►
From YouTube: sigs.k8s.io/kind 2019-12-02
A
B
After,
like
couple
days
and
I
I,
don't
know
I
still
fighting
some
misty
suggestion
like
a
week
later.
Sorry,
it's
not
the
worst,
but.
C
B
C
C
D
B
B
B
Though
I'm
not
inclined
to
merge
any
of
this
stuff
until
we
fix
the
tie
net,
the
image
missing
and
just
cut
a
tiny
release.
That
is
just
those
straightforward,
simple
fixes,
because
we
also
had
the
baby
we
had
since
the
last
release.
We
have
just
had
a
handful
of
things
that
fix
some
user
issues
like
the
v1
awful
for
config
had
slightly
more
restrictive,
like
defaulting
and
parsing.
B
D
B
D
B
B
C
I
started
down
that
path,
but
the
crux
of
the
problem
there
is
like,
although
I
can
get
going
and
some
of
that
stuff
working
and
it
just
handles
some
of
the
paths
we
still
have
for
the
build
image
and
I
don't
see
any
good
track.
They're
like
completely
blowing
up
those
build
scripts,
and
so
I
think
you
know,
based
on
what
we
showed
at
the
session
I.
Think
it's
reasonable
to
say.
You
know
if
you
want
to
use
kind
for
like
a
local
cluster
to
like
test
an
app
that
makes
sense.
C
C
B
B
The
other
thing
I
was
thinking
about
is.
If
we
modify
how
the
note
image
build
works
a
bit,
we
can
still
support,
there's,
probably
a
class
of
user
that
just
wants
particular
versions.
So
we
could
support
like
using
that
tarballs
or
something
but
I
think
they're,
even
there.
The
better
answer
is
to
just
like
figure
out
our
strategy
for
like
pre
publishing
images
of
all
the
things
so
that
you
don't
need
to
build
for
that
class
of
user
and
for
the
class.
That's
like
I
want
to
develop.
B
Kubernetes
I
think
we're
gonna
tell
you,
you
should
probably
use
Linux.
If
you
can
Mac
will
work,
but
given
the
odds
that
you're,
probably
on
a
MacBook,
it's
still
not
gonna,
be
as
good
of
an
experience.
If,
as
if
you
can
like
you
know,
get
a
beefier
Linux
box,
and
that
is
recommended.
If
you
can.
C
B
No,
no
so
so
so
kind
that
is
cross
kubernetes
version,
but
for
the
for
the
kind
node
image
you
we
that
needs
to
have
a
kubernetes
build
actually
building
it.
Isn't
that
hard?
The
question
is
stuff
like
how
are
we
gonna
handle,
knowing
if
it
works
with
your
kind
version
like?
How
do
we
schedule
these
because
we've
like?
B
We
can
probably
follow
the
pattern
that
the
project
has
been
doing
with
the
gcd
stuff,
where
prep
like
prowl
is
at
the
mix,
but
it's
on
the
trusted
cluster
and
all
its
doing
is
scheduling
a
cloud
build
somewhere
else
and
the
actual
like
docker
image
push
credentials
are
there
so
that
we
don't
have,
you
know,
are
sketchy
build
running
in
the
trusted
cluster,
but
you
the
credentials.
Are
there
transitively?
B
Because
we
don't
want,
like
you
know,
like
kindest
docker
hub
credentials
and
like
the
normal
brow
cluster
and
then
now
that
we
have
rerun,
is
a
thing
that
at
least
I
can
do
as
a
crowd
Minh.
If
we're
triggering
off
of
like
a
kubernetes
tag,
it'll
be
okay.
When
it
fails,
we
can
just
like
reschedule
it
by
like
I,
just
go
click
button
on
browse
page
and
that
wouldn't
be
of
the
worst.
But
there's
also
like
do
we
want
to
backfill?
And
none
of
that
solves
the
problem
that,
like
we're?
B
Communicating
that
is
hard,
I
think
we
could.
We
could
try
to
do
something
where,
like
we
list
images
and
then
we
do
our
own
discovery
process
to
like
figure
out.
