►
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
Okay,
hey
folks,
this
is
a
cluster
api
office
hours
on
the
2nd
of
february.
We
are
writing
to
the
cncf
code
of
connect,
so
please
be
nice
to
each
other.
Please
use
a
resense
feature
that
you
want
to
say
something
yeah.
If
you
have
anything,
you
want
to
talk
about,
feel
free
to
add
a
new
topic
here.
If
you
don't
have
access,
you
can
join
the
cluster
lifecycle,
google
group,
which
is
somewhere
up
here,
yeah,
and
that
should
be
it.
So,
let's
start
so
first
point
open
proposal.
Readouts.
A
Okay,
then
next
up
new
attendees
is
anyone
new
and
or
maybe
not
so
new
and
wants
to
introduce
themselves
feel
free
to
just
unmute
yourself
and
speak.
B
Hey
hello,
I'm
new
guillermo.
I
work
for
aws,
specifically
in
eks
anywhere
yup,
so
nice
to
meet
you
all.
A
Welcome
unless
you
also
have
a
topic,
a
few
minutes,
perfect,
okay,
so
then,
let's
start
with
discussion
topics.
So,
first
of
all,
I
called
it
upcoming
and
recent
releases
and
for
me
to
just
expand
on
whatever
I'm
saying
if
I
miss
something
so
the
1.1
release
is
planned
for
a
right
after
that
meeting
and
we
had
some
patch
a
first
round
of
patch
releases
yesterday.
A
So,
essentially,
all
old
versions
of
classic
cuddle
were
failing
now,
because
they
weren't
able
to
download
search
mentioner
affected
are
everything
above
0.4
and
for
pizzu
just
released
this
afternoon.
Fixed
versions
of
0.4
and
1.0,
and
the
upcoming
1.1
will
also
include
that
fix.
So
for
all
providers
which
are
trying
to
to
capture
1.1.
That
really
should
fix
it
again.
A
C
Stephan
only
a
quick
shout
out
to
you
you've
already
killian,
that's
jumping
inapp
and
in
fixing
the
supplemental
issue
in
two
hours.
Basically,.
D
B
Yeah,
so
this
is
my
first
time
presenting
a
proposal
here.
I'm
not
sure
what
is
the
right
process
here
like
do
we
give
folks
a
few
minutes
to
read
it?
Do
I
go
over
the
document?
Real
quick,
or
how
do
I
do
this.
A
Good
question
I
would
propose
maybe
something
like
giving
a
rough
overview
over
the
proposal
and
probably
then
some
kind
of
async
review.
First,
I'm
not.
E
A
Some
other
regular
hosts
here
differing
okay
sounds
okay,
so
I
think
I
can
give
you
rights
to
present,
or
I
can
just
open
the
post
for
you
up
to
you.
B
Yeah,
if,
if
you
don't
mind
open
it,
that
would
be
great,
no
problem.
B
Yeah,
it's
realized
that
google
docs
doesn't
understand
markdown.
So
if
you
prefer,
I
can
just
convert
this
and
then
paste
the
right
format
later,
but
this
should
be
enough
to
go
over
it.
Quick
summary,
so
this
is
a
very
simple
extension
on
how
cluster
resource
sets
work.
It's
just
adding
a
new,
a
new
mode,
this
mode
and
how
it
works.
It
was
kind
of
already
mentioned
in
the
renal
proposal
for
cluster
resource
sets.
B
It
was
like
left
out
of
the
scope,
but
it
was
mentioned
mostly
how
it
should
work,
and
if
you
look
at
implementation,
it's
like
very
clear
that
it
was
into
something
like
this,
so
the
idea
is
there
will
be
a
new
mode
called
reconcile.
B
It
will
just
check
the
hash
of
the
object,
definitions
that
we
are
already
storing
in
the
cluster
resource
binding
set,
and
then,
if
the
has
has
changed,
then
we
just
reapply
all
those
resources
in
the
targeted
clusters.
