►
From YouTube: Multi-Network community sync for 202306221
Description
Multi-Network community sync for 202306221
A
B
Yeah
thanks
Monte
yeah
I
just
wanted
to
bring
up
that
I
I
feel
like
sometimes
we're
getting
kind
of
stock,
slash
in
a
log
Jam
here
and
again
with
people
bringing
up
like
the
cases
that
are
important
to
them
and
we'll
we'll
have
a
situation
where
it'll
be
like
one
person
will
be
like.
Well,
that's
important
to
me.
Another
person
will
be
like
that.
Doesn't
fit
my
scenario
and
I
on
either
end
of
those
discussions.
B
I
I
think
that
that
stuff
is
important,
but
what
I
feel
like
is
or
not
necessarily
capturing
those
kind
of
conversations,
because
I
think
that
we
can
get
good
output
from
the
results
of
getting
people
on
the
same
page.
So
I
wanted
to
kind
of
bring
up
like
the
idea
of
maybe
forming
some
kind
of
strategy
so
that
we
can
maybe
both
capture
more
of
that
discussion,
but
also
give
us
a
way
to
maybe
like
potentially
Park
those
conversations
and
allow
people
to
work
on
them.
B
Asynchronously,
like
maybe,
have
some
kind
of
a
thing
like
where
we
can
be
like:
hey,
okay,
we've
identified
like
we've
been
on
this
problem,
for,
let's
say
seven
minutes
and
we're
stuck
like.
Could
we
say
like
okay,
hey
this
merits
more
discussion,
then
maybe
we
could
have
people
write
it
down
and
then
have
them
put
in
the
kind
of
data
or
evidence
that
backs
their
position,
and
then
maybe
we
could
cycle
back
through
this
kind
of
stuff
on
a
regular
basis
and
kind
of
what
I'm
thinking
is.
A
Yeah
I
hear
you
I
think
that's
that's
fair,
like
not
because
right
now
what
at
least
I'm
trying,
because
I'm
I'm
driving
most
of
the
time
trying
to
get
those
results
on
the
meeting
right
and
then
I'm
waiting
for
the
next
one,
and
then
that
takes
time.
I
agree
with
you.
That's
a
fair
point
to.
If
we
have
some
contention
around
anything
just
shelf,
it
have
some
documentation
set,
do
it
offline
and
then,
if
any
any
additional
doubts
then
touch
it
on
the
meeting.
B
A
We
can
we
can,
we
can
start
creating
action,
items
or
I,
don't
know
action
requests
or
whatever
you
want
everyone.
Every
company
has
their
own
names
for
that,
but
basically
items
that
someone
has
to
kind
of
work
on
for
the
next
meeting
to
to
kind
of
that
kind
of
things.
That's
what
you
have
in
mind.
D
B
And
I
think
that's
a
good
way
to
freeze.
It
too
is
like
yeah
assign,
assign
the
actions
and
or
whatever
we
want
to
call
them
and
then
yeah.
Let
people
go
off
and
and
do
that
and
then
also
I
think
it'll
be
good
because
it'll,
it's
you
know
it's
just
human
nature
that
there
are.
There
are
going
to
be
some
people
that
are
better
thinking
about
them
in
their
quiet
space
and
there's
going
to
be
other
times
when
people
are
have
a
loud
voice
about
stuff,
and
then
we
can.
A
The
thing
here
is
I
would
I
would
try
to
keep
myself
in
check
as
well,
but
I
would
welcome
other
folks
participate
and
catch
me
or
any
of
the
discussions
in
such
a
state
right.
So
we
don't
bog
Downs
ourselves
and
maybe
provide
this
top
Gap
kind
of
during
the
discussion
and
and
and
and
kind
of
to
take
a
step
back
and-
and
do
this
right,
create
this
action
item
from
from
a
specific
topic
so
that
we
can
continue
on
something
else.
A
Yeah
yeah
exactly
so
so
I
I
I
ask
everyone
to
kind
of
participate
and
then
maybe
say
that
right
that,
if
something
is
being
prolonged
and
and
being
kind
of
bogged
down,
let's
please
call
it
out
everyone
and
then
we
can
shelve
it.
And
then
let's
discuss
that's
next,
we
can
let
go
move
on
so
please.
Let's
do
that
and
I
encourage
everyone
to
kind
of
participate
and
and
catch
dogs
as
well.
B
A
All
right,
I
have
another
bullet
I
added
you
Tomo,
as
as
the
as
as
the
owner,
but
I
added
that
one.
This
was
from
the
two
weeks
before
a
meeting
the
recording
was
finally
there,
so
I
always
managed
to
look
at
it.
