►
From YouTube: sigs.k8s.io/kind 2019-06-17
A
B
B
Yeah
thing
to
find
it
I,
don't
know
where'd
it
go
well,
so
anyhow,
basically
H
a
proxy.
All
we
need
to
do
is
what
we
did
before.
We
just
need
to
build
an
image
that
has
a
that
has
a
dummy
config
in
it,
so
don't
crash,
and
then
you
can
send
signal
and
they're.
The
official
docs
on
the
image
basically
say
to
do
this,
though
they're
expecting
that
you
mount
your
you'll,
make
an
image
with
your
actual
config.
So
I
tell
you
do
like
from
their
image
copy
your
config
in
and
then
run
that
necklace.
B
A
B
B
You
need
to
take
this
image
and
make
a
custom
image
where
you
add
a
dummy
config
enough,
that
it
won't
die,
and
then
you
can
right
over
that
config
and
send
a
signal
and
it
will
reload
the
config
yeah.
We
we
already
implemented
that
we
just
need
to
go
back
and
like
bring
that
back
and
we
need
to
add
the
we
need
to
add
healthy
checks.
A
A
B
So
I'm
not
sure
about
that,
but
I
know
that
there
are
other
projects
that
are
that
switch
to
this
to
solve
this
problem.
So
we
should
be
able
to
check
that
we
can
look
at
what
what
keeps
Bri
is
doing
in
theory,
that
I
spoke
to
them
and
they
that's
why
they
have
engine
xm8
NH
a
proxy.
They
added
h
a
proxy
to
solve
this
problem.
Now,
though,.
A
B
I
spent
some
time
when
we
can,
we
should
not
send
you
an
extra
this
and
I
believe
it
should
be
pop
you're,
actually
working
and
the
issue
of
it
will
exit
without
the
config
is
trivially
solvable
by.
We
just
publish
our
own
image
that
that's
one.
What
we,
if
you,
if
you
dig
back
to
it,
appears
we
actually
briefly
defects
when
we
were
switching,
we
can
basically
just
bring
back
that
implementation.
A
B
Yes,
one
around
I
think
it's
what
everyone's
using
that's
actually
doing
this
successfully
there
might
some
people
using
on
Blake,
but
I
took
a
look
at
that
and,
if
you're
not
actually
programming
envoy
dynamically,
it
doesn't
look
that
doesn't
look
necessarily
better
for
this
problem
and
it
looks
much
more.
B
So
ipv6
right,
sorry
I
need
to
follow
up
on
your
if
you're
new
PR
again
with
the
the
config
thing,
so
we're
also
gonna
need
to
account
for
one
further
thing
which
is
actually
taking
in
each
those
inputs
because
there's
some
other
stuff
that
we're
gonna
want
per
node.
That
needs
to
be
config
anyhow.
So
this
is
just
a
good
change
to
figure
out
and
make
and
get
config
working.
B
We're
gonna
want
join
joint
configuration
for
each
node
soon,
so
that
we
can
actually
specify
some
stuff
I
didn't
get
this
on
the
agenda,
so
I
will
put
until
after
Patrick's
demo.
If
we
have
time.
B
Yeah
so
I
just
I,
don't
think
we
need
to
talk
too
much
about
these
here.
I
will
go
back
and
review
those
again
later
today.
I
think
we've
generally
got
the
right
approach
there.
Once
we
get
rid
of
the
needing
to
override
the
drop
in
the
ipv6.
Pr
is
mostly
I
think
what
we
want
and
then
we
can
for
there.
B
C
Yeah
yeah
sure,
let
me
get
this
shared
out,
so
basically,
I
was
just
doing
a
quick
experiment
over
the
weekend,
since
there
was
that
new
windows
subsystem
for
Linux
that
was
just
released,
yeah
so
hold
on
see
if
I
do
is
that
coming
through?
Yes,
okay,
cool
all
right
and
now
you
should
see
something
completely
different.
Yes,.
B
A
C
C
C
If
we
look
at
it,
I
didn't
send
a
particular
memory
limit
on
this,
but
it
will.
