►
From YouTube: sigs.k8s.io/kind 2019-11-04
A
A
A
A
A
This
week
that's
fine
did
anyone
here
have
anything
they
wanted
to
bring
up.
B
I
got
one
thing:
I
don't
have
results
for
it
yet,
but
there
was
a
mention
on
the
channel
from
of
the
doctor:
docker
for
Windows
just
released
a
new
version
using
the
WSL
two
subsystem,
and
so
I'm
gonna
take
a
quick
run
through
the
QuickStart
on
that
and
see
if
there's
anything
that
I'm
basically
impacts
the
steps
that
we
already
have
there
and
if
it's
different
enough,
I'll
probably
just
create
another
topic
page
for
that.
So
that
way
people
can
choose
between
using
the
stable
version
or
the
wl2
version.
A
Yeah,
on
my
end,
I'm
trying
to
get
some
last
fixes
and
clean
up
in
you
know
before
cube
con.
So
we
have
everything
good
to
go
and
kind
of
a
demand
for
this
meeting.
But
Patrick
is
here.
You
will
need
to
go.
Moreover,
the
workshop
stuff
I
know:
Duffy
has
gotten
a
job
start
on
that
I
feel
bad
for
her
behind
I
am
on
that,
but
I
think
you
know
we
saw
of
time
it
should
be
to
that
everyone
who
will
be
there
bulk
know
all
this
quite
well.
I,
probably
I've,
been
thinking
about
I.
B
A
B
A
Couple
gigs
kubernetes
is
main
if
they
like
it.
Also,
it
really
depends
on
like
how
they
bill
I.
Don't
think
we
want
to
go
the
Basel
drought
this
time.
So
if
we're
not
doing
that,
then
they're
definitely
downloading.
For
example,
the
kubernetes
build
image
and
the
build
container
last
time
I
checked,
which
has
been
a
while
losing
north
of
two
gigabytes
because
it
has
cross-compiled
toolchains
for
everything.
A
B
A
B
A
Alternatively,
we
could
like
we
could
make
a
VM
image,
but
I
kind
of
think.
The
whole
thing
I
think
we
want.
It
would
be
nice
if
we
think
we
can
get
people
to
run
this
ahead
of
time.
To
show
I
mean
part
of
the
point
is
you
should
be
able
to
do
this
on
your
laptop
yeah,
yeah
yeah,
and
it
should
actually
work
fine
on
conference
Wi-Fi.
If
we've
done
things
once
mm-hmm.
A
B
Okay,
I
did
have
one
other
other
question,
so
Ellen
and
I
have
also
been
looking
at
being
able
to
join
another
node
to
a
kind
cluster,
and
specifically,
this
would
help
with
getting
coverage
on
Windows
or
arm
and
without
having
to
rely
on
an
emulator,
yeah
and
so
using
port
for
work.
So
using
VX,
LAN
and
port
forward
seems
to
be
the
easiest
way
to
go
there,
because
then
you
don't
have
to
do
with
any.
B
A
B
A
Yeah
I
haven't
tried
specifically
overriding
the
advertised
address,
but
it
shouldn't
the
same
as
overwriting
other
fields.
It's
especially
since
last
I
checked
it.
It
is
just
a
it's
just
a
string
field,
so
it
shouldn't
need
like,
for
example,
a
JSON
patch
to
meet
a
specific
operation.
Just
writing
a
patch
that
matches
it
should
work.
Okay,.
B
A
Should
be
a
little
bit
easier
and
head
as
well
in
head,
we
have
a
new
patching
runtime
and
specifying
the
API
version
is
optional
mm-hmm.
So
you
could
just
specify
the
just
the
kind
field
on
the
patch
okay.
B
B
A
And
if
you
send
me
a
reminder
later,
I
can
follow
up
on
that.
What
I
am
currently
working
on
is
just
some
nasty
code
cleanup.
We
we
moved
off
of
using
like
la
giris
directly
to
make
it
a
little
bit
easier
for
embedding
and
things,
and
when
we
did
that
to
make
migration
easier,
we
made
a
little
package
that
just
has
like
a
global
logger
instance,
so
that
we
could
use
them
a
pattern
everywhere
and
plumbing
a
non
global
one
through
everywhere
and
getting
it
to
work
correctly.
A
It's
been
pretty
messy
like
without
we
don't
soon.
We
have
like
one
or
two
stability
fixes.