So
there
was
discussion
you
brought
up
a
a
topic
of.
Can
we
have
multiple
items
in
the
param's
reference
of
the
Pod
Network
I
think
that
was.
That
was
something
that
you
brought
up
and
I.
A
Think,
because
you
wished
to
kind
of
reference
like
not
only
the
custom
stuff
like,
for
example,
you
would
you
would
reference
Network
attachment
definition
and
you
would
reference
a
resource
claim
or
something
like
that
right.
One
of
the
examples
I
think
it's
it's
a
matter
of.
Do
we
do
we
want
it
because,
from
the
implementation
point
of
view,
it
doesn't
matter
really
because
core,
at
least
from
the
core
Point
from
the
core
Point
it
the
car
will
never
parse
those
elements
right.
A
It
will
only
look
at
whether
those
exist
or
not.
That's
the
only
thing
that
the
core
will
do,
and
then
it
is
up
to
the
implementation
on
how
this
is
going
to
be
handled.
That's
what
is
it
basically
is.
The
only
concern
I
would
have
this
kind
of
come
to
my
mind
is
the
conditions
that
I
was
kind
of
trying
to
build
behind
this
Let
Me
Maybe
show
what
I
have
in
mind
on
that.
A
So
we
have
three
conditions,
but
basically
what
we
have
is
the
params
already,
and
if
you
look
at
the
example
at
the
description
of
the
of
the
params
ready
is
it
should
be
populated
by
the
implementation,
if
the
if
the
parents
ready
field
is
anything
other
than
nil
right.
So,
basically,
if
I
set
anything
here,
the
car
is
going
to
expect
that
someone
says
this
field
right.
Someone
said
sorry,
not
field.
Someone
sets
this
condition.
A
Params
right
in
the
network
object-
and
this
is
kind
of
boils
down
how
conditions
work,
I'm,
not
sure,
folks,
how
familiar
how
folks
are
familiar
with
the
conditions
themselves,
but
think
the
conditions
that
we
have
on
a
pod
right
when
pod
starts
up.
There
is
a
I
think
phase
condition.
Basically,
there
is
always
a
ready
condition
and
what
happens
is
there
is
a
centralized
controller
for
the
ready
conditions,
but
then
there
are
other
conditions
for
the
other
elements
right
drawn
by
some
other
components
like
cubelets,
mostly
mostly
by
cubelet
right
and
basically,
what
happens?
A
Values
are
correct
right,
so,
basically,
if
the
pod
containers
didn't
came
up
ready
because
kublet
or
c
r
I
didn't
came
back
and
say:
okay,
I
I
started
all
the
conditions.
The
then
the
main
ready
conditions
will
never
get
set
ready
because
the
previous
one
are
not
there
yet,
okay,
so
there
is,
you
can
build
some
sort
of
dependency
or
between
the
conditions
with
that
right,
each
controller,
con
owning
specific
conditions
and
specify
and
providing
values,
and
then
you
can
create
some
sort
of
dependency.
A
A
If
you
were
to
have
a
list
right,
your
implementation
has
to
Now
look
for
all
the
all
the
all
the
parameters
and
then
check
whether
they
are
okay
to
say
that
the
whole
thing
is
ready,
because
you
have
only
one
conditions
for
that
field
right
and
for
the
Readiness
of
of
the
of
the
parameters.
So
this
is
where
the
complication
goes,
because
this
can
be
one
of
the
approaches
where
I
can
have
One
controller
handling,
all
the
parameters
in
the
list.
But
what?
A
A
I'm
going
to
show
you
did
I
explained
this
correctly.
Hopefully
I
did
so.
That
would
be
my
argument
to
just
keep
it,
as
is
to
single
field,
because
then
we
know
I
have
one
controller
for
this
field
and
for
this
condition
that
says
that
looks
at
the
origin.
The
parameters
object,
validates
it
do
whatever
it
needs
with
it,
and
then
reports
to
the
network
object.
Oh
okay,
my
my
reference.
A
F
E
I
mean
and
then
also
the
your
your
explanation
have
the
you
assume
that
the
each
the
so
that
this
part,
if
you,
if
the
performance
level
is
direct
at
that
time,
the
there
are
the
multiple
element
is
manipulated
by
the
each
controller.
I
mean
that
the
one
controller
man
is
the
one
object
only
and
then
the
and
then
the
hence
the
controller
is
two
or
more
but
I'm,
so
yeah
I'm.
Also
thinking
that
the
one
provider,
of
course
the
provider,
is
one
string.
E
I
do
not
changing
this
stuff,
so
one
provider
have
the
One
controller
which
managing
the
multiple
parameter
graphs.