Let
me
basically
use
all
up
tall
the
memory,
that's
on
the
system
so
rather
than
like
docker
for
Windows,
where
I
had
to
go
in
there
and
pick
how
much
memory
one
to
reserve
for
it.
This
basically
I,
don't
know
how
exactly
how
it
works
behind
the
scenes,
but
it
seems
to
just
be
able
to
page
stuff
in
and
out
as
needed.
That's
really
cool.
C
So
if
I
go
over
here
and
just
run
some
reason,
the
same
Linux
version
of
kind
I
put
it
in
these
are
been
local
because
I'm
kind
of
old-school
that
way
but
and
then
the
version
of
docker
that's
install,
is
just
whatever
they
get.
Docker
dot.
Io
script
installed
so
I
think
it's
I,
don't
know
16
or
sorry
1906
EE
or
something
like
that.
C
So
it
looks
like
that
stuff
was
already
we're
waiting
for
it
to
crash
to
work
great,
and
so
it's
fun
here
as
well.
This
is
going
if
I
bring
up
task
manager
here,
like
I
just
show
you
I
can
tell
that's
operating
differently
under
the
hood
here.
If
I
go
down
here,
I'm
looking
for
there's
this
process
here
called
vmm
yeah,
and
so
this
is
actually
like.
C
I,
don't
know
how
to
describe
it.
I
think
it's
like
a
fake
process,
but
this
is
basically
showing
how
much
memory
is
being
taken
away
from
the
Windows
operating
system
and
given
over
to
this
yeah.
This,
like
lightweight,
Linux,
VM,
that's
running
their
brain
and
we
can
also
see
that
it's
eating
what
a
core
yeah
yeah,
basically,
which
I
mean
I,
don't
know
if
that's
typical,
for
what
kind
is
doing
right
now.
A
A
C
Could
just
show
you
right
here
if
I
go
over
to
the
performance
tab,
I
can
see
that
I've
got
a
separate
v-neck
with
its
own
MAC
address
for
the
WSL
instance.
But
right
now
it's
behind
a
mat
and
so
I
think
that,
depending
on
the
networking
scenario,
I'm
going
to
check
and
see
if
there's
a
way,
I
can
get
it
to
run
without
in
that,
because
I
may
need
to
do
that
if
I
want
to
join
another
physical
node,
because
one
of
the
things
I
wanted
to
test
out
was
doing
a
cube.
B
C
B
Pretty
related
to
this
Duffy
Coulee
reached
out
a
while
back
about
the
idea
of
doing
a
tutorial
session
in
San
Diego
for
like
developing
kubernetes,
locally
I've,
been
debating.
Doing
that
and
like
this
would
be
a
very
nice
way
to
bootstrap
potentially
for
anyone
using
you
know,
Windows
laptop
I
will
have
to
poke
the
both
of
you
if
I
continue
further
on
that
proposal,.
C
Interesting,
so
I
could
like
copy
and
paste
my
cube,
config
from
the
WSL,
to
instance,
into
my
Windows
host
if
I
wanted
to
okay,
but
anyway,
once
I
get
this
work
and
I'll
get
some
steps
there,
so
that
the
main
dock
site
that
you
have
here,
the
that's
under
the
docks
or
sorry
kind,
cigs,
okay,
instead
IO,
is
that
just
published
as
like
github
pages
off
your
repo.
It's
never.
B
Fine
but
yeah
it
comes
off
of
the
repo,
so
there's
just
there's
a
site
content
er
that
has
the
actual
content.
Okay,.
B
C
B
Was
like
an
overall
like
user
guide
heading
that
has
like
a
couple
of
pages
in
a
now
and
I'm
playing
to
have.
You
know
this
would
probably
be
a
great
page
for
that.
Okay,
it
should
be
really
easy.
Potentially,
if
you
want
the
absolute
easiest
way
to
iterate
on.
This
is
just
if
you
send
pull
requests
that
will
five
will
render
it
and
they'll
be
like
a
link
in
like
one
of
those
contacts,
and
you
can
just
go
there.
A
B
C
B
B
That
looks
very
promising
if
anything
I'd
be
worried
that
well,
we
won't
need
containers
anymore
if
we
can
just
make
VMs
magically
clear
their
memory.
