like
six
nodes,
for
instance,
during
upgrade
of
replace
type
of
great
upgrades.
Very
you
removed,
odd
nodes,
add
new
nodes
and
for
which
one
folks
saw
failures.
So
that's
why
we
hadn't
very
dries.
A
Remove
this,
this
is
all
so,
he
tries
checks
and
this
PR
he
saw
that
cube.
Atm
join,
can
block
indefinitely,
and
this
is
logic
present
incubating
since
a
very
long
time
ago.
It
has
these
nested
paul
in
infinite
calls
in
a
couple
of
locations
which
are
problematic,
because
this
is
already
wrapped
in
a
white
group
and
I
think
this
pair
is
good
again
blocked
by
the
CI
I.
Think
at
this
point,
yeah
all
right,
necessary
tries.
This
is
a
so.
A
This
is
a
something
I
want
to
bring
as
a
topic
again
some
point,
because
we
are
consuming
customized
and
I
noticed
some
very
weird
stuff
that
the
customized
people
are
doing.
They
are
making
copies
of
the
staging
in
Co
British
colonies,
such
as,
for
instance,
case
dot,
io
/,
API
machinery.
They
are
making
a
snapshot
of
that
at
certain
version
and
they
are
placing
that
inside
the
customized
repository
which
is
taken.
You
know
taking
snapshots
of
API,
so
I
already
alarmed
a
bunch
of
people,
Phillip
responded
here,
I
think
we
have
here.
A
We
have
Daniel
and
Jordan
are
also
complaining
about
this
problem,
but
basically
we
I
want
to
bring
the
discussion
like
potentially
not
using
customizing
cube
a
tomorrow,
but
this
is
a
topic
for
another
meeting.
I
just
wanted
to
mention
that
is
I'm
not
liking,
some
of
things
they
are
doing
so,
even
if
we
import
a
very
small
portion
of
customized.
G
E
There
is
a
separate
Google,
Cloud
repo.
We
should
talk
with
Google
there's
a
separate
Google
Cloud
repo,
where
they
basically
forklifted
the
behavior
that
they
wanted
on
a
customized
into
their
own,
separate
library
and
I
do
know
that
a
couple
of
other
people
have
also
done
similar
things.
So
if
we,
if
we're,
if
you're
logging
an
issue,
let's,
let's
let's
loop
in
some
of
those
folks,
I
have
to
figure
out
where
they
were,
but
just
in
Santa
Barbara
would
be
able
to
connect
the
dots.
There,
too,
is.
F
A
I'm,
not
a
fan
of
customized
at
all,
but
yeah.
Let's
continue
the
discussion
another
time
because
you
have
a
bunch
of
stuff
to
cover
as
well.
Okay,
so
I
meant
the
hypercube
duplication
change.
This
is
a
small
change,
my
rusty
in
the
in
a
warning
that
we
have
for
hypercube
nowadays,
because
we
are
deprecated
in
it
and
it's
it's
already
merged,
actually
I
think
it
just
merge
like
when
I
opened
this
nice,
so
somebody
fixed
the
CI
I
think
so
this
is
a
bump
of
coordinates,
last-minute
bump.
A
It's
a
patch
bump,
the
sort
of
John
Sandeep
made
a
big
change
here
and
I
mean
Ross.
He
blocked
it
because
he
was
suggesting
that
the
affinity
of
the
coordinates
deployment,
which
is
not
a
great
idea,
so
close
recycle.
So
at
this
point
they
are
only
bump
in
a
punch
version
and
adding
this
lame
duck
chef
check
that
they
have,
which
is
fine,
I,
think
I,
think
this
is
waiting
on
Depp
approvers
at
this
point
and
I'm
going
to
watch
the
signal
after
this
patch
bump.
A
When
migration
fell,
so
this
is
yeah,
we
are
by
port
in
a
change
to
116,
I,
think
yeah,
so
so,
basically,
the
problem
here
is
that
some
users
found
that
the
they
have
custom
coordinates
built
and
they
want
to
upgrade,
but
they
are
completely
blocked
because
of
the
new
migration
Waja
Concordian
s.q
medium
for
coordination
cube
idiom
and
this
fix
is
going
to
unblock
them.
It
does
not
introduce
anything.
A
A
A
A
But
thanks
to
the
access
thinking
otherwise
possible
right,
a
clear
usage
of
Excel
OCA,
so
this
south,
so
we
have
some
kind
of
documentation.
Types
issues
here
and
I
usually
leave
this
until
last
minute,
because
some
of
these
can
be
fixed
for
the
K
website
master
branch.
So
it's
not
really
a
that's
something
that
is
particular
for
this
170
release,
but
I
just
leave
the
documentation
until
the
last
minutes.
If
somebody
has
opinions,
we
can
move
them
from
an
amount
on.