Oh
this
image
has
this
metadata,
so
we
know
it
only
works
or
this
version,
but
that
is
a
little
bit
magical
and
confusing
I
think
the.
But
the
pattern
I
like
the
most
is
we
just
get
to
a
point
where
we're
not
breaking
things
and
then
we
just
push
images.
Maybe.
C
A
A
C
Think
that
makes
sense
and
I
mean
no
right
now:
I
I,
don't
see
kinds
charter,
as
necessarily
you
know,
being
one
for
one
replacement
for
mini
cube
or
something
that's
supported
and
like
docker
for
Windows
Enterprise
or
something
like
that
I
mean
if
users
want
to
be,
maybe
I
think
that's
an
eventual
goal,
but
I
mean
I'm.
That's
just
my
view
of
it
feel
free
to
disagree.
But
it's
like
I
feel
like
at
least
doing
the
first
thing.
C
I
want
to
do
for
the
first
use
case
to
these
help
peeps
dear
people
back
onto
a
path
to
succeed,
building
and
then
because
we're
we're
talking
about
a
different
use
case.
If
we're
trying
to
talk
and
build
up
kind
of
a
version,
compatibility
matrix-
and
you
know
that's
something
that
I'm
doing
in
my
other
project,
as
part
of
you
know
with
the
cap
with
holster
API
as
your
team-
and
you
know,
we've
got
this
whole
code
base
called
a
GIS
engine.
That
does
that
and
it's
a
very
non-trivial
task.
Yeah.
B
It's
yeah,
I,
agree
totally
and
I
think
that,
like
eventually
this,
like
as
we
as
we
hopefully
reach
a
point
where
the
stuff
that
seems
reasonable
to
test
with
kubernetes,
just
works
and
kind,
is
relatively
stable
and
make
sense
to
put
more
effort
into
like
you
know,
it
has
all
the
versions
available
in
it.
It
can
be
used
more
MIDI
cubish,
but
we
still
have
a
lot
of
things
to
do
with
like
the
ipv6
tests
are
not
reliable
enough.
D
B
D
B
C
B
B
New
bad
that
you
can't
like
I,
don't
think
you
can
like
a
retro,
actively
label
a
docker
container,
I'd
love
to
be
able
to
like
start
the
container
and
then
just
like,
add
a
label
to
it.
That
says
it
was
started
with
this
IP
and
then,
when
you
like
restart,
do
that,
which
would
be
a
little
bit
less
brittle.
I.
Think
then
like
depending
on
the
start,
order
was
what
we
could
just
do
like
it.
We
could
we
could
just
do
I,
don't
know
it
means
it
needs
more.
B
B
B
We
still
I
still
at
the
moment,
I'm
a
little
bit
more
interested
in
fixing
the
like
hack
on
kubernetes
flow
I.
B
Don't
think
we're
too
far
out
from
that
more
or
less
being
good
to
go
though,
but
there
are
a
couple
possibly
big
things:
I
want
to
look
at
external
cloud
provider
is
something
I
think
we
might
have
to
do
eventually
like
right
now,
we're
not
setting
a
cloud
provider
and
we're
letting
like
kind
of
some
fallback
built-ins
happen,
but
I
think
we
might
need
a
kind
cloud
provider
to
like
do
just
the
bare
minimum
there
like
go
through
and
set
the
NotI
peas
that
kind
of
stuff
I,
don't
think
we
need
it
immediately,
but
it's
more
of
a
like.
B
There
are
some
signs
that
it
might
be
something
we
have
to
do
long
term.
So
it's
worth
looking
into
to
stabilize.
We
also
I
think
I
want
I,
still
want
to
finish
and
submit
that
integrated
test
command
so
that
people
do
not
need
to
like
run
this
batch
script
or
figure
out
a
fiddle
with
that.
You
should
just
be
able
to
run
what
we're
running
in
presubmit.
B
B
B
No,
no,
no,
no
there's
a
cloud
provider
interface
and
so
not
not
no
you're
thinking
of
not
implementing
a
cloud.