We
also
need
to
when
there's
still
some
flakiness
with
log
export,
for
example,
but
I'm
kind
of
hoping
to
get
all
that
in
this
week
and
then
give
him
the
proximity
just
like
make
sure
we
cut
a
release
before
cube
comm,
so
that
we
don't
inevitably
have
like
some
trivial
bug
fix
or
something
unreleased
before
we're
trying
to
have
a
bunch
of
people
workshop
with
it.
Okay,.
A
Yeah
we're
pretty
close,
but
there
just
a
few
things
like
that
that
are
not
quite
set
right
up.
We
also
Antonio
point
out
to
me
so
and
and
in
head
right
now
we
now
do
cube.
Config
management
like
you'd,
expect
from
like
say,
cops
or
UK
or
etc,
where
it
just
exports
it
and
merges
it
into
your
cube
config
file,
unless
you
set
like
dollar,
cube,
config
or
cube
config
to
point
to
a
different
location,
mm-hmm.
C
A
C
A
Very
mirja
below
so
we've
like
done
all
of
that,
and
then
Antonia
pointed
me
to
a
github
search
of
kind,
get
cube,
config
path
and
they
were
like
1300
hits
or
something
just
deleted
that
it
was
like
yeah.
You
should
just
migrate,
so
I'm
gonna
put
that
back
with
a
warning,
but,
like
you
should
migrate
off
of
rely
on
that.
But
just
have
it
like
guess
at
the
same
path
that
it
used
to
export
so
that,
like
most
people,
probably
won't
actually
break
okay.
B
B
Okay,
good
good,
that's
something!
I
was
about
to
look
at
today
because
with
the
docker
for
Windows,
stuff,
they're,
adding
and
they've
got
a
crew.
Bernays
15.5
service
set
up,
and
so
I
wanted
to
make
sure
that
I
documented
how
to
switch
the
cube
config
back
and
forth
between
that
and
kind
for
a
development
cluster
yeah.
A
It
should
be
a
lot
easier
in
when
we
cut
that
release,
because
it
will
just
be
a
context
in
your
normal
unless
you
specify
otherwise.
Okay
and
it'll
be
like
kind
dash
cluster
name
so
for
D
by
default
to
be
kind
kind,
which
is
a
great
name,
but
it
doesn't
mean
they're,
they're,
definitely
identifiable
now,
and
you
won't
have
the
nonsense
or
like
you,
switch
terminals,
and
you
forgot
to
export
dollar
cube
config
and
your.
A
B
A
A
Times
it
was,
it
was
much
easier
to
implement
correctly
without
heavy
dependencies,
then
just
use
a
different
file,
but
I
spent
some
time
very
extensively
mimicking.
How
client
go?
Does
this
and
writing
a
lot
of
tests
to
try
to
make
sure
that
we're
not
going
to
break
anybody's
config
files
and
I'm
pretty
calm
than
that
now?
But
we
realized
that
the
not
having
that
sub
command
is
going
to
break
a
ton
of
scripts
and
that's
probably
a
little
too
harsh.
We
should
probably
still
have
the
sub
command,
but
have
it
a.
C
A
C
A
Warning
that,
like
that
flag
is
gone,
so
it
just
need
to
do
a
couple
small
things
like
that.
We
should
be
we're
pretty
stable.
Now,
we've
things
are
running:
pretty
smooth
I'm,
actually
planning
to
put
out
a
PR
today
to
make
kind
of
pre
cement
blocking
and
current
IDs,
and
just
like
wait
for
and
like
send
out
of
the
node
and
wait
for
the
consensus
on
that.
But
I
think
we're
ready
for
that
and
I
want
to
get
that
in
before
code.
Freeze,
mm-hmm.
A
So
hopefully
that
will
also
mean
that
we
won't
code
to
the
cube
con
and
find
some
that
the
other
nice
side
effect
of
that
is.
We
won't
find
that,
like
you
know,
somebody
broke
kind
and
head
again
for
some
trivial
change.
Oh
yeah,
you
know
fix
kind
at
the
very
worst
we
can
tell
people
to
install
from
head
if
we
have
to-
or
we
can
cut
a
small
patch
or
something,
but
we
shouldn't
have
shouldn't
get
into
a
state.
We're
like
it's
not
possible
to
use
kubernetes.
A
C
A
Only
go
patch
up
kind
of
work
around
it
or
something
like
that.
We
will
defect
of
not
have
that
issue
is
going
forward,
which
will
be
nice,
so
hopefully
you
can
have
that
in
before
the
conference.
I
think
we
could
have
done
it
earlier,
but
I'm
I'm
gonna
do
it
now:
okay,
cool
yeah
and
I
will
try
to
mess
around
with
the
new
stuff
to
I'm.
