►
From YouTube: sigs.k8s.io/kind 2019-03-11
A
A
A
Okay,
oh
hey:
what
ended
this
week?
Hello!
Everyone
welcome
to
the
kind
sub-project
meeting.
This
is
Monday
March,
eleventh
nineteen
or
just
quick
reminder
that
this
meeting
follows
the
CN
CF
code
of
conduct,
which
basically
filters
down
to
toffee
a
jerk,
and
this
meeting
will
be
recorded.
Posted
publicly,
so
looks
like
first
up,
we
have
Antonio's
discussion
of
c9.
B
B
Well,
I
just
wanted
to
know
if
please
pick
one
week
or
one
month
or
when
they
it'll
be
today
but
I,
don't
know
the
name,
the
one
one
one
person
suggested
to
use
the
to
mount
the
folders.
So
with
that
we
can
use
difference
here,
nice,
so
that
problem
is
solved
and
if
you
have
been
working,
feel
we
can
parse
the
IP
address
and
the
data
for
is
ipv6
or
not
so
I
think
that
that,
with
that
I
can
submit
a
pull
request
to
have
a
PVC
support.
Okay,.
B
Am
doing
I,
try
it
quickly
and
and
I
wasn't
able
to
to
make
it
work
at
the
first
time.
So
I
used
the
time
to
study
a
bit
more
CNI.
It
was
alone
scene
I
was
in
contact
with
kubernetes
anyone,
and
then
this
time
once
I
have
this
working
I
will
try
to
set
up
calico,
but
I
want
first
to
have
something
working
with
a
BBC
and
then
start
to
able
to
see
a
nice,
because
if
I
do
the
opposite,
I
start
to
test
to
see
a
nice
and
I
kind
of
married.
A
B
Suggest
to
go
to
the
standards,
but
first
I
want
to
have
some
in
this
table.
Testing
IP
resist
to
know.
You
know
if
very
well.
If
the
problem
is
in
Coronet,
if
the
problems
in
the
cni,
the
problem
is
in
kind.
This
way
we
can,
we
can
identify.
What
is
the
problem?
I
have
I
set
up
and
invited
me
inside
KCI,
so
I
I
would
like
to
know
once
I
have
something
working,
I
will
try
calico
and
I
wear
red.
You
know,
okay,
great
thank.
A
You
so
Timothy
has,
when
is
the
next
planning
cycle,
/
115
plans,
so
I
think
we've
gotten
a
bit
out
of
sync
with
kubernetes
and
that's
probably,
okay,
we're
trending
towards
like
monthly
ish
releases,
I'm,
hoping
that
we'll
be
ready
this
week
mostly
wanted
to
land
the
new
configuration
version.
So
we
can
go
ahead
and
roll
that
out.
That
one
makes
it
a
bit
easier
for
us
to
start
extending
some
other
things
that
should
happen
today
and
we
figured
later
this
week.
So
the
next
release
will
be
next
month
after
after
this
week.
A
C
D
D
C
A
Yeah
I
think
we
that
would
have
been
nice,
but
we'll
probably
need
to
you,
know
slow
up
and
push
that
back
and
let
alpha
and
beta
take
their
natural
course.
It
was
a
good
forcing
function
for
trying
to
think
about
how
to
get
some
of
these
things.
Disabled
but
I.
Think
it's
a
bit
optimistic
to
say:
oh,
it's
1.0
now
that
quickly,
so
we
will
also
just
need
to
have
a
discussion
about
when
that
should
be
yeah.
C
A
A
C
That
that
will
give
a
nice
exit
strategy
to
to
the
upgrade
scenario
where
you
know,
if
you
don't
want
to
plumb
through
upgrade,
which
is
totally
fine.
If
we
have
individual
phase
support,
then
they
could
stop
and
do
the
customization
or
upgrades,
and
we
can
even
build
that
into
CI
via
script
or
whatever.
If
we
want
to
do
right.