B
There
are
some
clarifications
about
drift
and
detecting
drift
between
the
clusters,
the
research
definitions
in
the
management
cluster
and
the
actual
objects
in
the
target.
Warlock
clusters,
this
dog
in
on
purpose,
leave
those
an
address.
Even
though
it
commands
some
possible
solutions
more
than
happy
to
expand
this
if
we
want
to
cover
those
in
this
mode,
but
in
my
opinion
this
is
the
simplest
approach
that
solves
the
problem.
B
A
Yeah
any
comments
already,
I
assume
most
will
be
async
after
the
fox
redder
document.
C
First
of
all,
thank
you
for
for
this
proposal.
Second,
maybe
it's
part
of
the
drift
this
past
the
discussion,
but
what
happened
if
something
gets
deleted
from
the
resource
that
I'm
I'm
applying.
B
C
B
Good
question
so
for
deletions
specifically,
this
doesn't
address
that.
It's
also
called
in
some
of
the
clarity
classification
paragraphs
it
will.
It
will
detect
that
there
is
a
change
in
the
secret,
which
is
where
the
resources
are
defined
and
it
will
reapply
all
the
resources
defined
in
that
secret,
but
you
won't
delete
the
original.
Let's
say
that
you
remove
the
config
map
right,
it
won't
delete
the
configmap
from
the
clusters.
B
It
will
just
reapply
all
the
resources
that
are
defined
in
in
that
secret
and
if
those
have
haven't
changed,
then
the
end
result.
Is
you
end
up
with
the
same
objects,
but
it
won't
delete
it.
So
yeah
deletions
are
also
not
included
in
this
design.
They're
they're
called
out
because
they
won't
work
but
they're
not
including
the
same.
A
Yeah
thanks
yeah,
I
would
say
quite
a
few
requests
in
the
past
and
a
few
duplicate
tickets
about
that
mode.
So
it's
nice
to
see
it
addressed.
Okay,
any
other
questions,
I'm
pretty
sure
we'll
get
back
to
it
pretty
soon.
A
D
A
Okay,
I'll
just
give
you
a
host
that
should
be
enough:
okay,
cool,
okay,
so
stop
sharing,
and
then
you
should
be
able
to
share
sure
I
I
probably
should
have
given
you
a
co-host,
because
now
I'm
not
hosting
one
but.
D
Yeah,
can
everyone
see
my
screen
and
hear
me
probably
okay,
cool?
I
can
see
you
okay,
awesome,
so
in
this
demo,
we're
going
to
talk
about
a
new
command
cluster
tool
called
a
new
alpha
command
in
first
of
all
called
topology
plan
with
cluster
class
in
its
current
state.
D
The
goal
behind
this
new
command
is
to
help
users
to
adopt
cluster
class,
improve
the
usability
of
cluster
class
and
basically
give
more
insights
into
how
cluster
class
is
actually
working
underneath
the
scenes
and
how
the
topology
controller
operates
on
certain
inputs
that
the
user
wants
to
play
with
right.
So
there
are
many
use
cases
for
this
particular
command,
we'll
be
taking
a
look
at
a
few
of
them
in
this
demo.
D
So
first,
let's
imagine
I
am
a
user
who
is
designing
my
own
faster
class
right,
and
I
want
to
see
how
this
new
cluster
class
that
I'm
designing
will
will
turn
out
when
a
cluster
uses
that
particular
cluster
class
right.
So,
let's
take
a
look
at
that,
for
example,
so
I
I
have
two
files
called
new
plus
new
class
and
new
cluster,
and
these
have
the
definitions
for
the
cluster
class
that
I'm
writing
and
an
example
cluster,
that's
using
that
particular
cluster
class.
D
So
let's
see
how
we
can
use
this
command
for
a
user
to
figure
out
what
kind
of
bundling
objects
will
the
topology
controller
create
if
if
the
user
uses
this
particular
class,
so
what's
the
cuddle
alpha
topology
plan
and
then
the
sign
yeah.