Kubernetes
has
a
concept
of
a
cloud
provider
and
you
can
tell
the
controllers
Harry
I'm
running
this
cloud
provider,
and
that
is
responsible
for
things
like
specifying
that
the
nodes
have
these
IPS,
because
it's
expecting
that
you
have
its
your
integration
with,
like
the
actual
cloud
storage
has
its
own
separate
integration
credentials
have
their
own
separate
integration,
load,
balancer
and
ingress
of
their
own
separate
integration,
but
the
rest
of
Corcoran.
B
Any
use
has
its
has
an
your
face
for
this.
So
it's
a
little
weird
because
yeah
it
doesn't
cover
storage,
it
doesn't
cover
ingress,
it
doesn't
cover
low-down,
sir,
take
out
those
classes
of
things,
there's
still
there's
still
an
interface
for
dealing
with
cloud
and
that's
moving
towards.
You
run
a
controller
that
is
the
cloud
controller
manager
for
your
cloud
and
right
now
we're
basically
in
the
fallback
mode.
That's
like
we
don't
have
a
cloud
controller
and
that's
where
you
were
running
into
the
issue
with
like
cubelet
owning
the
node
IP
I.
D
B
You
know
not
quite
on
the
level
of
like
if
we
weren't
using
cube,
Adam
or
something
but
like
it
definitely
is
different.
Cluster
bring
up
different
yeah
it'll
be
another
thing
that
needs
orchestrating
and
we
might
need
to
do
that
if
we
were
running
a
like
quote-unquote
real
cluster,
we
like,
if
we
were
doing
a
really
good
job
providing
kubernetes
as
a
service
like
I
know.
B
C
It's
some
hockey
I'd,
wasn't
aware
of
in
the
Google
Doc.
Okay
I
just
wanted,
so
the
experiment
that
we
were
looking
at
is
now
we're
currently
at
a
spot
where
we
can
join
Windows
nodes
with,
can
join,
and
so
we've
had
two
different
people
trying
to
basically
make
a
smaller
make
a
faster
test
run
for
Windows,
because
today,
what
we
do
is
we
spin
up
a
couple
VMs
to
run?
You
know,
controller
manager,
scheduler
all
that
stuff
on
Linux
and
then
a
couple
VMs
that
are
running
Windows
that
join
as
nodes.
B
C
B
Sorry
so
like
when
we,
when
we
bootstrap
a
kind
node,
we
grate
a
config
just
for
that
node,
like
kind
of
rates
went
out
and
says,
like
you
know,
here's
all
the
options
I
want
you
to
use
with
queue.
Madam
and
I
think
that
includes
the
joint
info,
but
it
might
not
yeah.
If
not.
The
contract
map
is
a
well-known
one.
So
you
could
probably
just
like
edit
that
it's
used
by
queue
Menem
for
upgrade
as
well.
C
D
C
It's
I
mean
I,
think
it
I
think
it's
definitely
feasible.
I
was
gonna
figure
out
what
the
last
couple
workarounds
are,
but
if
that
works,
that
means
we
may
be
able.
This
is
something
that
can
provision
a
cluster
faster
than
waiting
on
clouding
it
on
the
other
nodes,
and
it
would
get
us
closer
to
being
able
to
make
a
Windows
job
PR
blocking,
because
today
our
provisioning
time
is
still
up
in
the
I.
B
B
A
C
B
B
C
C
B
C
B
C
And
so
the
the
thing
I
could
do
for
Windows
is
there's
two
paths
that
I've
done.
One
I
actually
have
your
on
Duffy's
repo
that
was
I've
just
got
a
vagrant
file
that
Ubuntu
VM
with
everything,
and
then
the
floor
works
fine
and
so
I
think
that
would
be
the
direction
that's
most
compatible
with
different
Windows
versions.
B
Yeah
I
don't
know
I
in
general,
have
less
strong
opinions
about
how
to
develop
kubernetes
well
from
Windows.
I
knew
that
to
be
enough
of
an
issue
in
the
past
that
I
just
didn't
try
but
I
do
think.
We
want
to
support
so
I'm,
pretty
willing
to
just
like
totally
defer
on
that.
Like
I,
don't
know
what
people
are
actually
using
to
do
stuff
these
days.
There's
a
lot
of
options.