B
So
I
don't
know
if
all
gets
written
up
on
I
can
type
more.
This
later
so
I've
been
thinking
about
where
you
need
to
run
a
service
in
the
cluster
and
serve
traffic
to
it,
and
what
we
can
do
to
make
that
work
well
and
there's
like
a
bunch
of
options
for
this,
but
I
think
actually
the
most
straightforward
and
stable
thing
to
support
is
just
to
let
users
have
more
port
for
words.
That's
gonna
work
everywhere.
B
It
doesn't
have
any
extra
overhead
it's
on
docker,
to
support
that
which
is
great,
but
the
trick
for
this
is
gonna,
be
in
the
multi
node
world.
In
particular,
you
need
to
know
you
know
how
do
I
make
my
workload
run
on
a
node
that
actually
has
the
host
port
being
forwarded
to
the
to
the
actual
host
right.
So
if
we,
if
we
let
you
specify
like
for
this
extra
port
from
here
to
here
on
a
node
and
the
config,
ideally
you
don't
have
to
like
say
predict
the
node
name,
Donna
Seba.
B
How
I
actually
do
this
real
bare
metal
clusters?
And
the
answer
is
you?
Would
you
would
label
that
note
and
maybe
even
taint
it
to
say
you
know
this
node
is
I'm,
actually
sending
ingress
traffic
from
the
from
the
rest
of
the
world
and
I
want
to
schedule.
You
know
an
ingress
engine
X
set
on
this
labeled.
B
No,
that
actually
has
the
appropriate
port
forwarded
to
it
like
if
I,
if,
instead
of
using
docker
port
for
it
I'm
doing
and
doing
this
at
home
and
I'm
like
matting
with
my
browser
right,
I'm,
gonna,
say:
ok,
the
routers
pointed
at
this
machine
and
it
has
port
forward,
put
a
label
on
it,
and
then
you
can
schedule
the
ingress
controller
to
actually
run
on
that
node.
So
what
you
need,
then,
is
you
just
need
to
be
able
to
match
up
labels
to
nodes
in
the
config,
so
simple
on
the
node
objects
in
the
config.
B
We
let
you
specify
labels
and
taints.
We
propagate
that
through
to
keep
items,
join
configuration
down
to
the
cubelet
args,
and
then
we
allow
you
to
specify
additional
port
forwards.
On
top
of
that,
you
can
deploy
things
like
the
socks,
proxy
or
whatever,
but
for
the
simplest
case
of
just
like
I'm,
a
user
that
wants
to
run
a
web
server
and
a
one
node
cluster.
You
just
need
to
specify
the
web
server
port
forward
and
run
with
like
host
network
and
on
the
on
the
web
server
pod
and
that's
it.
C
B
Node
name
and
guess
the
node
name,
or
something
like
that
and
like
in
general,
your
problem.
This
could
be
useful
for
testing.
I
spoke
to
various
potential
internal
users,
and
one
of
the
things
that
came
up
is
when
they're
running
multi
node,
and
they
want
to
do
multi
node
testing,
even
if
they're
not
on
something
like
gke.
They
kind
of
make
node
pools
by
like
labeling
like
these
nodes
are
first
a
full
workloads
and
these
loads
nodes
or
not.
B
So
if
they
want
to
run
the
same,
manifests
and
keep
those
on
different
nodes
to
get
more
similar
behavior
to
their
production
deployment,
then
they're
gonna
want
labeling
them.
You
could
accomplish
this
right
now
by
doing
this
by
hand
in
your
own
script
after
it
boots,
but
I
think
we
should
just
add
it
to
the
config,
because
then,
whenever
we
have
some
advanced
stuff
like
the
mounts,
the
extra
mounts
or
the
ports,
you
can.
You
know,
label
a
node
to
say.
C
B
We
can
just
add
support
for
patching
a
per
node
config
and
you
can
jump
through
the
hoops
of
configuring,
a
queue
Batum
config
patch
that
accomplishes
this,
but
the
UX
for
that's
going
to
be
pretty
bad,
and
this
is
a
common
feature
and
for
common
enough
settings
that
we
think
kind
should
be
aware
of
you've
been
doing
top-level
fields,
so
I
think
we
should
probably
do
a
top-level
field
for
this
and
I
think
that
actually
solves
like
most
people's
problems.