So
I'm
passing
in
my
class
definition,
I'm
passing
in
a
cluster
definition,
which
is
using
this
particular
class
that
I'm
designing
and
I'm
specifying
an
output
folder
we'll
look
at
why
we
need
to
specify
an
output
folder
in
just
a
minute
but
yeah.
Let
me
run
this.
D
So
let's
take
a
look
at
the
output,
so
it
says
this
changes.
This
inputs
that
we
are
applying
will
touch
that
will
touch.
Something
called
a
new
class
is
acting
on.
A
new
class
is
acting
on
something
called
a
new
cluster,
and
applying
these
two
files
on
a
real
cluster
would
result
in
the
following
set
will
result
in
the
following
change
summary
right,
so
it
means
it's
going
to
create
all
of
these
objects
as
specified
here,
and
it's
going
to
modify
the
cluster
object.
D
So
it's
going
to
create
dot
cluster,
it's
going
to
create
the
kcp
object,
the
machine
deployment
object
and
so
on.
What
we
would
expect
the
topology
controller
would
do
when
creating
a
new
cluster
with
a
cross
class
and
to
get
a
better
insight
into
the
structure
of
the
objects,
because
I
am
a
user
who
is
designing
my
own
customers.
I
want
to
want
to
see
how
these
objects
turned
out.
D
We
actually
write
this
command,
actually
writes
the
outputs,
writes
the
objects
into
onto
disk
into
the
following
paths.
Both
of
them
are
written
into
the
output
path
as
passed
into
the
initial
command,
so
we
can
take
a
look
at
that,
so
it
creates
two
folders
called
created
and
modified
so
under
created.
It
prints,
it
stores
the
yaml
objects
of
the
final
objects
that
were
created.
As
a
result
of
the
reconciliation
and
for
modified
object
objects,
it
gives
us
the
original.
It
gives
us
the
modified,
basically,
the
final
state.
D
It
gives
us
a
json
patch
between
the
two,
and
it
also
gives
us
a
diff
between
the
two.
So
this
gives
the
user.
This
gives
any
user
the
insight
on
okay,
how
particular
things
are
working
or
how
particular
inputs
will
modify
the
state
of
my
cluster
and
so
on.
Now,
let's
shift
gears
and
take
a
look
at
another
use
case.
D
D
So
if
I
run
that
it
tells
me
that,
okay,
if
we
apply
this
particular
changes,
it's
going
to
change
the
kcp
object
and
it's
going
to
delete
one
of
the
machine
dependents
and
we
can
go
into
the
modified
folder
and
then
see
what
are
the
changes
it's
doing
on
that
kcp
object
and
then
gather
and
then
say
now,
let's
say
instead,
I
want
instead
of
modifying
the
topology,
I
try
to
modify
the
cluster
class
itself
right.
So,
let's
take
a
look
at
that.
D
Okay,
so
I
have
something
called
a
shared
class
already
installed
on
the
management
cluster
and
now
I'm
trying
to
apply
a
change
to
that
share
class,
and
I
want
to
see
how
that
would
affect
the
state
of
my
clusters
right.
So,
if
I
run
this
now,
this
output
is
interesting.
It
tells
us
okay,
applying
this
change
would
modify
or
would
touch
would
affect
the
share
class,
and
it
would
also
affect
the
following
clusters.
D
So
here
we
see
that
modifying
that
one
particular
class
would
affect
multiple
classes
and
it
just
listses
all
those
buses,
and
in
this
case,
since
there
are
multiple
clusters,
it
doesn't
give
us
like
a
detailed
output
of
what
changes
on
each
cluster
and
we
can
force
it
to
do
that
by
passing
in
the
cluster
parameter
and
then
asking
it
to
specify.
Okay,
give
me
the
detail,
output
of
one
particular
cluster,
let's
say
the
first
cluster,
and
then
it
would
tell
us
okay,
applying
this
class.
D
Applying
this
changes
on
cluster
class
would
affect
the
first
cluster
in
the
following
way.
