►
From YouTube: Multi-Network community sync for 20230412
Description
Multi-Network community sync for 20230412
A
All
right
welcome
everyone
to
the
multi-network
Community
sync
today
is
April
12th
and
we're
going
to
start
with
Philip.
He
wants
to
kind
of
talk
about
how
we
can
integrate
with
gate
API
some
ideas
and
let's
talk
about
kubecon.
That's
thanks
to
Antonio
for
reminding
that
let's
touch
on
who's
gonna
be
there.
Should
we
cancel
next
week
meeting,
probably
we
should
and,
lastly,
we're
just
going
to
continue
the
discussion
on
our
cap.
So
Philip
floor
is
yours.
Do
you
want
to
share
something
or
just
want
to
discuss.
B
Let
me
give
a
little
background.
So
I've
been
working
on
a
product
that
is
really
focused
on
a
service
provider
and
launching
5G
networks,
and
when
you're
working
in
service
provider
networks,
they
have
very
complex
Networks.
B
B
This
can
happen,
of
course,
in
a
Enterprise
where
maybe
you
have
a
secure,
back-end,
Network
and
a
network
can
reach
the
internet
or
something,
but
in
service
providers.
You
always
have
to
integrate
with
a
radio,
Access
Network
and
operation
Administration
and
maintenance.
B
Network
networks
for
the
network
functions
to
communicate
to
each
other
Etc,
and
so
you
have
to
put
kubernetes
into
this
sort
of
mix
and
what
a
lot
of
folks
have
been
doing
has
been
to
ask
for
multis
interfaces
and
what
they
do
with
maltes
is
to
do
like
a
direct
srav
access
to
the
outside
world,
and
each
pod
basically
connects
to
external
networks
and
has
to
handle
its
own
sort
of
interaction
with
the
routing
firewalls
Etc
and
it
what
it
results
in
is
sort
of
spilling
the
complexity
from
what
should
be
contained
and
orchestrated
inside
of
kubernetes.
B
It
spills
that
complexity
out
and
and
so
what
we've
done
is
we've
We've
made
a
product
where
you
know
sort
of
everything
can
come
through
our
product,
so
we
can
connect
with
multiple
networks.
We
can
do
multiple
protocols.
All
of
the
reasons
why
people
ask
for
direct
network
access
and
I
was
planning
so
in
the
Gateway
API,
we're
very
interested
in
using
the
Gateway
API
there's
a
model
with
different
layers
with
the
Gateway
class
gateways
and
routes
that
come
off
of
those
gateways
and
I.
B
We
are
getting
ready
to
suggest.
Essentially,
this
model,
where
we
would
configure
when
a
Gateway
class
is
installed.
You
could
say
you
know,
let's
install
the
F5
Gateway
class
or
the
Kong
Gateway
class
or
the
nginx
Gateway
class,
and
at
that
stage
you
could
configure
multiple
networks
and
then
each
Gateway
would
essentially
be
a
gateway
to
a
single
Network
right.
B
So
if
you
wanted
to
have
a
gateway
to
your
radio,
Access
Network,
you
would
allow
only
the
functions
that
are
allowed
to
talk
to
your
radio
Access
Network,
to
get
routes
on
that
Gateway
and
that's
true
for
egress
routes,
which
is
a
new
thing
that
we're
going
to
try
we're
going
to
submit
soon
about,
but
also
Ingress
routes.
B
You
know.
Multiple
networks
within
kubernetes
and
I
am
a
little
concerned.
That
I
think
that
that
is
great
I'm,
a
little
concerned
that
that
the
pattern
will
still
end
up
being
something
like
this,
where
it
just
allows
for
a
more
sort
of
kubernetes
way
to
be
not
Cloud
native,
and
that
a
good
pattern
would
be
to
to
have
the
access
to
these
networks
be
through
the
gateways.
And
so
it
would
be
good
to
figure
out
how
we
can
coordinate.
B
You
know
how
we
configure
the
networks
at
the
Gateway
class
level
with
the
modeling
you
guys
are
doing
for
for
how
you
model
multiple
Networks.
Did
that
make
sense
or
did
I
just
ramble.
A
I
I
think
it
it
it
does.
But
Antonio
you
have
a
hand.
C
Yeah
I
have
a
question
when
you
can
go
to
the
previous
slide,
where
you
saw
the
services
that
I
want
to
understand
the
model,
because
that
service
is
a
kubernetes
service.
C
B
Yeah
so
with
the
Gateway,
though
you
could
you
could
model
this
as
a
service.