So
this
is
also
the
body
to
design
I
think
so.
Maybe
the
Euro
design
you
talked
about
I
mean
that
the
promise
rest
are
managing.
The
first
element
is
managed
by
controller.
A
second
is
managed
by
the
controller
B
at
that
time.
Maybe
the
Airport
network
status.
E
A
E
Be
some
no
no
I'm
just
thinking
the
One
controller
money
in
the
both
pumps,
I
mean
at
the
one
network,
object
having
an
apartment's
left
and
then
here
and
the
multiple
I
mean
that
okay,
that's
pumps,
they
have
the
object
there
and
object
B
and
object
C,
and
then
these
three
object
is
managed.
One
controller
by
the
provider
field
of
their
Network.
G
A
It
it
can
in
some
ways,
I
I
know
that
there
is
always
because
there
is
at
least
finalizers
that
can
prevent
you
from
deleting
an
object.
So
basically
you
can
create
a
finalizer
put
the
finals
on
an
object.
Of
course,
finalizer
can
be
deleted
by
Asia.
Then
the
other
ways
is,
you
can
create
a
web
hook
that
will
prevent
complete
deletion.
If
some
condition
is
met,
then
you
can
delete
the
web
hooks.
A
But
at
that
point
we're
talking
about
like
kind
of
hacking
right
and
yes,
of
course,
when
you
hack
you
can
you
can
eventually
get
yourself
to
a
broken
state
and
that's
okay,
someone
hacked
you,
someone
kind
of
not
hacked
you,
someone
just
on
purpose,
just
trying
to
break
stuff,
but
in
normal
usability.
That's
that
that
would
work
right
and.
A
Okay,
so
the
con
those
conditions
are
for
those
conditions
are
for
the
the
rest
of
the
implementation.
So,
basically,
if
if
an
airport
network
is
not
ready,
it
will
never
be
used
by
a
cubelet,
it
will
see.
Oh,
this
object
is
unused.
It's
not
ready.
I
I,
you
reference
an
unready,
unready,
Network,
pod,
Network
and
basically
I
cannot
create
this
pod.
A
G
G
A
So
those
would
be
twofold
right.
It
will
be
definitely
I,
think
we're
just
trying
going
over
the
main
topic,
but
those
with
be
two-fold
first
debuggability
right
and
the
other
thing
will
be
kind
of
a
signal
for
your
implementation.
It's
what
you're
saying
is
your
implementation
has
to
use
it,
but
then,
if
we
don't
have
it
here,
how
what
is
the
other
way
to
kind
of
to
to
Define
that
right,
I
agree
with
you.
A
If
your
implementation
doesn't
respect
those,
then
they
are
meaningless,
but
then
add
another
more,
but
otherwise
how
would
I
do
that
right?
There
is
no
other
thing
that
I
can
do
this
right
and
I
can
have.
This
is
very
simple
controller
that
I
can
have
in
core
that
will
check
whether
you
set
the
palms
or
not,
and
whether
you
set
the
params
ready
condition
and
then
I
will
set
the
ready
for
you
and
then,
if
there
is
any
pod
using
this
network
because
it
was
already
and
anyone
use,
a
pot
is
referencing.
A
This
network
I
will
set
the
in-use
condition.
So
basically,
those
are
debuggability
in
information
right
metrics,
even
you
might
say
so,
whether
you're
gonna
use
them
or
or
not.
That's
that's.
A
kind
of
up
to
implementation.
I
agree
with
you,
but,
for
example,
the
in
use
is
very
useful
right
because
this
can
be
very
well
handled
by
the
core
by
the
controller
in
the
core
which
we'll
see
okay,
someone
reference
this
pod
Network
in
a
pause,
so
I
will
mark
it
as
in
use
and
basically
I
agree
with
you.
A
If
you
want
to
do
anything
else
with
it,
then
then
do
or
that,
but
the
at
least
the
in
use
is
very
useful
because
I
see
then,
okay,
there
is
some
pod
in
my
cluster.
That's
using
it
is
the
spot
Network
and
then
there
is
some
dependency
on
this
one,
because
then
you
cannot
use
a
delete
that
pod
and
since
I'm
in
core
there
there
is
no
workaround
for
this
one.
So
the
in
use
is
very
useful.
The
ready
I
agree
with
you.
A
This
is
up
to
the
implementation,
whether
they
were
respect
it
or
not,
right,
but
still
it
gives.
This
is
a
very
cheap
way
to
do
and
gives
you
a
lot
right
and
whether
you
you
use
it
or
not,
then
that's
another
thing,
so
I
don't
see
I,
don't
think
what
you
said
that
oh
it's
it's
not
really
it's
not
going
to
be
used
by
the
core.