A
A
F
A
Link
to
the
other
document
and
the
other
document
we
have
to
summarize
like
what
are
the
steps
for
creating
using
examples
here.
Okay,
in
any
case,
we
can
keep
this
in
the
mouse
tall.
I
mean
you
should
probably
move
it
to
the
next
one.
At
this
point,
big
nobody's
going
to
have
time
for
this
right:
118,
update
release
cycle
and
D,
so
here,
I
didn't
have
the
time
to
update
the
release
cycle
and
D
document.
A
D
A
A
I
saw
it,
I
saw
it.
It
was
already
in
there
okay,
so
talking
about
the
test
and
Cuba
I'm
upgrade
backup
process.
I
think
Marek
assigned
him
so,
but
I
said
him
so
that
somebody
else
can
have
a
look
so
whisk
the
product
here
is
that
we
have.
You
know
the
backup
folder
that
we
create
on
the
HDD
kubernetes
do
an
upgrade.
It's
not
documented
how
to
use
this
folder.
A
A
A
A
Sorry,
probably,
okay,
this
this
one
is
still
pending
problem
with
the
load,
balancer
and
a
cheat
documentation.
We
got
a
big
discussion
here
and
a
lot
of
people.
Basically,
the
TLDR
here
is
that
every
call
provider
has
different
semantics
in
terms
of
how
they
give
you
one
load
balancer
how
this
role
panel
supports.
A
So
it's
not
clear
to
us
how
to
document
the
best
practices
for
using
external
load
balancers
on
our
hae
document,
and
somebody
has
to
sit
down
and
formulate
these
paragraphs
of
texts
that
explain
like
what
we
have
seen.
What
is
the
best
practice
in
like
probably
create
a
brief
outline
of
the
different
core
providers
because,
for
instance,
our
jury
doesn't
work
very
well,
I
think
out.
The
AWS
is
the
best
one
things
like
that.
I
Okay,
just
a
heads
up
I,
don't
want
to
get
too
much
into
details
of
this,
but
go
is
going
to
have
this.
How
this
let's
say
this
ping
interval
in
which
see
it
will
it
will
be
in
either
its
TP
to
end
point,
so
we
sell
it
for
load
balancers.
You
won't
get
connection
closed,
I
open
up
a
request
for
Q
Burnett
is
but
then
go
is
going
to
have
that
so
I
close
that
for
requests.
I
can
follow
up
on
this
one.
How
we
do
the
commentation
remember.
I
D
Be
careful
in
doing
piece
because
a
couple
may
should
remain
agnostic
to
the
different
route
rider.
So
in
my
opinion,
this
should
be
stated
in
the
form
that
kubernetes
require
a
load
balancer
with
such
requirement,
and
then
it
is
up
to
the
user
to
the
cloud
provider
to
tell
disculpa
provider
his
best
water.
A
D
A
D
A
Go
to
the
next
milestone.
I
usually
bring
issues
back
into
the
current
milestone
if
needed.
Okay,
provide
no
competition
for
NCIS
compliant
questions,
wiki
medium.
This
was
critical
edge
and,
however,
I
see
that
the
deal
is
not
working
on
that
so
I'm
going
to
reduce
the
priority,
and
also
internally,
we
had
these
discussions
that
we
really
don't
trust
like
the
CAS
report.
At
this
point,
it's
not
competent
into
determining
if
key
building
is
secure.
Not
so
I
am
going
to
reduce
the
priority
here
to
long
term.
A
A
Making
great
progress
window
support,
so
I
spoke
with
Michael
Michael
from
sig
windows,
like
correctly
some
folks
from
pivot
or
working
on
some
aspects
of
the
windows
deployment,
and
something
that
this
group
knows
about
is
that
we
don't
deploy
to
proxy
and
CNSE
ni
as
daemon
sets,
and
they
are
actually
working
on
fixing
this
part,
which
is
kind
of
annoying
and
they're,
going
to
allow
these
components
to
be
deployed
as
daemon
sets
of
windows.
However,
the
windows
installer
now
has
to
have
this
extra
component
that
it
deploys.
A
It's
like
a
binary
that
it's
a
proxy
service
manager
or
something
like
that
that
manages
these
demon
sets,
and
this
is
not
cube.
Atm
related.
We
are
not
going
to
care
about
it.
It's
going
to
be
managed
by
cygwin
those
folks
but
other
than
this
news.
Nobody
is
working
on
the
end-to-end
test
for
Windows
and
also
on
the
upgrades,
which
was
which
is
a
requirement
for
beta
Grand
Vizier
graduation.
So
we
are
moving.
This
118
I
already
spoke
with
Michael.
It's
fine.
D
A
What
one
potential
consumer
of
this
to
mind?
The
cynic
was
that
I
saw
some
mentions
about
the
sequester,
sorry,
the
the
coaster
epi
provider,
a
sure,
maybe
one
day
they
want
to
support
cube
ATM
with
Windows
Phones,
but
other
than
that.