Basically,
it
would
change
the
kcp
object
and
we
can
go
and
take
a
look
at
that
modified
object
to
see
what
actually
changes
now
again
left
more
to
another
use
case.
So
technically
the
users
can
modify
the
cluster
class
and
then
roll
and
then
basically
affect
multiple
clusters.
But
cluster
api
recommends
that
users
instead
follow
the
operation
practice
of
rebasing
clusters.
D
What
that
means
is
if
I
am
a
user,
and
if
I
want
to
change
the
cluster
class,
I
should
instead
create
a
copy
of
the
cluster
class
with
the
changes
I
care
about
and
then
rebase
my
cluster
to
the
target
cluster
vb's,
the
cluster
that
I
want
to
the
target
cluster
and
let's
take
a
look
at
and
see
how
this
command
can
help
us
in
figuring
out
if
a
rebase
is
going
to
work
or
not
right.
So
cluster
cuddle
alpha
topology
plan
minus
f
e
base.
D
So
I
have
a
cluster
called
target
cluster
that's
already
installed,
and
I
have
the
from
cluster
that
it
is
that
it
already
is
working
with,
and
I
have
a
two
cluster
that
it's
already
that
is
trying
to
rebase
two
and
I
am
applying
the
change
where
the
cluster
is
now
changing
the
cluster
class
from
the
from
cluster
from
class
to
the
two
plus.
So
if
I
run
that
it
will
tell
me
okay,
so
none
of
the
cluster
class
will
get
affected,
but
your
target
cluster
will
get
affected
and
it
will
get
affected
with
these
changes.
D
So
the
target
plus
is
going
to
have
all
of
these
changes,
but
this
command
also
helps
us
in
identifying
if
a
rebase
operation
is
dangerous
or
if
it's,
if
it's
going
to
fail
certain
things
right.
So
let's
I'm
trying
to
rebase
the
clusters
to
across
the
class
that
is
not
compatible
with
its
existing
cluster.
D
So
if
I
try
to
do
that,
it
shows
us
that
the
rebase
operation
would
have
certain
negative
effects
on
it
and
in
this
example,
it's
basically
saying
that
the
cluster
that
we
already
have
is
using
a
machine
and
in
the
classes
that
we
already
have.
There
is
a
machine
deployment.
That's
using
something
called
a
default
worker
machine
deployment
class,
and
this
default
mercur
worker
machine
deployment
plus
doesn't
exist
in
the
new
target
plus
the
bad
class,
and
therefore
this
warns
us
and
lets
us
know.
D
Okay,
so
the
rebase
operation
would
fail
if
we
move
our
cluster
from
from
our
existing
cluster
class
to
a
target
cluster
class.
There
are
many
other
use
cases
for
this
command,
so,
for
example,
we
can
also
imagine
what,
if
I
change
a
particular
template
involved
like
a
machine
template
or
a
kcp
template
like
we
can
figure
out,
okay,
what
all
cluster
classes
will
this
template
effect?
D
What
all
clusters
will
be
specific
and
so
on,
and
this
can
also
be
used
in
cases
where,
let's
say
I'm
trying
to
use
patches,
and
I
want
to
iterate
on
patches
and
see
how
patches
are
going
to
affect
my
cluster
class
or
how
patches
are
going
to
affect
my
cluster
and
how
can
I
consume
them
and
quickly,
iterate
and
update
my
patches
to
see
that
it
works
as
I
as
I
would
want
them
to
and
many
other
use
cases
there
is
a
documentation
for
how
to
use
this
particular
command
within
our
capi
book
right
now,
it's
in
the
main
branch,
so
please
go
ahead
and
try
and
use
it
and
all
feedback
is
welcome
and
that
that
that
is
all
for
this
presentation.
D
For
this
short
demo,
I'm
happy
to
take
any
questions.
If
there
are
any.
D
For
no
changes
would
the
output
folder
be
empty
yeah.
If
there
are
no
changes,
the
output
folder
will
be
empty
if
nothing's
created
or
if
nothing's
modified
the
output
folder
will
just
remain
empty.