So
we
shouldn't
have
it
I,
I,
I,
I,
don't
agree
with
you
on
that
one,
because
it's
it's
for
the
at
least
debuggability
is
very
useful.
A
A
If,
if
you
won't
leverage
that,
then
then
that's
your
implementation,
then
you're
gonna,
maybe
miss
out
on
something
on
debuggability
items
that
you
your
implementation
could
have,
and
if
you
long
term
you
kind
of
wish
to
really
support
multi-networking.
You
definitely
would
want
to
have
that
for
your
own
sake
right.
If
you
want
to
support
something
like
that
did
I
I.
Think
apart
took
you.
First
did
I
answered
your
kind
of.
A
G
Seems
a
bit
subjectable,
and
this
is
important
enough,
but
I'll
just
take
it
at
face
value,
and
if
you
say
that
it's
important
then
I
guess
it
is
I,
don't
have
any
particular
stake
in
this
I'm,
not
implying
I'm,
not
implementing
a
a
a
provider
myself
I'm,
just
looking
at
it
from
a
from
a
cap
and
API
perspective
and
I
wonder
what
is
it
good
for?
Is
it
called
out
in
the
documentation
how
it's
supposed
to
use?
What
is
the
must
for
a
provider,
for
example,.
C
A
A
I
am
saying
they
are
because
they
are
a
great
most
of
them.
Most
of
this
will
be
implemented
like
all
of
this
will
be
implemented
by
condition,
except
for
the
Palms
red
which
responsibility
falls
on
the
implementation
of
the
referenced
parameters,
ref
right.
So
basically,
if
if
this
field
is
empty,
nobody
is
going
to
expect
this
guy.
A
But
if
someone
provides
this
field
in
the
Pod
Network
object,
then
the
the
automation
will
expect
the
parents
ready
condition
should
be
set,
and
then
that's
that
responsibility
falls
on
the
controller,
and
this
is
in
the
text
as
well,
yes
and
now
boiling
down
to
whether
this
can
be
a
list
or
just
a
single
element,
I
think
a
list.
This
is
just
Tomo
I,
think
if
we
were
to
have
a
list
to
be
honest,
I
don't
have
a
strong
feeling
about
this
any
other.
Anyone
else
has
some
some
opinion
on
this.
A
A
But
then
you
could
same
way
reference
the
additional
object
in
your
own
reference
right
so
to
make
a
kind
of
a
chain,
but
maybe
not
maybe
we
should
have
a
list
here,
but
then
this
complicates
on
this
params
ref,
ready
condition,
because
then
you
would
have
to
have
a
centralized
element
that
understands
all
of
the
objects
in
the
refs
right
and
then
can
summarize
everything
together
and
then
finally
set
the
condition
as
one
thing
that
will
be
the
the
kind
of
the
the
hurdle.
If
we've
had
this
a
list.
A
A
F
One
General
thing
on
this
kind
of
stuff:
it
feels
to
me
like.
If
it
were
a
list,
would
there
be
an
impact
on
people
who
only
need
a
list
of
length
one,
and
there
may
be
some
people
who
do
have
a
use
case
that
care
about
it.
So
my
instinct
would
be
if
there
is
a
use
case.
If
we
come
up
with
one,
we
should
probably
be
fairly
permissive
about
it,
I'm
quite
relaxed,
but
let's
leave
it
till
we've
got
the
right.
E
Yeah
I'm:
okay,
that's
the
right
thing
now,
but
the
one
example
is
about
the
one
network:
Port
Network
object,
maybe
shared
with
the
multiple
net
attach
theft
case
I
mean
that,
let's
imagine
that
the
one
network-
okay,
they
are
from
the
network-
topology
side.
One
network
is
connected
by
SRB
interface
and
the
Mac
B9
interface,
then
of
course,
and
then
also
they,
this
network
won't
using
the
default.
The
ipam
is
managed
by
the
kubernetes.
E
You
know
like
not
external,
and
then
the
user
is
thinking
that
the
they
have
the
they
want
to
launching
the
two
kinds
of
the
port.
Roughly
one
pot
is
important
part.
This
is
requires
the
high
speed
network
bandwidth
or
some
stuff
that
at
that
time,
they'd
like
to
attaching
the
SRB
interface
this
network,
and
then
it
took
another
kind
of
leopard-
is
the
cheap
kind
of
Port.
E
D
E
A
So
no
I,
don't
definitely
don't
want
to
go
for
that.
The
only
definitely
that
I
would
be
against
to
make
ability
to
select
multi
kind
of
one
of
this.
