►
From YouTube: Network Plumbing WG Meeting 2018-09-13
Description
Kubernetes Network Plumbing Working Group meeting for 2018-09-13
A
A
So
that
should
be
up,
and
hopefully
that
will
serve
as
the
long
term
home
for
the
v1
specification,
I'm
kind
of
waiting
for
somebody
else
like
Tim
or
somebody
to
say.
Yes,
this
is
ok.
To
put
it
there,
I
think
we
talked
about
it
in
sig
network,
maybe
about
a
month
or
so
ago,
and
decided
to
all
try
this
Cabernets
community
repo
as
a
first
pass.
A
B
B
Okay,
let
me
paste
a
copy,
but
he
gets
cable
to
see
that
one
actually,
the
link
in
the
and
indigent
actually.
So
the
use
case
here
is
like
it's
like
deceased
cases
there
for
like
many
months
actually
so
they
the
use
case
the
Peters
were
saying
is
that
he
wants
to
expose
services
in
the
additional
network
as
well.
B
So
one
of
the
valid
use
case,
he
has
put
it
in
the
kubernetes
sig
group
on
his
particular
project,
and
there
was
multiple
people
who
are
asking
me
the
same
thing
when
I
was
working
and
the
multi
reference
implementation.
So
it
goes
back
to
the
same
conversation
we
had
like
one
year
before
what
will
be
the
kubernetes
service
impact
on
multiple
networks,
I
I,
remember
like
last
September,
2017
I.
Think
you
put
your
proposal
like
gathering
or
every
inputs
of
various
companies
has
put
various
proposals.
B
Then
you
consolidate
those
proposal,
so
I
just
took
that
proposal
and
then
trick
it
a
little
bit
for
the
multi
networking
with
use
of
reference
implementation
using
our
version,
one
spec,
so
how
it
looks
like
on
that
particular
implementation.
That's
why
I
came
up
with
a
suggestion.
Linkage
just
idea.
Actually,
so
you
can
do
the
multi
networking
with
service
supports
as
well.
C
B
B
B
B
Okay,
cool,
yes,
I
will
go
through
a
little
bit
actually
from
the
introduction,
so
this
pack
is
actually
you
were
spec
if
you
remember
Dan
long
back,
actually,
70
2017,
so
I
just
used
that
one
so
I
didn't
change
anything
on
the
kubernetes
core
or
EPA's,
or
this
particular
proposal
is
nothing
to
do
with
kubernetes.
So
the
entire
implementation
will
remains
outside
the
kubernetes.
B
B
There
may
be
minor,
changes
will
be
there
which
I
will
explain
later
and
what
I
was
proposing
when
trying
to
put
idea
is
that
the
service
impacts
should
have
kind
of
on
a
creation
in
which
we
can
specify
which
network
this
particular
service
you
want
to
expose.
So
you
just
can
Reeves
the
same
on
occasion
which
we
were
using
in
the
version:
1
spec
and
then
so
I'm,
currently
not
considering
into
the
load,
balancing
and
node
ports
I'm,
just
considering
only
for
that
lesser
IP
address,
so
that
these
things
will
be.
B
We
have
to
explore
a
little
bit
more
and
it
also
require
lot
of
efforts
on
the
load
balancing
stuff,
so
I'm,
not
considering
that
one
for
the
particular
initial
phase
discussion
or
for
the
POC
and
then
the
next
one
is
that
the
how
the
speck
of
over
all
the
services
and
the
pot
should
will
look
like.
So
it's
the
same
thing,
the
part
1
and
part
2
will
have
the
same
annotation
as
we
said
in
the
version
1.
B
The
interesting
part
is
the
next
one,
the
network,
so
the
one
thing
I
planned
thinking
of
introducing
is
the
service
er
D.
So
we
mentioned
in
the
network
attachment
definition
itself
which
servicio
D
it
has
to
be.
You
wants
these
particular
things
to
be
connecting
to
and
then
come
it's
the
main
part
of
routing
the
information.
Let
me
see
it
whether
consuming
it
out.