I
really
don't
see
any
consumers,
and
also
we
have
to
discuss
it
with
sig
windows,
also
signalled
that
have
to
be
involved
because
they
know
how
to
provision
GCE
windows
VMs-
and
this
is
a
discussion
we
have
to.
A
A
A
Right
so
this
is
suppose,
I
Kosovo,
it's
a
book,
it's
not
a
critical
book.
So
again
we
have
to
point
it.
Do
you
remember
this
one?
It's
about
wheat
ain't,
the
control
plane,
notes
too
late
and
Warhol's
already
scheduled
on
them,
which
is
a
problem
in
the
cube
ADM
in
it
phases
in
the
way
we
basically
retained
too
late.
That's
a
problem.
A
A
Something
else
that
we
want
to
do
is
that,
like
my
accommodation
is
remote
using
porpoise
are
all
in
the
cube
ATM
repository
I
can
write
you
a
fifty
line,
go
program
that
is
going
to
do
a
better
job,
I
think,
and
it's
going
to
allow
us
to
indicate
what
dependencies
we
introducing
the
new
ripple
in
the
new.
With
all
the
report.
Let
me
change
so
I'm
going
to
close
this
create
a
new
issue
later.
A
D
A
I,
usually
prefer
keeping
the
old
ones
just
a
naming
stuff,
but
yeah
it's
okay,
evaluate
API
object,
fetch
retire
Roderick,
so
this
is.
This
is
the
the
logic
that
referral
added
for
the
config
map,
mutation
and
fetching
functions,
and
basically
for
the
folks
that
know
the
file
that
is
called
in
it
impotency
dot
go
this
file
needs
a
major
refactor
as
a
whole.
It's
problematic
it
pretty.
Much
has
arbitrary
different
timers
for
different
operations.
A
We
need
to
click,
create
a
story
for
all
the
functions
in
there
that
to
use
common
timeouts,
everything
should
be
organized
better,
in
particular,
the
the
function
that
we
are
doing
for
getting
a
config
map
uses
the
default
back
off,
which
is
pretty
much
one.
Second
in
total,
and
that's
a
problem
I
think
I'm,
seeing
problems
from
user
reports
about
this.
It
Italy
tries
for
one
second,
any
things:
I.
C
A
D
We
should
the
general
review
all
of
this
piece
of
code
and
try
to
standardize
on
I
think
each
API
call
for
a
TCC
do
for
the
case
ever
rapid
in
a
exponential
back-off
and
the
width
time
out.
So
we
are
resilient.
We
are
more
resilient
to
everything
resilient
to
everything
can
happen
on
the
network,
on
quantity
city
or
on
the
abuser.
So
I
think
that
now
we
are
asked
a
turret
approach
to
be
so
before.
I
organized
this
for
etcd,
but
I
would
like
to
make
this
Co
system.
A
Yeah,
we
have
to
have
backups
your
your
change
for
HDD
in
general,
can
be
you
can
have
a
utility
function
in
the
HDD
dot
go
file.
The
idempotency.
However,
problem
is
a
bigger
mess,
because
there
we
have
the
API
client
and
we
can.
We
try
to
retry
different
objects
with
different
ways,
even
and
I.
The
way
you
added
severely
tries
fine.
It's
come
it's
consistent
for
the
HDD
client,
but
for
the
API
server
client,
it's
it's!
It's
a
mess.
Yeah.
I
I
think
we
need
to
go
through
this,
and
maybe
it's
series
a
special
case,
but
the
unique
is
we
need
to
do
a
full
audit
of.
When
are
we
reaching
the
API
server
when
we
can
retry
how
we
how
we
can
have
for
one
story
regarding
this,
at
least
for
the
API
server?
How
calls
that
we
are
making?
Maybe
we
can
also
make
it
CD
into
that
same
story,
or
maybe
it's
a
different
one,
but
yeah?
We
will
pay
the
time
to
do
that.
Yeah.
A
A
A
A
This
is
probably
going
to
take
a
couple
of
years
to
solve
it's
problematic
in
terms
of
how
these
throws
name,
their
paths
for
certificates,
and
they
are
different
on
different
distance
and
also
we
generally
mount
very
top
level
paths
like
this
one
or
this,
and
we
ideally
want
to
stop
doing
that
in
our
static
pods
but
yeah.
This
is
next
spinach.
A
F
A
E
A
Machine
read
about
4
billion
commands,
so
this
is
still
some
117
I
spoke
with
the
release
team.
Today
we
are
going
to
release
alpha
came
out
for
support
because
we
already
have
the
back
end,
but
cubed
M
upgrade
plan
command
is
not
ready
yet
so
we
only
have
the
cube
ADM
config
images
command
supporting
this.