This
input
Network
represents
one
connection
to
a
pod.
If
you,
if
you,
if
we
make
this
a
list,
it
will
be
only
to
for
you
to
provide
additional
parameters
for
that
single
network,
not
to
provide
multiple
additional.
A
A
I
think
at
that
point,
probably
not
if
you
you
were
to
use
the
kubernetes
iPhone
if
and
you
want
to
share,
because
what
you're
trying
to
achieve
is
shared
aside
there
or
a
subnet
between
two
networks
that
you
defined
and
if
you
want
to
do
that,
do
your
iPhone
on
your
own,
not
through
kubernetes,
because
kubernetes
would
assume
yes
separate
subnets
per
in
pod
network.
But
if
you
want
to
share
them,
you
can
do
it
same
way
today.
You
do
it
today,
right,
you
have
the
whereabouts
or
a
static
or
one
of
those.
A
So
and
that's
what
you
would
do
it
here
in
the
reference
them
input.
Network,
it's
not
like
you
can
have
what
you
just
described
with
one
network
attachment
definition.
You
have
to
have
two
separate
anyways
that
will
I
mean
ipam
Define,
those
things
for
you.
Yes,
kubernetes
here
uses
whatever
the
default
icon
can
do.
A
If
you
want
to
do
something
what
you're
describing
then
you
would
have
to
that's,
that's
not
the
topic
we
are
not
talking
about,
because
now
what
you're
saying
is
that
can
be
completely
a
separate
cap
where
you
would
want
to
change
how
the
iPhone
is
done
in
kubernetes,
where
okay
I
want
to
support
a
case.
What
you
just
described,
where
I
can
share
a
single
subnet
across
multiple
pod
networks,
and
this
is
something
that
can
be
an
extension
in
in
the
future
to
what
we
are
creating
here.
E
F
B
F
E
A
E
A
E
You
mean
that
the
port
network
is
just
their
net
attacks
there
right,
yes,
but
so
from
I'm
I'm.
Just
talking
about
the
let's
imagine
we
have
some
newbie
is
joined.
Is
that
it's
looking
in
the
multinetal
cap
at
that
time?
They
are
this
kind
of
then
they
are
this.
This
newbie
is
reading
the
port
Network
right.
F
E
That
time
that
this
user
must
understanding
this
is
the
network
object
because
the
portal
Network
it's
similar:
okay,
Auto,
Network,
okay,
so
this
means
the
user
is
thinking
about.
This
is
kind
of
the.
If
some
someone
is
good
at
the
openstack
openstack
neutral
right
at
that
time
there
this
is
pretty
one
network,
and
then
one
network
is
just
the
computer
network.
I
mean
that
yeah,
which.
F
A
Just
to
let
you
know
how
a
network
is
represented
because
you're
trying
to
get
and
push
one
model
of
how
a
network
is
viewed,
yes,
that
user,
that
you
described,
views
the
network
as
such,
but
then
that
is
that
boils
down
to
how
the
network
is
visible
by
the
implementation.
Not
by
this.
This
just
is
an
API
AP,
a
common
API
across
the
board,
so
that
you
don't
have
to
use
annotations
and
then
have
the
integration
with
core.
That's
the
core!
A
That's
the
main
concept
of
this,
and
this
represents
a
network
and
you're
trying
to
say
now:
oh
it
represents
Network.
So
it
has
to
mean
that
no,
we
said,
that's
why
we
don't
call
it
a
network.
We
call
the
Pod
Network,
because
pod
network
is
just
a
handle
to
wherever
you
have
in
mind
what
a
network
means
for
you.
A
So
don't
please
don't
make
it
in
a
in
a
in
a
stable,
like
definition,
one
definition
of
it,
because
we
almost
explicitly
call
it
pod
Network,
because
it's
just
a
representation
of
a
network
that
your
implementation
think
it
is.
If
your
network
thinks
a
network
means
just
a
a
slice
of
a
subnet.
That's
what
it
is.
If
you
want
to
make
a
pod
Network
a
whole
subnet,
it
can
be
that
if
you're
interested
supports.
E
Okay,
let
me
let
me
sharing
that.
My
concern
is
if,
okay,
let's
imagine
the
one
one
network
is
connected
to
SRB
and
then
the
market
beta
and
then,
as
you
mentions,
with
the
user,
is
creating
the
two
Port
Network
object
and
then
in
the
future,
of
course,
the
other
one
they
want
to
configure
the
amounts.
The
network
policy
run
service.
At
that
time.
The
current
video
I'm
thinking
the
the
for
each
Port
Network
yep,
the
network
policy
and
services
created
at
that
time.
