►
From YouTube: Network Plumbing WG Meeting 2018-12-06
Description
Kubernetes Network Plumbing Working Group meeting for 2018-12-06
A
B
You
know
I've
got
the
thread
on
sig
net
worth,
but
as
Mike
noted,
it's
like
not
great
for
like
expressing
a
range
of
when
is
good
in
terms
of
like
this
is
a
goodish
versus
this
is
really
good
type
of
thing.
It's
a
little
bit
more
binary,
so
maybe
you
could
see
I
mean
it's
like
Fung
had
put
down
that
Tuesday
morning
was
good
like
early
on
Tuesday
morning.
That's
good
ish
for
me
as
well.
B
C
If
it's
an
hour
that
I
think
it's
a
little
easier
to
fit
in
I
feel
like
I
think
so.
Tuesday
does
not
work
for
me
at
all
other
than
early
morning.
That's
the
same
for
now.
I
would
also
vote
against.
I
would
vote
against
Thursday,
because
I
suspect
people
leave
yep
at
least
Wednesday
right
so
Wednesday
or
early
Tuesday
morning.
If
people
want
to
get
up
like
I
mean
not
that
early
right,
so
I
mean
like
what
9
or
something
nine
o'clock
might
work
on
Tuesday
yeah.
So
something
like
that
is
one
option.
A
E
Right
I
mean
the
yeah
I
mean
the
problem
is
basically
this
conference
has
no,
you
know
open
or
unconference
time
right,
yeah,
no
space
for
buffs
or
anything
else
and
formal
like
this,
so
I
think.
For
me
the
least
interesting
sessions
are
actually
the
first
ones
on
Tuesday,
so
like
from
1020
to
1140.
E
A
A
F
C
F
E
C
B
B
B
C
B
C
Don't
we
do
to
a
four
o'clock?
Why
don't?
If
that's
the
case?
Let's
not,
because
if
the
party
starts
at
6:00
and
we're
throwing
or
if
it
is
somebody
is
going
to
the
party,
then
they
probably
need
transportation
at
like
five
or
something
like
that
right.
So,
let's
not
even
risk
that
you
know
why?
Don't
we
just
do
four
and
then
you
know
and
then
we'll
schedule
four
to
five
and
who
can
join
in
you
know
all
that
right:
okay,
okay,.
B
C
F
F
B
A
C
A
C
A
A
There
have
been
people
in
the
past
from
the
west
coast
and
I.
Don't
think
we
want
to
you
know
intentionally.
You
know,
shut
people
out,
but
we
haven't
had
too
many
from
the
west
coast
recently,
okay,
so
yeah
I
think
that's
okay.
Why
don't
we
make
that
decision
on
the
mailing
list?
Give
people
one
more
chance
to
object
and
if
they
don't
we'll
just
call
it
a
time
or
call
it
D
time
for
the
next
meeting,
yeah.
A
I
think
like
I
mean
the
next
meeting
December
20th,
you
know
which
I
think
we
all
said
that
we
would
hold
yeah.
Let's
just
put
that
out,
there
say
hey.
We
are
gonna
change
this
time
unless
anybody
we're
going
to
change
to
this
time
unless
anybody
really
strenuously
rejects
and
Norman
next
week,
we
can
just
change
the
invite
and
everything
like
that.
If
we
don't
get
any
objections,
okay,
okay
and
hopefully,
then
we
can
try
to
get
a
little
bit
more
involvement
from
Europe
as
well,
which
you
know
it's
been
a
problem.
You
asked
right.
E
F
B
B
C
C
B
A
The
problem
is
basically
that
we
were
following
some
convention
from
CNI
when
we
specified
the
static
IPS
in
the
network
selection,
annotation,
and
that
convention
did
not
include
a
network
prefix
for
the
IP
address.
So
our
documents
do
not
talk
about
that
explicitly,
but
it
turns
out
that
for
static,
IP
use
cases
where
there
is
no
other
configuration
around
subnets
or
ranges
that
doesn't
really
work,
because
then
that
plug-in
has
no
idea
what
that
network
prefix
should
be.
So.
A
So
those
two
parts
marked
in
green
are
the
changed
parts,
basically
just
a
state
that
the
network
prefix
is
optional
for
those
addresses
just
to
allow
existing
CNI
plugins
like
host
local,
to
continue
working,
but
that
if
plugins
require
a
prefix,
they
are
allowed
to
fail.
The
attachment
request,
if
they
do
not
receive
one
basically
I,
think
the
upstream
recommendation
will
be
that
plugins
should
accept
a
prefix
and
that
users
should
probably
is
a
prefix,
no
matter
what,
when
they
can
configure
these
things
and
by
when
I
say
upstream,
I
means
CNI
upstream.
A
Should
be
yes,
so
host
local
requires
basically
host
local
will
farm
out
addresses
out
of
the
subnet
that
it
is
configured
for,
but
you
can
request
a
specific
address
out
of
that
subnet
and
if
that
address
is
not
already
allocated,
it
will
give
you
that
address
right
and
it's
very
easy
for
a
host
local
to
take
a
bear,
IP
address
without
prefix
and
say:
oh
hey.
This
range
that
I
already
have
configured
contains
the
scipy.
Therefore
I
know
what
the
prefix
would
be
right.
E
A
A
D
A
E
E
D
A
A
We
could
just
go
with
that
and
I
could
update
the
document,
but
what
I
think
I'm
going
to
do
is
leave
it,
because
I
think
we
probably
want
to
update
this
document
the
spec
a
little
bit
more
to
clarify
that
we
are
now
on
like
version
1.1
or
something
like
that,
so
I
will
go
through
and
try
to
do
that
and
then
bring
in
this
text.
When
I
do
that,
and
then
we
should
probably
have
to
change
your
section
to
note
which
stuff
has
been
changed.
A
All
right,
if
not
so
this
one
is
kind
of
a
carryover
from
November,
8th
I
proposed
to
add
section
7.5,
which
is
in
me
like
one
time.
Excuse
me,
rent
time,
considerations,
implementation
considerations.
Sorry,
let
me
figure
out
what
that
section
is
actually
named
that
run
time
and
implementations.
Considerations
is
the
section
and
we
would
add
a
section
or
an
item
to
that
section
to
say
that
any
runtime
configured
Cabernets
to
the
implementation
via
CNI
Jason
configuration,
must
not
be
passed
to
you
or
applied
by
network
attachments
other
than
the
cluster
wide
default
network.