B
I
have
personally
found
it
easiest
to
just
also
boot
Linux,
when
I
need
to
develop
Linux
but
I'm,
not
sure
how
viable
that
is
in
it
for
everyone
and
I,
don't
like
I,
don't
know
these
days
like
WSL
or
something
may
actually
be
more
convenient.
You
know.
C
C
My
machines
and
our
running
insiders
and
I
used
bsl-2,
but
the
speeds
comparable
between
them,
but
then
there's
no
weirdness
like
any
copy
and
paste
good.
Do
you
have
from
the
docs
just
one
SSH
to
your
VM
and
your
said:
yeah.
C
B
B
B
C
B
A
B
I
think
to
improve
that
we
would
have
to
significantly
change
upstream
I.
Think
that
may
actually
happen,
though,
with
projects
splitting
out
of
the
repo
they're
starting
to
realize
they'd
like
something
like
cube.
Adam
can
just
be
built
with
like
similar
to
how
we
build
with
the
makefile
I
was
putting
them
to
that.
It's
like
yeah,
you
can
just
use,
go
on
the
hoax,
but
you're.
B
Never
ibly
gonna
want
to
get
rid
of
that
because
you
have
problems
with
someone
has
the
wrong
version
installed:
yeah
the
binaries
aren't,
reproducible
and
stuff,
but
the
little
shell
thing
that
we
have
that
like
runs
the
build
in
a
go
container.
It's
pretty
simple!
It's
very
quick!
You
know
you're
not
linking
a
massive
binaries
like
if
we
moved
more.
If
the
project
in
that
direction,
I
think
we
could
get
it
to
not
be
so
painful
again:
yeah
yeah.
C
B
Yeah,
if
you
so
the
trick
there
with
kind,
we
don't
request,
building
everything
which
is
way
more
than
you
would
normally
build
with
a
make
release.
So
if
you
do
like
kinds
build
with
Basel
on
a
Linux
box,
that
is
the
optimum
experience
currently
I.
Think
with
some
caveats
that,
like,
for
example,
you
probably
want
the
Basel
in
your
path
to
be
basilisk
and
on
a
recent
kubernetes
branch.
We
should
point
that
at
like
the
right
version
that
should
just
work
as
opposed
to
like
I.
B
Don't
know
you
have
the
latest
installed
and
we're
using
some
super
old
one
and
see
I,
know
but
long
term
I,
don't
think
that's
going
to
work,
especially
as
we
fragment
the
repos
and
we
won't
need
it
as
bad
because
well,
you
won't
need
all
that.
Caching,
the
real
problem
is
gonna,
be
how
do
we
test
this
in
CI?
I
still
need
to
catch
up
on.
That's
a
that's
a
big
one.
We
we
should
be
looking
into
cube.
B
B
D
B
B
I
think
we
can
drive
it
in
the
other
direction,
but
I
think
we
want
to
see
if
we
can
hack
that
on
to
the--
like
move
out
of
tree
stuff,
because
that's
gonna
shake
up
the
building.
You
know
I
expect
that
will
actually
in
the
Basel
stuff
I.
Don't
think.
There's
gonna
be
enough
effort
to
maintain
that
when
it's
like
10
repos
but
we'll
see.
B
Don't
know
I
I
think
we
have
got
to
make
the
actual
build
cheaper
as
much
as
kind
can
like
just
avoid
building
things
that
doesn't
need
building
the
things
that
does
need
are
is
expensive,
they're,
also
a
lot
bigger
than
I,
like
one
thing
that
we
get
a
lot
that
I've
not
been
trying
to
take
head
on
is
people
saying,
like
oh
just
use
k3s
instead
I
had
a
friend
reach
out
the
other
day,
because
the
heads
like
discovered
how
far
kinda
come-along
seeing
it
on
like
hacker
news
and,
of
course,
like
every
comment,
had
been
replied
to
by
someone
from
Rancher.
B
B
B
And
some
of
them
aren't
clear
yet
how
we
can
remove
some
of
these
things
like
it
doesn't
remove
things
like
credential
provider,
the
Eataly
binaries.