C
A
D
D
I
think
that
a
different
view
of
what
Tom
was
saying
or
I
might
be
wrong,
but
something
that
or
how
he
interpret
it
is
that
when
we
are
specifying
the
port,
the
port
networks
we
allow.
We
have
these
parameters,
ref
that
allows
implementations
to
add
their
own
parameters
to
that
Network
right
and
then
the
spec
has
also
some
parameters
for
the
Pod
Network
attachment
Port
network
name
interface
name.
So
the
this
spec
is
recognizing
the
data
model
that
an
attachment
is
intrinsically
different
than
a
network.
D
A
A
So
I
would
ask
everyone
here
who
has
any
mistake
in
the
in
the
initial
stories
to
and
end
the
requirements
to
go
through
them
again
because
they
might,
they
might
have
slightly
diverge
to
what
we
initially
kind
of
come
up
in
like
January,
that
was
kind
of
half
a
year
ago,
right
now
I've,
as
I
mentioned
last
week,
the
cap,
what
we
are
making
is
we
create
this
one
cap
and
we're
gonna.
Keep
iterating
on
it
with
what
we're
doing
today.
That's
what's
gonna
happen,
so
please
give
the
cap
another
read.
If
you
want
to.
A
There
is
a
lot
of
comments,
so
you
can
open
in
it
in
my
branch
and
just
go
browse
to
that
file,
you
can
have
a
clean
file
with
all
the
markdown
scene,
so
you
can
read
that
and
then
comment
on
the
ad
on
the
pr
itself.
So
I
recommend
going
through
that
again.
If
anyone,
some
of
you
didn't
look
at
it
for
some
time,
because
I'm
trying
to
push
the
leadership
to
finally
get
LG
TMS
on
this
and
finally
merge
it,
because
it's
just
taking
so
long.
A
Okay,
now
getting
back
to
the
reference
so
last
week
we
I
think
finalized
on
the
discussion
of
the
kind
of
how
things
can
be
done
in
terms
of
like
cnis,
because
I
think
the
initial
discussion
was.
What
does
this
primer
really
means
right
here,
because
that's
what
we
cannot
Boggle
down
on
on
six
versus
four
and
then
how
can
that
be
done
and
managed
I
think
Michael?
We
are
on
the
same
page
now
on
what
I
was
saying.
A
H
H
A
H
Let's
tie
this
back
a
bit
to
the
prior
conversation
quickly,
we
can
once
you
have
what's
called
a
dual
home
network:
I,
don't
care
if
it's
kubernetes
doesn't
make
a
difference.
You
have
two
different
cases.
The
ietf
compliant
implementations
can
use
both
of
them.
Concurrently,
you
can
send
packets
on
one
receive,
applies
on
the
other,
etc,
etc.
H
Ip
address
functionally
is
on
loopback,
whether
you
put
it
right
on
the
interface
or
in
the
IPM,
for
a
given
net,
attach
whatever
you
want
to
call
it
Linux
and
every
implementation
I
know
of
you
can
reach
that,
regardless
of
what
interface,
what
why
are
you
using
and
any
definition
contrary
to?
That
is
just
confusing
users.
H
I
agree
and
I'll
argue
now.
If
you're
on
a
cloud
you
can
set
things
up
such
that
the
cloud
will
enforce
that
it's
bi-directional,
even
though
IP
doesn't
doesn't
force
you
to
the
cloud
right.
If
your
implementation,
if
you
pick
a
VPC
on
AWS,
for
example,
you'll
get
that
bobbing
changes
in
the
last
six
months.
That's
fine,
but
we
don't
need
to
do
anything
here.
Then,
the
same
with
there
are
implementations
on
beer
metal.
A
Okay,
but
then
how
do
we
address
the
other
use
cases?
This
is
what
you're
saying
or
your
use
case,
but
but
then
we
have
to.
We
want
to
kind
of
apply
to,
like
the
other
use
cases
that
don't
only
con
kind
of
boil
down
to
just
the
conflicts
right
so
having
I
think
somewhere,
I
was
I.
Think
somewhere
in
the
text,
I
will
have
to
read
it
again.
A
I
think
I
had
somewhere
that
one
of
those
elements
has
to
have
a
primary
field
set,
and
this
is
something
that
I
need
to
remove
for
your
use
case
right
to
say
that
okay,
I,
don't
have
to
set
it
right,
I'm,
not
sure
whether
you
can
use
in
conflicts.
The
interface
name
field
at
all.
At
that
point,
if
it's
a
cni,
because
okay,
can
you
take
that
as
an
argument?
Oh.
H
Yeah,
in
fact,
it's
required
for
cni.