A
C
A
A
There
is
like
a
config
file
for
all
of
the
nodes,
Cubana
voice,
and
that's
going
to
make
it
a
little
difficult
to
properly
do
like
multi
version
clusters
and
I
think
during
one
of
the
phases
cleanup
there's
a
way
we
can
fix
that
and
that
that's
also
something
that
we
should
really
land
so
that
you
can
have
right
now,
technically
as
possible
to
set
them
to
different
images
with
different
versions,
but
there's
not
really
any
support
beyond
it.
We'll
use
that
image
it.
It's
not
smart
about
it.
C
A
Yeah
so
I
started
to
do
this
also
and
then
realize
that
we
should
everyone,
I
made
a
milestone
for
0.3
and
I
adjusted
the
date
for
point
to
that.
We
didn't
get
to
to
be
by
the
end
of
this
week
and
then
I
just
said
point
free
for
a
month
out,
because
that's
been
about
the
cadence
that
we've
actually
gotten
to
it's
about
a
month.
But
we
should.
We
should
really
get
everyone
in
like
discussed.
A
This
should
be,
it
should
be
good
right
away,
I
mean
so
we're
always
running
against
head
for
the
CIA
by
the
ideal
model
I
want
to
get
to
is
where
you
should
be
able
to
just
pull
the
latest
binary
always,
and
we
don't
know,
we
don't
break
you.
We've
we've
prioritized
changing
things
that
we
think
should
be
changed.
Breaking
wise
like,
for
example,
you
used
to
prefix
the
cluster
name,
and
we
drop
that
pretty
early
I
I.
The
other
model
is
that
we've
been
pretty
careful
about.
A
If
you
have
an
image
that
you
want
and
you're
using,
you
should
be
able
to
switch
between
versions
of
the
binary
and
it
should
work.
Even
if
we've
like
enhanced
things,
it
will
gracefully
degrade
to
the
old
behavior
if
something
isn't
present
in
the
image
so
like
we
now
pre
include
CNI,
so
it
doesn't
have
to
pull
that
if
it
can't
find
one,
it
will
fall
back
to
the
old
behavior
of
just
installing
we've
from
upstream.
C
A
A
E
A
Probably
right
that
makes
sense,
I
think
one
of
the
biggest
ones
that
isn't
actually
explicitly
listed
anywhere
here
is
like
having
the
config
to
at
least
fade.
That's
great.
So
we're
saying
like
we
are
actually
not
going
to
break
that
since
we're
still
calling
that
alpha
yeah,
really
it's
kind
of,
doesn't
make
a
lot
of
sense
to
have
1.0
in
the
primary
way
of
turning
any
knobs
is
marked
alpha.
E
A
A
A
lot
of
these
look
like
the
things
that
we
have,
but
in
the
mouth
already
look
like
things
that
are
gonna
fall
in
pretty
pretty
quickly
pretty
easily,
or
some
of
them
are
possibly
already
done
and
just
need
updating
like
the
windows
usage
improvements,
I,
imagine
it
it's
not
actually
going
to
need
to
be
that
far
out.
We
just
need
to
get
to
where
we're
comfortable
with,
like
the
the
knobs
configuration
I
mean.
E
C
D
C
You
made
some
milestone
that
says,
you
know
like
default,
CI
integration
tool
for
KK,
for
you
know
other
CRD
integration
or
migration
tools,
maybe
Travis
CI
or
other
standard
CI
tools
that
that's
the
default
way
of
for
them
to
actually
through
their
testing.
That
might
be
a
good
way
to
make
sure
that
you
have
all
the
radical
signals
that,
through
deploy,
50,
GA,
yeah.
A
For
me,
other
than
closing
out
some
of
these
things
like
phases,
the
biggest
thing
I
want
to
be
looking
towards,
but
I
don't
think,
is
a
1.0
blocker
he's
better
support
for
serving
traffic
from
a
cluster
locally,
so
that
you
can
like
test
that
you
have
some
application
and
that
you
can
have
some
integration
test
run
and
hit
all
the
endpoints.