B
B
So
the
mechanism
of
what
I
was
trying
to
put
together
is,
like
you,
have
admission
controller,
which
could
be
able
to
watch
the
service
creation
and
if
it's
service
creation
is
happening
able
to
detect
what
is
the
network
which
is
specified
for
that
particular
service,
and
then
it
can
connects
to
the
controller
or
service
manager
or
whatever
the
mechanism
is
that
could
able
to
hook
into
mulches
and
say
it
for
this
particular
network.
You
have
to
connect
this
particular
interface
to
this
particular
service.
Virtual
IP
address.
B
And
the
DNS
is
the
one
I
was
discussing
with
Peter
whitey
is
serving
angel.
Work
is
currently
doing
on
the
DNS.
I
get
to
see
is
proposal
actually
what
he
is
currently
doing,
but
once
that
proposal
is
then
I
will
try
to
link
that
one
to
this
particular
Pivo
see
or
try
to
get
more
information
around
this.
But
this
is
just
a
initial
area
I'm
having
it
may
be.
A
Right,
it
looks
interesting.
One
thing
that
I'd
like
to
see
is
some
explicit
language
around
what
clients
that
aren't
multi
network
aware
would
do
with
a
service
that
specifies
a
network
like
how
would
that
work?
How
do
we
expect
that
to
fail?
How
do
we
expect
that
client
to
actually
behave?
In
that
case,
we.
A
I,
just
something
that,
because
I
mean
the
cube,
API
is
pretty
rich
and
there's
a
lot
of
things
that
talk
to
it.
So
assume
you
have
some
random
client
that
listens
for
services
may
be
tries
to
do
something
talk
to
that
service
whatever
what
would
happen
if
that
client
encountered
a
service
that
had
a
network
annotation
and.
D
A
A
A
B
C
B
I
will
do
that
one
so
I'm,
currently
working
on
that
one.
There's
endpoints
creation
actually
able
to
do
that,
one
so
that
that's
the
area
I'm
currently
working
on,
so
how
to
create
an
additional
endpoints.
As
for
that
particular
services,
actually,
so
that's
my
goal
to
explore
more
on
that
particular
thing
actually,
but
this
proposal
is
nothing
to
do
on
the
load
balancing.
That
is
something
little
bit.
We
have
to
see
more
use
cases
on
don't,
but
this
is
just
to
create
the
connections
as
for
the
additional
network.
A
A
B
B
F
B
B
C
A
F
A
B
Yeah
yeah
this
one.
Fortunately
we
got
a
appear
actually
for
the
bandwidth
and
capacity
flag.
Actually
I
mean
I
forgot.
The
Easter
cou
did
this
particular
implementation
work
actually,
so
that
PR
is
currently
there
and
it's
kind
of
working
actually
I
tested
the
P
R,
so
I
think
we
can
make
a
tick
for
this
bandwidth.
It's
almost
done.
B
Maybe
if
you
want
to
look
at
the
pier,
maybe
you
can
look,
but
it's
I
think
it's
done,
but
Oh
back
to
this
capacitive
bandwidth,
okay,
I
think
we
are
not
doing
this
on
iteration,
but
the
border
notation
it
can
get
the
pod
a
notation,
I
think
if
that
particular
PS
only
doing
the
capacity
flag,
actually
that
I'm
pretty
sure
yeah,
so
the
bandwidth
is
still
be
able
to
do
it.
Actually,
so
I
will
take
that
air
and
I
will
continue
that
one
and
on
the
v1
there
is
another
implementation
we
did.
B
B
It's
a
standalone
actually,
but
it's
wasted
in
the
repo
multi
stripper,
but
it's
a
standalone
admittance
similar
to
other
admission
control.
You
can
run
it
in
any
class
hid'
or
you
can
run
it
anywhere.
Actually,
the
admission
control
is
something
like
that
and
on
the
device
plug
in
and
CNA,
we
pretty
much
have
like
good
branch
on
masters.
That
is
a
boy
it's
aligning
with
the
device
plug
in
Maine
for
the
first
the
base
plane,
he
took
his
SR
away,
a
network
to
base
plugging,
and
it's
currently
working
actually
on
offer
engineer
called
up
made.
B
It
work
actually,
so
that
alignment
is
happening
and
on
the
Numa
I
we
have
angel
POC,
which
is
done
by
one
of
our
engineer,
Louise
on
the
Numa
on
the
alignment
with
device
plug-in.
So
that
is
also
currently
in
the
code
review.