B
A
C
B
B
B
A
A
B
B
C
B
Whole
thing
to
do,
and
then
so
then
it's
the
most
flexible
and
what
my
point
is
that
we
can
reuse
this
labeling
taint
stuff
to
test
some
other
things
where
you
want
to
have
like
sets
of
notes
for
different
things,
I'm
going
to
firm
what
the?
How
did
you
do?
The
full
run
like
this
is
a
at
cluster
creation
time
you
specify
on
the
now
or
for
NCR.
A
B
A
concept
of
this,
so
we
can
kind
of
lift
that
concept.
It
doesn't
move
you
think,
I'm
gonna,
say:
okay,
you
can
specify
a
port
for
it
on
a
note
and
then
we
will
simultaneously
give
support
for
you
can
like
label
and
paint.
The
note
so
you
know
like
this
is
my
port
for
denote.
I
want
something.
You'll
add
these
to
the
APA
right
to
the
kind
APA
and
you,
and
we
can
actually
say
we
could
do
this
without
first-class
support.
B
For
most
of
it,
we
need
first-class
support
for
the
extra
port
for
it
and
I
think
we
might
as
well
do
it
for
labeling
taint
just
because
those
are
pretty
awkward.
The
way
that
you
would
do
that
and
how
is
there
will
Flags
only
not
config
so
you'd
need
to
like
patch
cubelet
extra
args
and
the
key
Batum
config,
and
that's
an
ideal
and
I
know
it
doesn't
look
like
they're
moving
to
a
proper
configure
object
at
any
point,
so
we
can
just
add
a
field
and
plumb
that
through
and
it.
B
Node
right
so
like
whichever
like,
and
you
could
have
a
bunch
of
nodes
with
different
port
for
its
to-
and
you
could
say,
like
I'll-
run
this
in
grips
with
this
port
in
this
ingress,
with
this
port
in
this
ingress
of
this
port
or
whatever,
but
they
need
to
match
up,
and
this
solves
I'm,
not
sure
I'm,
not
sure
about
the
wsl
to
stuff
how
that's
gonna,
work,
yeah
free
wsl
to
Walker
for
Windows
has
the
same
problem.
Where,
like
it's
bouncing
traffic
and
like
you,
you
know
have
like
an
IP
to
do.
B
B
B
The
problem
was
always
like:
what's
the
UX
for
that,
once
you
go
past
one
node
and
I
I,
don't
know
why
we
went
against
labeling
and
tainting
before,
but
I
looked
around
in,
like
most
mature
deployers
support
that
it's
a
very
useful
thing
to
be
able
to
do
to
say,
like
this
node
has
some
label
on
it
for
some
dedicated
scheduling
and.
B
A
A
B
If
we're
actually
going
through
the
process
of
taking
those
inputs
and
mapping
it
through
that
one
internal
command
that
takes
based
on
the
Kurban,
a
DS
version
the
template
and
generates
the
equivalent
configuration,
then
we
should
be
able
to
have
an
easier
time
doing.
Something
like
this
note,
is
using
this
image
with
this
turbinates
version
of
this
node
is
using
this
image.
With
this
community's
version.
I
think.
B
They're
using
totally
different
logic
for
the
actions
and
stuff
because
that's
been
too
unstable
internally,
it
kind
o
but
yeah.
So
this
will
should
all
come
together
pretty
nicely
and
I.
Think
all
of
the
other
ways
that
you
can
you
know
approach
will
can
kind
of
be
built
against
this
I
give
it.
If
you
wanted
to
build
some
kind
of
thing
for
dynamically,
affording
arbitrary
things
you
could
still
tunnel
it
over
having
a
port
open
up.
B
B
C
B
Thing
I
would
like
to
stress,
but
I
think
we
need
to
talk
to
more
people
is
where
I
think
so
kind
is
coming
up
on
about
a
year
and
config
mostly
seems
pretty
stable,
we're
just
adding
things.