Pretty
hopeful
the
the
WS
old
mode
should
be
pretty
nice
with,
like
the
memory
reclaim
and
everything
yeah.
B
B
A
B
A
C
A
A
A
Except
image,
build
pretty
much,
doesn't
actually
touch
docker
directly.
At
this
point
there
small
there's
a
provider
package
that
implements
the
provider
interface
and
everything's
behind
that,
so
most
of
the
code
is
just
doing
things
like
giving
a
note.
I'm
gonna
execute
some
command
on
it
to
like
yeah.
B
A
A
B
B
All
right,
yeah
yeah,
we
could
talk
about
that
more
later,
cuz
I
think
I'm
gonna
figure
out
what
you
know,
what
we're
actually
trying
to
solve.
At
least
you
know
with
the
different
variants
there,
because
I
don't
want
to
differentiate
things
too
much
from
the
Linux
workflow,
because
part
of
the
reason
why
I
like
using
wsl
is
that
I
can
reuse
all
the
same
scripts.
B
C
A
With
vm's
and
I
can't
really
do
that
when
I
have
a
bunch
of
containers
on
the
same
post,
with
the
same
kernel
without
like
with
swap
enabled
on
host
stuff,
like
that
there
there.
A
few
places
where
like
can.
The
privileged
container
thing
is
to
leaky
yeah.
If
we
have,
if
we're
able
to
just
like,
take
a
container
image
and
run
it
as
a
VM,
then
it
shouldn't
cost
us
too
much
to
do
that.
A
B
B
B
A
C
A
Important
to
anyone
so
far,
yet
yeah,
but
the
other
thing
being
there
like
if
docker
pulls
some
nonsense
with
their
command-line,
were
insulated
from
a
now.
It
shouldn't
be
too
hard
to
Migrator
forever
mm-hmm,
there's
other
than
the
build
code,
which
is
still
pretty
terrible.
It
means
we
work.
The
all
the
node
management
stuff
is
pretty
nicely
decouple.
Yeah.
A
C
A
Coats
been
like
edit
works,
whatever
almost
everything
about
it
was
wrong
I,
but
it's
just
been
one
of
those
things
where,
like
nobody
really
cares,
how
good
that
code
is
cuz,
it
doesn't
get
done
much
it's
just
like
it
builds
it.
It's
fine.
A
Yeah
and
I
mean
some
really
like
kinda
build
base,
I
mean
it
doesn't
really
need
to
be
a
thing
anymore,
like
it's
just
doing
a
docker
build
so
but
yeah
we're
more
cleaner
to
figure
out
I.
Think,
though,
we
are
also
getting
a
lot
closer
to
kinda,
be
much
easier
to
embed
and
the
api's
are
slowly
getting
less
terrible,
I
think
getting
more
stable
card
headed
that
way,
also
just
because
they
are
smaller,
like
random
stuff.
A
That
is
incidental,
is
not
the
API
anymore
and
we
don't
import
like
huge
dependencies
like
client
go
and
the
binary
anymore.
It's
actually
we've
got
kind
down
under
10
bytes
now
for
the
main
CLI.
Just
because
we're
not
importing
all
this
random
stuff
indirectly
that
took
a
bit
of
work,
but
it
is
much
more
now.
The
set
of
dependencies
is
pretty
much
all
things
that
any
kubernetes
project
is
definitely
gonna
have,
and
it's
not
super
version
sensitive.
Do
you
like
ya
mole
and
like
Cobra
and
PFLAG,
and.
C
A
The
the
like
power
of
the
core
stuff,
like
all
the
things
that
kind
of
map
to
the
command
line,
is
not
going
anywhere,
but
like
the
kind
of
the
names
and
like
pulling
through
logging
and
stuff
like
that,
is
then
rejiggered
and
cleaned
up
a
bit.
I'm
feel.
A
You
know
you
can
actually
like
in
the
like:
okay
I
want
to
like
pass
cube,
configures
like
this
path
or
something.
Instead,
you
know
and
we're
making
the
we're
starting
to
try
to
treat
like
the
commands
themselves,
also
as
more
of
an
API
so
like
I'm,
working
on
plumbing
through
like
standard
and
standard
out
as
things
that
are
injected
the
logger
etc.
So
put
one
more
global
stayed
and
and
like
everything
is
explicit
and
hopefully
getting
stable.
That's
part
of
why
I've
taken
so
long
in
this
release.
It's
like!