A
lot
of
people
have
these
test,
setups
that
aren't
nice
like
Samui
and
they
don't
run
in
the
clusters.
A
So
they
want
to
like
talk
through
it's
not
like
an
ingress
or
load
balancer
or
and
I
some.
Something
like
that
right
now,
it's
difficult
with
kind
to
connect
up
things
on
the
host
to
without
using
something
like
you
could
have
proxy,
which
hasn't
been
great
so
far,
so
got
a
couple
of
ideas
with
that
James
did
you
wind
up
playing
with
the
IP
route
at
there
Liam
adding
the
address
on
the
host
at.
E
A
E
E
Like
for
visibility
for
people,
so
you
can
kind
of
see
the
overarching
thing
we
actually
go
beyond
that
and
create
a
project
per
milestone
to
and
everything
in,
but
I
think
that's
a
separate
thing.
We've
also
picked
up
on
seed
cluster
life
cycles,
next
milestone,
which
I
found
quite
useful
to
sort
of
be
like
right.
You
know,
I
know
that
we
do
need
to
do
this
pretty
soon
it
kind
of
overlaps
with
the
priority
leg
was
a
bit,
but
that
might
just
be
the
way
we're
using
it.
A
C
I
think
the
the
reason
we
do
next
is
because
we
used
to
pretend
and
actually
have
many
milestones
good.
We
used
to
lie
to
ourselves
every
cycle
about
like
trying
to
get
to
something
into
certain
milestones,
so
instead
we
just
put
it
into
next
every
single
time
when
we
go
through
next
every
time
to
actually
see
like
what
can
we
really
accomplish
this
next
cycle,
because
when
we
always
found
that
we
did
this,
we
could
we
would
we
would
overestimate
we
could
possibly
accomplish
I've.
A
Learned
some
from
testing
fro
where
we
have
these
like
20
18
goals,
2019
colds
and
then
I
haven't
bred
yet
because
we
haven't
been
that
far
out,
but
test
infer
also
has
a
some
day
milestone.
That's
like
yeah!
This
isn't
slated
at
all
like
we're
going
to
track
the
business
and
I,
but
it's
no
one's
taking
this
on.
This
isn't
happening.
Yeah.
C
A
Right
like
we're,
we're
we're
pony
would
be
great,
but
it's
not
happening
I,
don't
think.
We've
had
any
of
those
open
yet
they've
mostly
fallen
into
things
where
it's
like
yeah.
It
would
be
good
to
do
it
this
year,
ideally,
but
I
could
see
that
happening
eventually
like
what
one
that
would
probably
make
sense,
for
that
is
the
homebrew
formula.
That
seems
like
a
thing.
That's
like
some
day.
Maybe
if
we
ship
one
oh,
we
should
do
that,
but
really
not
important.
E
A
A
Yeah
I
will
also
be
sending
you
that
v1
alpha
3
one
there's
a
little
bit
of
hairy
API
machinery
stuff
going
on
with
that
with
the
conversion
between
2:00
and
3:30,
not
an
exact
one-to-one
had
to
write
some
manual
conversion
stuff,
I,
think
that
it
makes
sense,
but
I
haven't
done
a
manual
conversion
before
so.
We
appreciate
a
pair
of
eyes
on
that.
A
B
I
was
working
with
LeBlanc
in
he
was
using
day
and
what's
her
name,
the
cupid
mean
the
DI
doctrine,
doc,
I,
don't
know
one
bash
script
but
issues
to
implement
it.
So
the
goal
is
to
try
to
to
other
job
in
the
CI,
with
with
kind
with
a
PVCs
I
started
to
to
clean
some
of
the
tests
that
were
failing.
Well,
we
see
how
long
how
long
it
takes
to
to
cut
this
test
work.
Oh
thank.