You
have
to
specify
the
network
name,
space
and
Linux,
which
is
implied
here.
We
all
because
pods
already
do
it
and
if
you
don't
specify
the
interface
name,
you
get
eat
Zero
by
default.
So,
okay,
so
this
is
you
have
so
well.
This
calls
them
Net
One
net,
two
Etc.
H
H
H
H
You're
doing
if
you're
doing
an
ietf
compliant,
everyone
in
the
world
knows
what
you're
talking
about.
Yes,
if
you're
doing
something,
that's
sort
of
like
ice
IPv6,
light
and
you're
going
to
specify
things
your
own
way,
then
we
may
need
and
then
I
agree,
we'll
need
to
specify
things
your
own
way,
but
don't
say
it's
compliant.
A
A
But
I
think
we
are
on
the
same
page
and
what
I
want
to
touch
on
all
this
from
all
this
discussion
is
and
and
and
I
think
we
have
Michael
here,
Michael
Zappa,
you
can
tell
me,
can
can
I
say
that
can
I
say
that
the
cni,
what
what
Mike
Cambria
is
saying
that
support
for
having
the
name
taken
as
conflict
and
basically
having
provider
some
support?
Can
that
be?
Can
I
say
that
that
is
one
of
the
implementations.
C
A
H
G
A
A
H
Me
I,
don't
I
won't
speak
behind
me,
but
you're
never
gonna.
This
is
the
issue.
You're,
never
gonna
get
everything
that
will
be
needed
with
CMI
cni
implementation
will
so
essentially
you're
trying
to
bless
the
Caps
trying
to
bless
certain
Network
implementations
by
making
them
first-class
citizens,
and
the
rest
of
us
are
going
to
be.
A
Okay
can
so
maybe
maybe
tell
you
this?
What
if
we,
because
right
now
we
are
discussing,
we
are
just
having
two
two.
We
already
know
that
said
on
the
interface
name
is
a
valid
field.
So
that's
something
that
cni
does
support
in
the
conflicts
right,
so
you
can
pass
that
so
that's
good,
so
the
I
think
the
the
the
first.
The
other
thing
is
the
the
primary
field
can
that
be
supported
by
by
the
conflicts.
Probably
could
right.
H
Don't
need
it
for
V6
at
all
and
you
if
you
use
host
local
or
some
of
the
other
cni
ipms
or
you
write
your
own.
Of
course.
That's
always
seems
to
be
missing.
You
can
always
write
your
own
that
will
do
the
same.
You
know
the
essentially
pointed
default
route
to
a
specific
to
egress,
a
certain
interface.
In
fact,
I'll
log
you
I,
want
Prime
mineral
multiple
for
V4
because
of
ecmp.
H
H
Certain
things
certain
things
we're
going
to
go
to
we'll
go
there
and
others
we
won't
I'll
argue
we
don't
need
this
at
all.
So
you
want
to
compromise
if
you
could
put
primary
change
primary
to
the
default
Road
and
let
us
put
it
on
multiple
Network
attachments
and
then
we're
getting
closer
to
agreement.
A
H
No
no
ecmp
I
want
them
to
be
exactly
the
same.
I
want
to
get
to
the
API
server
over
multiple
interfaces
because
I
live
in
a
real
redundant
world
and
if
somebody
decides
to
take
a
backhoe
and
remove
my
attachment
to
one
tour,
everything
survives
over
the
other
and
other
than
a
slight
latency.
Spike
user
won't
even
know
about
it.
A
H
A
So
this
is
where,
where
I
think
the
next
thing
that
I
wanted
to
get
to,
finally,
is
the
the
thing
that
I
added,
because
this
is
what
Haney
was
asking
for
so
and
that's
why
I'm
I'm
I
I
am
leading
towards
that
that
was
the
dra
piece,
this
guy
network
interface.
So
what
we
could
have
is
this
guy,
that's
something
that
I
was
I
was
proposing
in
my
initial
draft
from
last
year.
A
But
what
what
this
guy
means
is
that
this
object
would
be
a
description
of
the
whole
network,
but
of
the
single
interface
into
a
pod,
all
right
and
and
let's
Let
It
sing,
because
this
is
important
it's
it
cannot
be
referenced
multiple
times
it
can
be
referenced
only
once
it
cannot
be
used
in
deployment
sets
or
any
replica
sets.
Unless
it's
a
repeat,
the
number
is
just
one:
the
number
of
replicas
equals
one
because
it
cannot
be
replicated.
It
can
be
used
multiple
times
right.
So
that's
what
this
would.
The
guy
would
mean.