So
it
may
take
some
that
couple
of
weeks
that
we
will
flush
everything
in
the
open
source.
Actually,
so
those
things
still
in
work
actually
yeah.
B
A
A
F
Let's
say
so:
I
think
we
take
talked
a
little
bit
about
the
sra1,
so
I
think
we
are
pretty
much
ready
to
talk
about
that
I
think
we
can
well.
The
next
meeting
is
is
OH&S,
so
we
probably
won't
be
able
to
do
that,
but
but
likely
the
one.
After
that,
we
should
be
able
to
do
a
demo
of
the
SRV
enablement
like
how
you
know
other
components,
work
together
and
but
I
think.
F
A
A
A
See
we
also
had
punted
until
this
meeting
seen
any
other
Duggars
suresh
around
for
dynamic
attachments.
We
had
questions
around
making
parts
of
the
spec
optional.
For
example,
a
plug-in
would
implement
selection
and
status,
but
would
not
implement
the
network
attachment
definition.
Part
of
the
specification.
F
A
I
I
think
that
actually
came
from
Peter,
maybe
right
originally
yeah
see
Peter
also
had
the
network
attachment
selection
annotation
for
replica
sets.
I
feel
like
I,
understand
the
problem,
and
so
I
added
that
text
there
about
replica
sets
and
IP
ranges,
but
it
would
be
good
to
have
him
on
the
call
to
discuss
that
further.
F
B
A
B
A
Okay,
I
guess
we'll
kind
of
put
this
list
again.
I
mean
again.
The
goal
for
all
those
kind
of
v2
stuff
is
to
identify
some
things
that
we
feel
like
we
can
do
in
the
next
couple
of
months
and
then
things
we
can
do
in
like
the
next
six
months.
For
like
a
larger
v2,
it's
not
looking
like
anybody
is
really
yelling
about
specific
things
that
they
want
for
a
v1
and
a
half
other
than
potentially
the
was
it
bandwidth
and
capability
flags
things
which
right.
E
B
B
The
is
that
that
something
that,
like
overall
signet
work,
is
trying
to
push
like
dual
stack
support
like
currently
what
happens
in
communities
like
either?
You
can
run
it
with
ipv4
or
ipv6,
but
the
problem
in
this
case
that,
if
you
want
both
the
things,
then
you
have
to
have
expensive
net
between
these
two
components,
or
else
you
have
to
have
one
cluster
on
ipv4,
another
cluster
on
ipv6
and
it's
kind
of
difficult,
actually
yeah.
E
B
A
Okay,
given
all
of
that
I
think
we
will
probably
try
to
wait
until
the
next
meeting
for
some
of
these
things,
but
Fung
you're,
saying
that
you
and
possibly
Doug
and
maybe
Toma-
will
not
make
the
next
meeting
due
to
ons,
correct,
okay
and
and
crawl
and
I'm
guessing
a
bunch
of
other
people.
Okay,
I
guess
the
ask
until
the
next
meeting,
so
that
you
know
maybe
we
can
try
to
get
at
least
a
couple
of
these
knocked
off
for
the
next
meeting
would
be.
If
your
name
is
next
to
any
of
these
items.
A
D
F
So
a
question
I
have
about
regarding
just
this
list
the
way
we're
going
to
work
through
this.
It
looks
like
you
know.
We
said
there.
There
are
a
list
of
potential
short-term
items
that
you
know
we
can
do.
But
if
we
are
not
seeing
a
lot
of
traction
in
some
of
those
sort
of
short-term
items,
can
we
go
ahead
and
and
say:
let's
go
ahead
and
look
at
some
of
the
potential
longer-term
items
and
maybe
start
working
on
those?
Yes,
maybe
we
don't
need
a
1.5
right,
yeah.
A
F
E
A
Or
two
because
obviously
yeah
everything
sure
mark
the
ones
that
you're
interested
in
talking
about
okay
and
maybe
fill
in
some
detail
around
existing
efforts
there
or
what
the
kind
of
status
of
that
is.
If
anybody
has
thoughts
on
where
that
could
go,
and
then
we
can
start
like
shedding
on
all
of
that
kind
of
stuff.
All
right,
yep
sounds
good,
all
right
anything
else.
Anybody
wants
to
discuss.