A
B
A
So
fun
thing:
so
people
have
tests
where,
like
they
run
like
an
HTTP
server
app
in
their
cluster
and
then
they
want
to
like
talk
to
it
and
see
it
routed
to
different
versions
so
like
when
I
was
looking
as
K
native.
So
Canada
has
like
scale
to
zero
in
multiple
versions,
and
they
have
this
component
that
selects
two
different
backends.
A
A
That
should
mean
that,
from
inside
the
container
network,
you'll
see
you'll
send
a
request
out
to
the
external
IP
address.
Thanks
for
coming
to
you
so
you'll
see
it
a
request
going
to
this
IP
address
that
doesn't
exist
in
the
cluster
Network,
so
it
should
bounce
out
and
then
one
that
gets
to
the
host.
A
It
should
see
that
that
IP
address
exists
on
the
interface
and
then
go
to
the
thing
you
have
listening
on
the
load
balance
reports,
which
can
be
some
kind
command,
and
then
that
can
wrap
back
into
the
cluster,
and
so
you
can
actually
have
in
theory
and
load
balancer
IP
and
support
load.
Balancer
services,
okay
across
the
across
the
container
network
and
I-
should
even
work,
hopefully
unlike
Windows
and
Mac
as
well.
We'll
just
need
something.
The
thing
that's
listening
on
the
host
will
need
to
like
bounce
traffic.
A
A
This
4k
need
of
someone
else
implemented
this
thing
where
you
like
just
go
ahead
and
add
a
whole
like
edits
out
on
your
host
to
the
mini
cube
address,
and
then
you
run
a
controller
that
just
overwrites
the
external
IP
on
load
balancer
services
to
be
the
the
note,
IP
and
I'm,
not
sure
how
well
that
works,
but
that's
also
a
little
bit
of
a
different
problem,
because
it's
doing
the
VM
on
the
VM
interface,
as
opposed
to
like
the
containers
on
the
container
network.
So
we've
already
talked
for
a
while.
A
Now
we
should
be
able
to
do
something
like
maybe
SSH
tunnels,
due
to
like
feed
arbitrary
numbers
of
ports
into
the
network
is
the
alternative
is
that
we
would
need
to
like
bind
ports
to
the
containers
ahead
of
time,
an
arbitrary
number
of
them,
which
is
a
problem.
But
if
we
have
some
tunnel
mechanism,
then
we
can
just
open
up
ports
on
the
host
whenever
we
want
ports.
So
then
I
had
a
suggestion
from
Justin
sp2.
It
lets
us.
A
Because
you
could,
just
you
could
just
add
an
IP
address
to
a
near
face,
so
we
can
just
start.
We
could
just
add
a
private
IP
potentially
and
find
listening
for
traffic
on
that
IP
and
then
that
should
let
you
actually
create
multiple
load
balancers,
even
on
like
the
same
port,
not
sure
how
well
it's
going
to
actually
work.
There
may
be
various
problems
with
that,
but
it's
probably
worth
exploring
so.
B
A
Be
great,
finally,
I'm
also
planning
to
play
with
just
like
letting
you
put
host
ports
on
a
node
and
just
say
up
front
that,
like
okay,
I
need
to
have
these
ports
bound
to
the
host,
and
then
you
can
like
you,
use,
host
host
networking
for
your
pod,
hopefully,
and
that
will
go
through
to
the
container
port.
That's
bound
to
the
host
so
that
you
could
set
up
something
like
ingress
engine
X
and
serve
traffic
locally,
because
there.
A
A
B
D
A
B
A
A
yeah
I
think
it
I
imagine
that
the
the
biggest
problems
we'll
see
are
be
like.
If
you
add
this
interface,
if
you
add
this
IP
and
you
don't-
and
you
add
it
to
like
one
of
the
actual
one
of
the
actual
network
interfaces,
then
suddenly,
your
host
machine
is
like
advertising
IP.
Is
it
written
so
I
want
to
avoid
that?