This
may
be
more
interesting
to
your
actual
work
in.
They
are
going
to
be
really
really
interesting
for
everyone
long-term.
When
we
realize
that
we
want
to
have
you
do
a
dot
test
and
everyone
use
it,
but
like
we
don't
want
all
the
cloud
providers
and
the
repo
and
stuff
it
heavily
depends
on
them
mm-hmm
and
is
generally
a
mess.
How
are
people
supposed
to
build
that?
B
B
A
B
Don't
think
anyone
has
a
plan
for
that
yet
mm-hmm
and
that's
going
to
be
a
mess
for
all
of
the
clouds,
but
we're
like
oh
I,
want
XYZ
test
I,
don't
know
what
the
I
don't
know
what
the
approach
is
gonna
be
there
yet.
But
that
is
that
that
is
another
place
where,
like
we
can't
actually
make
it
smaller
and
get
rid
of
all
the
cloud
providers.
B
C
C
A
D
B
Using
kind
and
having
you
like,
REM,
kubernetes
and
stuff
that
way
and
the
workshop
is
oriented
towards
contributing
kubernetes
I,
don't
think
they
had
as
much
around
like
you
know,
run
it
and
configure
it
and
whatever
by
nominally
it's
a
kubernetes
contributor
workshop,
they
he's
leveraging
it
I
found
out
about
that
at
San,
Diego
and
I've
been
planning
to
try
to
be
on
the
contributor
summit
staff,
helping
anyhow
so
I'm
hoping
for
Amsterdam
to
actually
be
on
that
and
have
a
better
idea.
B
What
they're
doing
and
I
think
we
can
I
think
we
have
a
lot
of
room
to
collaborate
there
and
that
might
be
a
better
spot
in
the
future.
It's
you
know
not
a
speakers
at
the
main
conference,
but
I
think
all
of
the
people
involved
are
already
going
to
be
speaking
as
like
sig
representatives
or
something
anyhow.
So
if
we
can
do
it
during
the
new
contributor
workshop,
that
may
be
more
appropriate
in
the
future
anyhow
well
see.
C
C
C
B
And
it
doesn't
it
not
only
thing
I
like
about
it
is
that
it
I
was
pretty
happy
doing
a
workshop
could
take
anything
but
I
do
want
to
let
you
know.
Other
people
speak
and
stuff.
I
have
not
spoken
during
the
main
I've
only
spoken
during
the
main
thing
once
besides
the
workshop,
but
I'm
pretty
happy
with
that
I'm
planning
to
just
see
if
we
can
get
another
kind,
deep
dive
and
try
to
pull
in
someone
else
on
that.
So
we
can
keep
rotating
in
people
and
do
the
do
the
contributor
summit
stuff.
C
Yeah
I
definitely
encourage
others
to
do
it
and
then
also
maintain
your
track
ones.
They
do
allow
two
speakers,
and
so,
if
there's
someone
that
that's
contributing,
especially
as
they've
got,
you
know
a
new
thing.
They
want
to
talk
about.
That's
another
way
that
you
can
that
they
can
get
an
opportunity
to
do
that.
Yeah.
B
I'm
gonna
wait
and
see
closer
I
think
what
we've
done
before
is
you
can
just
put
a
placeholder
and
then
switch,
but
tentatively,
my
hope
for
this
time.
I'm
I've
gotten
permission
to
steal
a
bunch
more
of
a
colleagues
time
here,
who's
been
doing
a
lot
of
peer
review,
but
hasn't
said
much
code,
I'm,
hoping
by
the
time
we're
at
Q.
Khan
bill
has
been
like
writing
code
and
then
maybe
they'll
want
to
speak
and
to
rotate
in
someone
new.
No,
because.
B
Else
that's
been
involved
that
is
likely
to
go
to
Amsterdam,
has
spoken
on
this
yeah,
but
we'll
see
definitely
pulling
someone
there.
Aaron
did
the
same
thing
with
me:
I
didn't
submit
the
deep
dive.
I
was
going
to
just
do
the
workshop
and
like
mentoring
sessions,
and
it's
gonna
be
great,
and
he
just
like
just
signed
you
up
beside.