A
A
This
would
mean
that
maybe
this
reference
here
would
would
have
another
field
here
and
it's
similar
to
what
I
think
in
resource
claims.
We
saw
Source
right,
so
this
will
be
pod
network
name
or
pod
network
interface
name,
and
you
can
reference
one
or
the
other.
Only
right,
you
cannot
reference
both.
You
cannot.
You
have
to
have
at
least
one
of
them
and
they
are
not,
and
then
they
are
optional
right,
but
there
is
a
validation
to
make
sure
that
at
least
at
least
one
is
referenced
right.
A
So
I
can
reference
this
guy
or
this
one.
So
this
is
where
there
is
very
similar
to
what
the
the
dra
is
doing.
The
the
big
difference
here
is
this
single
resource
claim
can
be
referenced
multiple
times
and
I
have
Patrick
here
to
correct
me
if
I'm
wrong,
but
I
think
that's
the
case
right
here.
Network
interface
cannot
be
referenced
multiple
times
because
we
cannot
have.
A
Let's
say
if
that
field,
one
of
the
fields
I'm
proposing
that
it
has
is
Mac
address.
We
cannot
have
multiple
multiple
interfaces
having
the
same
Mac
address
in
the
cluster,
because,
though
possible,
if
they
are
in
separate
separate,
let's
say
physical
L1
layers-
it's
still
possible,
but
maybe,
but
at
least
can
kind
of
that
will
be
kind
of
the
issues
you
will
have
to
deal
with.
If
you
were
to
do
that
right.
A
What
I'm
proposing
here
is
this
will
have
only
like
free
Fields
right
now,
and
this
doesn't
fully
solve
what
Jaime
is
asking
was
is
asking
for
initially,
but
maybe
this
can
be
the
gate
to
that
right,
so
bear
with
me
initially.
This
would
have
things
like
which
net
which
network
I
am
from
because
it
has
to
be
in
this
guy
is
per
Network.
So
it's
multiple
of
multiple
of
network
interfaces
pointing
to
a
single
Network
right.
So
that's
a
given
same
like
pod
pod
referencing,
multiple
pods
reference
same
pod
Network.
A
This
is
just
the
intermediary
between
the
two
right
that
can
provide
additional
parameters
to
attaching
a
this
in
this
network
to
a
specific
pod
right
right
now.
It
has
just
things
like
IP
addresses
in
a
spec
you,
you
would
Define
what
sort
of
things
you
want
to
do
and
then
the
status
Pro
kind
of
has
the
additional
meaning
queue,
because
that's
very
important,
because
this
object
has
could
the
kind
of
serve
as
an
additional
meaning
in
the
whole
kind
of
multi-networking
word.
And
what
I
mean
by
that
is.
A
It
could
provide
the
information
that
we
could
today
we
lack
like,
for
example,
most
of
you
probably
is
familiar
with
maltus
maltus.
All
the
information
about
whatevers
has
been
set
in
the
Pod
puts
as
an
annotation
in
a
in
the
thoughts
and
annotation
right.
You
have
the
whole
network
status.
I.
Don't
know
that
the
title
of
the
key,
but
basically
there
is
a
network
status
and
it
contains
all
the
information
about
each
interface
that
it
added
right,
so
name
Mac,
address
IP.
A
If
and
maybe
some
other
parameters
if
there
are,
but
basically
by
the
way,
this
annotation
is
the
kind
of
the
the
main
kind
of
inspiration
that
this
object
got
created
because
of
that
annotation
I.
We
don't
like
I,
don't
like
annotation,
so
everything
has
to
be
properly
structuralized
from
my
point
of
view.
So
this
is
where,
where
this
is
the
inspiration
for
that
right,
so
basically,
this
object
can
hold
all
the
information
wherever
the
implementation
set
inside
the
pod
right.
A
It
will
boil
down
to
that
implementation
to
create
this
object
or
update
it,
because
I
don't
think
we
can
rely
on
cubelet
to
do
that.
Maybe
but
I'm
not
sure
whether
cubelet
could
handle
kind
of
all
that,
and
there
is
a
lot
of
question
around
the
whole
thing
right.
So
that's
what
kind
of
the
kind
of
the
the
gist
of
this
object
is
a
Tomo
you
have
had
please
go
ahead.
E
A
With
you
Tomo
thanks
for
for
keeping
your
time,
let's
continue
on
the
on
the
discussion
on
this
one
next
week
and
please
read
through
that.
There
is
some
description.
There
is
not
much
to
it
but
familiarize
with
yourself
with
this,
and
let
me
explain
it
a
bit
more
next
week
and
let's
call
it
it
for
today.
Okay,
thanks,
everyone
hear
from
you
next
week.