I
am,
I
think,
planning
to
propose
moving
to
beta
soon,
I,
don't
think
we
actually
I
think
we
can
meet
the
requirements
for
data
to
like
not
freak
people.
The
logging
changes
one
of
the
last
ones.
B
That
technically
is
going
to
violate
that,
because
the
flag
change
so
I
mean
if
I
want
to
slip
that
one
in
and
then
once
we've
slipped
that
one
and
I
can't
think
of
anything
else.
We're
like
we
definitely
want
to
tear
out
and
replace
you
think,
that's
at
the
command
line
or
the
config
level.
Instead,
we're
gonna
be
adding
stuff
like
the
port
for
or
changing
internals
that
are
not
specified.
Like
you
know,
what's
in
the
node
images
unspecified
other
than
it
boots,
communities.
A
Yeah,
but
before
going
to
bed,
I
suggest
that
you
select
some
internal
functions
or
methods
on
these
things
to
add
unit
tests,
because
it's
growing
too
fast
the
project.
What
would
select
what
forbid
to
choose?
Something's
the
need
unit
tests,
you
know
troubled
arrested,
I
mean
very
vessels
adopted
some
punches
that
we
should
do.
B
But
it
shouldn't
be
a
blocker,
because
so
the
other
thing
is
with
German
is
obviously
all
this
is
taken
in
good
faith,
that
you
don't
intentionally
break
things
once
you
go
to
beta,
but
that
doesn't
prevent
bugs
from
happening
and
if
they
do,
then
you
revert
and
patch
back
but,
like
you
know,
I
mean
criminals.
Has
this
happen
occasionally,
where,
like
there's,
some
breaking
change
to
the
API
services,
not
quite
intentional
to
him
there
there's
there's
somewhere
around
that,
like
security
bugs
or
things
the.
B
Especially
because
it
wasn't
something
that
was
actually
specified,
as
you
were
supposed
to
be
able
to
do
this,
it
just
happened
that
you
were
I
kind
should
be
in
the
same
boat.
We
should
not
break,
we
should
not
be
it.
We
should
not
need
to
intentionally
break
things
like
the
config
format.
At
this
point,
I,
don't
believe
also
beta
does
allow
you
to
eventually
change
it.
B
You
just
need
to
keep
her
out
for
a
while,
so
I
think
we
should
be
good
to
like
there's
nothing,
there's
nothing
that
jumps
out
about
the
API
service
right
now.
The
biggest
one
I
can
think
of
is
actually
that
that
logging
flag
is
very
loggers
specific
and,
as
far
as
I
can
tell
to
get
into
the
ecosystem.
K
log
is
just
perfect
suck
it
up
and
use
K
log
everything,
kids
using
K
log,
and
we
are
making
it
symbol
for
library
consumption
to
like
not
to
use
K
log.
B
I'm
hoping
we
still,
we
still
have
time
I
think
too,
when
these
bring
up
to
how
we
do
convicted
and
then
probably
hopefully,
stove
ipv6
doesn't
seem
particularly
at
risk.
I
think
most
of
the
other
features
like
no
restart
I'm,
still
kind
of
hesitant
about
how
we've
implemented
that
and
I
want
more
time
to
review.
A
Okay,
well,
any
the
people
would
experience
in
the
templates
with
you
might
mean
to
review
the
the
pull
request
with
the
countries.
They
are
working
in
the
pull
request,
but
I
don't
know.
I
took
some
approach,
super
set
for
everything
as
a
control,
plane
and
party
good
theories.
I
use
it
for
all
viruses,
but
you
do
have
only
one.
No,
this
is
they
were
not.
Is
the
control
planning
point
right.
B
A
I
know
senator
s,
but
the
thing
is
that
I
had
not
I,
don't
have
experience
with
it
in
place
and
I
need
to
check
that
everybody
check
and
with
a
BBC
is
I
water.
Probably
when
again-
and
yes,
the
cold
now
is
only
check
if
it's
a
BBC's,
you
change
the
IP
address
that
is
passes.
Us
parameter
of
fancy
regret
in
theory,
all
day
of
the
methods
are
at
PVC,
so
maybe
before
they
don't
branch,
they
don't
do
different
things
depending
on
the
PVC
sorting.