Probably
and
the
other
thing
will
be
just
the
tongue.
The
mechanism
that
actually
gets
the
traffic
into
the
cluster
is
probably
going
to
be
awkward.
A
B
A
A
Well
so
I
think
maybe
it's
probably
one
container
per
instance
of
the
load
balancer
controller
running
which
will
need
to
be
actually
on
the
host,
because
it
will
need
to
open
a
port
on
the
host
to
listen
on
the
IP
address
and
then
route
back
into
the
cluster
Network.
But
then
you
can
talk
to
it
from
inside
or
outside
the
cluster
on
the
same
address,
which
is
what
you
would
expect
for
an
external
IP.
B
A
Not
sure
but
I
think
the
main
thing
is
that
you
want
it
as
closely
as
possible
to
be
able
to
have
like
a
set
of
manifests
for
your
app.
It's
like
the
same
thing,
but
like
running
in
a
fake
environment,
right
I,
guess
what
we're
trying
to
do
is
we're
trying
to
fake
out
kubernetes
just
on
your
machine.
D
B
A
A
D
B
A
A
Yes,
so
the
other
thing
there's
I'm
just
saying
I'm
just
take
some
intent
that,
like
for
going
into
like
kubernetes
one
one,
five
traffic
serving
is
one
of
the
areas
that
I
want
to
explore.
Better
I've
got.
Some
people
are
very
interested
in
using
it
but
they're
running
into
like
they
can't
run
their
whole
test
suite
because
anywhere
that
they
were
anywhere
that
they
had
traffic
surfing
and
doesn't
really
work
with
these.
A
A
B
B
A
A
A
B
A
F
Trying
to
run
the
conformance
test,
basically,
they
try
to
see
whether
they
are
running
or
not,
but
if
I
have
one
convict
I'll
and
try
to
see
whether
they
are
working,
the
kind
cluster
node
so
most
of
the
time
I
face
the
errors.
I'm,
not
certain
about
which
exact
error
I'm
facing
so
I
just
wanted
to
get
confirmation
like
whether
I
would
be
able
to
utilize
those
config
files
inside
to
see
like
how
the
bill
is
behaving
on
that
kind
cluster.
Is
it
possible
good.
A
Work
I've
brought
up
how
I
did
the
test
when
I
submitted
conformance
for
one
one.
Three
I
can
pull
that
up
and
we
also
have
continuous
tests
running
I've
noticed
that
the
if
you
try
to
run
them
in
parallel,
they're
a
bit
flakier
but
serially,
it's
been
pretty
reliable,
just
the
occasional
flake
it
should
work,
the
only
thing
being
that
you,
unless
you
have
the
latest
ones,
you
need
to
make
sure
that
it
knows
about
well.
A
F
I
use
the
master
plan
itself,
first
of
all,
create
the
base
image
in
the
node
image
and
then
question
so
and
also
like
I'm
working
on
promoting
existing
ii
ii
ii
ii
ii
conformance
and
before
that.
I
just
wanted
to
get
certain
guidelines
from
you
like.
What
kind
of
the
kind
of
the
area
would
be
there,
which
I
can
focus
on
like
before,
promoting
it
to
confirm
his
like
supposed
beings,
conformance
specific
version
or
seeing
whether
this
feature
is
good
to
promote
or
not
so
just.
A
F
A
Thanks
H,
let
me
know
if
you
have
any
specific
problems
with
this,
please,
like
pile
an
issue
or
say,
put
some
details
in
the
chat
or
something
I
should
work
I'm,
looking
through
the
dashboards
right
now,
and
we
didn't
have
like
a
flake
on
master
but
and
it
the
ones
that
are
using
the
Sun.
A
buoy
image
are
that
job
set
up.
It's
not
quite
right
yet,
but
the
other
versions
like
112
have
run
continuously
for
about
a
month
without
any
problems
now
running
the
full
and
confronted
sweet.