C
Yeah,
first
of
all,
thanks,
I
I
think
that
it's
coming
super
powerful.
It
is
the
first
time
that
that
we
kind
of
expose
we
make
it
visible,
what
happens
by
our
conciler,
so
it
is
put
super
far.
I
encourage
people
with
that.
C
I
started
with
cluster
class,
and
now
there
is
documentation
about
writing
cluster
class
or
changing
process
class
like
stuff
like
that
to
go
and
basically
play
with
the
alpha
the
topology
plan
command,
because
they
will
basically
understand
what
will
be
the
effect
of
their
change
without
trying
without
trying
them
live
in
cluster.
So
this
is
super
powerful.
C
We
we
kind,
we
also
try
to
explain
all
of
these
in
documentation
so,
but
there
is
really
a
lot.
So
we
are
really
looking
forward
for
feedback
how
to
improve
documentation,
how
to
make
things
more
more
clear
for
the
newcomers.
D
And
one
one
final
note
to
close
it
out:
this
command
can
actually
be
used,
even
if
you
don't
have
a
management
cluster,
so
you
can
just
pass
into
the
file
with
the
class
with
the
cluster
and
it'll
just
work
right.
You
don't
even
need
the
acting
management
cluster
to
back
it
up.
So
that's
even
easier
for
someone
to
just
start
and
try
and
see
how
cluster
class
operates.
A
Great
stuff,
can
you
give
me
backhoes,
I
think.
D
A
E
I
think,
yesterday
it's
kind
of
want
to
get
feedback,
so
the
the
issue
is
that
you
know
today,
it's
possible
to
have
a
a
machine
deployment
that
is
updated
to
a
version
that
is
ahead
of
the
control
plane
and
for
the
you
know,
for
all
the
controllers
to
to
begin,
reconciling
that
into
for
the
infrastructure
machine
to
be
created
and
for
the
bootstrap
to
be
initiated
and
for
that
to
eventually
fail
because
the
the
kublet
well,
it's
not
guaranteed.
E
But
you
know
in
case
of
like
1
21
versus
122.
If
you've
got
a
122
kublet
connecting
to
a
121
api
server,
it
won't
be
able
to
register
it'll.
It
will
fail.
You
know
trying
to
use
certain
certain
apis
that
I
think
have
been
removed
any
case
it's
possible
to
to
violate
the
diversion
sku
policy,
the
kubernetes
version,
skew
policy,
and
that
seems
like
a.
E
I
guess
what
you
know,
what
you
might
call
a
foot
gun
right.
It's
it's
a
way
for
the
user
to
you
know
to
inadvertently.
You
know
make
a
change
that
to
the
cluster
that
that
doesn't
you
know
they
can't
possibly
work
and
yeah.
I'm
just
interested
in
you
know
if
others
see
this
as
a
problem,
I've
got
a
a
pr.
E
C
Yeah
thanks
for
reporting
this
in
detail,
I
have
to
catch
up
with
this
stuff
because
I
was
busy
with
batteries
but
yeah.
We
discovered
something
similar
while
starting
with
cluster
class
cluster
class
is
the
the
first
point
where
we
have.
Let
me
say
overall
view
on
of
the
version
of
of
the
entire
cluster,
and
we
discovered
that
these
I
I
was
searching
for
the
issue,
but
I
I
I
need
a
little
bit
more
time,
so
we
find
out
these.
C
Basically,
there
is
this
possibility
that
you
are
lighted.
C
It
is
also
even
more
subtle,
so
you
can
basically
try
to
change
the
version
of
your
control
plane.
If,
if
I
remember
well
so
yeah,
we
discussed
how
to
get
this
fixed.
C
C
It
was
like
cluster
cluster
was
the
only
point
where
we
get
we
can
get.
Let
me
see
this
entire
view
of
the
of
the
cluster
and
those.
So
we
were
discussing
about
the
idea
of
surfacing
a
current
kubernetes
version
on
on
or
at
least
for
the
control
plane
on
cluster
level.