A
A
I
don't
know
it's
definitely
going
to
be
a
little
bit
annoying
to
move
on
those
things,
but
I
figure
if
we
can
kind
of
get
them
up
front
now
and
it'll
be
easier
to
get
to
a
point
or
happy
to
make
a
commitment,
and
hopefully
people.
Hopefully,
people
just
need
to
deal
with
migrating.
This
release
I
think
previously,
the
only
real
migrations
we've
had
have
been
the
images
and
similarly
getting
closer
to
just
declaring
that
stable.
A
C
The
migration
goes
for
us
because
we're
using
it
right
now
and
in
our
testing
pipeline
as
well,
it.
A
Should
be
pretty
easy,
but
it
will
require
some
work
like
if
you
want
the
previous
behavior
with
cute
config.
You
can
just
you
can
do
the
expert
cube
configure
before
you
create
the
cluster
and
you
can
code
the
path
that
you
want
like
dollar
home,
cube
kinds,
something
you
can
write
that
yourself
and
you'll
get
the
exact
same
behavior.
A
You
just
have
to
move
that
export
line
up
and
instead
of
using
the
sub
command
like
right,
the
path
you
want
like
that,
shouldn't
hoping
that
won't
be
super
difficult
and
then,
like
the
logs
flags,
you
can
use
the
same
one
for
this
release,
but
like
should
move
to
there's
now
a
verbosity
flag
and
there's
a
quiet
to
suppress
it's
everything,
to
bring
it
more
in
line
with
kind
of
the
K
luggage
stuff.
The
word
okay.
A
Think
so
I
mean
it
works,
the
same
as
like
cube
cuddle
itself.
So
there's
a
dash
dash,
cube,
cuddle
flag
or
create
delete
and
it'll
read
the
cube
config
and
if
you
don't
set
that
and
it
will,
if
the
cube
configured
isn't
set,
then
it
will
look
up
your
home
der
the
same
way
as
Conoco
and
it'll
take
home
duct
cubes
left
config.
It
follows
exact,
same
rules
as
to
cuddle,
but
the
plan
for
now
is
to
put
a
a
guide
over
that
in
the
police
notes:
okay,.
C
A
A
A
A
A
Yeah
I
are
like
similarly
things
like
so
like
the
API
for
dealing
with
clusters
directly.
Now
you
create
a
provider
currently
there's
no
options
to
it,
but
it
takes
up
like
very
etic
options.
So
plaid
is
like
if
we
need
to
add
things
that
you
might
want
to
override.
Like
I,
don't
know
you
can
fake
going
forwards
to
fit
fields
where
it's
obvious
that
we
may
need
to
extend
them.
A
They
take
like
the
option
pattern
where
we'll
just
add
more
options
with
defaults
and
so
like
the
command
line
may
start
passing
more
these
options,
but
we
won't
have
actually
broken
any
of
the
api's
and
we'll
just
make
the
defaults
on
obscene
behavior
so
like
the
logger
is
gonna
be
like
that
the
default
for
that,
if
you
import
it,
is
this
no
longer
there
just
doesn't
doesn't
blog
anything
ever
I
mean
if
you
want,
like
you,
can
opt
into
like
adding
our
blogger
or
or
wiring
up
your
a
blogger.
A
That
sort
of
thing
should
let
us
avoid
breaking
hopefully
and
I've,
been
putting
an
eye
towards
that
in
the
obvious
place
in
the
API
at
like
cluster
creation,
is
one
that's
pretty
obvious.
That's
going
to
wind
up
having
more
things,
plums
there
at
some
point
and
will
just
need
to
have
same
defaults,
so
hopefully
we're
most
delayed
towards
where
we
won't
find
more
places
where
we
need
to
do
that.
I'm
sure,
there's
still
something.
A
Yeah
well,
do
you
guys
have
any
other
error
questions
kind
of
thing
we
have
this
booked
and
told
you,
but
I
didn't,
have
a
ton
of
an
agenda
so
I'm.
B
I,
don't
have
anything
else,
just
got
that
other
docker
build
installed.
I'm
gonna
start
digging
into
that
this
afternoon.
So
cool.
A
A
B
Yeah
I
mean
that
was
that
was
why
I'd
had
pre
provisioned
vm's
before
because
getting
you
know
remote
desktop
or
SSH
on
conference
Wi-Fi
worked
quite
well,
but
it's
just
not
going
to
give
them
quite
the
same
experience,
because
if
I
give
them,
you
know
a
VM,
that's
different
from
what
their
laptop
is.
Well,
that's
not
going
to
really
be
what
we
intended
yeah.