So
so
we
you
can
take,
let
me
say,
informal
decision
about
which
version
to
allow
stuff
like
that.
So
I
I'm
yeah
I'm
aware
of
this
problem.
C
I
don't
have
a
solution
yet,
but
thank
you
for
bringing
this
to
the
attention
and
try
to
make
it
is
getting
a
formal
shape.
A
Yeah,
I
agree
I
mean
on
a
cluster
class
level,
we
can
avoid
the
problem,
but
I
think
it's
on
on
high
level.
It
makes
absolutely
sense
so
also
try
to
address
this
in
the
base
controllers.
I
think
the
main
main
issue
we
had
at
that
point
was
like
yeah.
How
do
we
bubble
up
the
information
so
that
we
can
consume
it
in
the.
E
A
Right
parts,
without
I
don't
know
making
dependencies
from
the
booster
provider
to
the
control,
pin
provider
or
some
strange
things
like
that.
But
I
also
have
to
take
a
look
at
the
pr
first.
E
I
I
in
the
pr
actually
ended
up
using
the
the
contract
package,
which
came
out
of
the
cluster
class
work,
and
then
I
guess
get
it.
It
moved
out
of
the
topology
package,
but
that
that
made
it
that
made
the
pr
you
know
much
smaller
than
I
think
it
otherwise
would
have,
would
have
been
perfect.
A
Okay,
so
yeah,
I
think
we
take
a
look
soon.
Any
other
concerns
questions.
Otherwise
I
think
that
was
the
last
topic.
Oh
no,
there's
a
new
one.
Let's
see.
F
Yeah
I
added
the
last
one,
so
so
in
some
in
some
of
the
providers
I
started.
F
I
actually
started
looking
at
the
way
cubelet
talks
to
the
api
server
and
it
seems
like
in
our
cp
cube
lists
in
the
control
plane,
we're
using
the
control
plane
endpoint
to
reach
out
to
the
api
server.
This
has
this
can
have
like
some
weird
drawbacks
when,
for
example,
you're
rolling
out
a
new
kubernetes
version,
a
new
machine
pops
out
with
a
new
cubelet
version,
say
123,
and
it
needs
to
talk
back
to
the
api
server
and
it
can
talk
to
pretty
much
any
api
server.
F
It
might
be
the
1231,
it
might
be
the
122
one.
So
this
has
like
this.
That's
the
basically
the
major
drawback.
The
second
one
is
in
some
of
the
providers.
When
you're
using
private
clusters
to
front-end
things
within
internal
load
balancers,
you
might
be
facing
a
hairpin
issue
where
the
back
end,
that
is
part
of
the
load
balancer
might
not
be
even
able
to
talk
to
the
load
balancer.
So
I
think
that
it's
at
least
true
for
azure
and
aws.
F
It
may
be
also
true
for
for
cabgie,
but
I
need
to
double
check
that.
So
I
don't
know
if
folks
have
considered
that
and
if
they
have
to
surround
it.
C
C
Not
so
so
long
time
ago,
in
order
to
make
sure
that
the
current
configuration
on
how
kubrick
talks
cannot
create
versions
too,
that
this
was
the
problem
that
we
was
we
were
talking
about.
So
currently
there
are,
there
are
differences
between
worker
nodes
and
control,
brake
nodes
and
and
those
differences
are,
are
meant.
C
We
were
introduced
to
basically
fix
a
problem
when
a
api
version
when,
when
new
cube,
api
version
deprecate
some
api
and-
and
so
we
we
fixed
this.
If
the
you
have
more
detail
about
the
problem
that
that
you
are
like
you
are
liking,
I
would
like
to
take
a
look
but
yeah.
We,
we
already
eat
some
problems
in
that
area
and
it's
kind
of
critical
to
to
get
to
the
right
session.
So
looking
forward
to
to
understand,
what's:
what's
not
working.
F
Yeah,
I'm
currently
working
on
a
document
to
compile
this,
hopefully
for
the
next
office
hours.
I
should
be
able
to
share
it.