B
So
so,
instead
of
saying
you
know,
I
need
to
connect
directly
to
the
outside
world,
because
I
need
to
connect
to
the
Rand
Network.
You
could
create
a
service
that
gets
a
route
through
a
gateway
to
the
ram
Network,
and
it's
rather
than
just
doing
the
direct
connection
to
the
outside
world.
B
It
could
be
L3,
it
could
be
L7
connections
to
the
pods
right
now.
You.
E
C
D
When
you
say
service,
it's
an
5G
core
or
is
it
the
stuff
that
runs
on
the
end?
That's
all
we
put
on
top
of
the
FRG
core.
B
No
this
this
is
like
a
service
for
them,
so
each
Network
function
would
have
services
for
different
interfaces,
and-
and
so
you-
you
know,
if
you
had
I-
don't
want
to
get
into
the
protocols
and
specifics
too
much.
But
but
a
network
function
would
have
a
whole
series
of
services
for
connecting
to
other
network
functions.
B
B
So
this
is
so
the
first
you
know,
5G
stuff
going
into
kubernetes
is
control
plane,
so
service
chaining
is
oh
kind
of
a
slightly
different.
D
B
Yes,
okay,
so
so,
like
an
AMF,
for
example,
would
have
it's
probably
it's
service-based
interfaces
coming
in
through
a
standard,
kubernetes
service,
but.
E
B
Ngap
interfaces
going
out
toward
the
Ram,
they
would
tend
to
do
this
where
they
get
direct
access.
Yeah.
B
Plan,
yeah
and-
and
it's
not
even
just
completely
standard
interfaces-
I
mean
you
know
there.
There
are
a
lot
of.
There
are
a
lot
of
interfaces
that
Network
functions
can
have
I
mean
I,
mean
the
Legacy
functions
for
of
data
replication,
all
kinds
of
things
again.
A
B
I
I
I
I'm
sort
of
talking
at
a
high
level,
though
in
this
group
about
so
figure
out
more
about
how
to
connect
to
the
the
networks
than
the
protocols.
B
A
Philip,
what
I'm,
if
it's
just
as
you
can
see
the
pulp
network,
is
a
very
high
level
here.
We
don't
it's
a
just
representation
of
a
network.
What
it
is
right
once
you
be
able
to
just
model
it
to.
However,
you
want
to
implement
this
one
that
work,
because
I'm
saying
that,
like
our
phase
three,
if
you're
not
mistaken,
is
to
integrate
this
with
Services,
of
course,
where
possible
right,
so
we
would
have
ability
to
expose
services
on
those
additional
interfaces
and
I.
A
As
you
say
yourself
said,
the
L2
networks,
the
SRO
Vista
data
plane,
still
has
to
be
this.
What
we
are
doing
right,
you
still
will
need
this,
so
so
your
pod
will
have
to
connect
to
multiple
interfaces.
What
we
are
doing
here
is
giving
you
additional
kind
of
means
of
exposing
interfaces
to
the
Pod,
and
it's
never
said
I.
Don't
think
we're
saying
that
anywhere
in
the
in
this
design
that
we
are
creating
that
a
pod
network
has
to
be
a
separate
physical,
like
a
VF
or
whatever,
whatever
Linux
interface,.
A
It's
completely
up
to
you
and
it
can
be
just
a
logical,
L3
networking
words
only
IPS
on
the
same
physical
interface
and,
of
course
your
implementation
has
to
handle
it
inside
out.
But
then
it's
just
up
to
you
that
the
only
thing
would
be
that
I'm
not
sure
whether
that's
that's,
that
would
be
a
goal
for
you
or
not.
If,
if
you
want
to
use
a
single
IP
for
all
those.
D
I
think
we're
talking
differently,
so
so
we
are
offering
L2
and
L3
connectivity
yeah
of
the
pods,
but
I
think
what
what
Philip
is
trying
to
do
is
it's
in
reality
to
see
if
it
can
use
La
for
a
lot
of,
say,
layer,
7,
construct
an
API
server
to
connect
the
pods
instead
of
do
the
call
it
use.
You
say
physical,
you
say:
Network
orchestrator
to
set
up
the
networks
and
then
mapping
the
networks
into
kubernetes,
so
they
become
visible
for
the
pods
and
so
on.
D
So
you
can
attach
networks
to
them
when
I
say
this
is
another
approach
of
how
to
do
sort
of
5G
and
started
using
other
constructs
to
address
the
the
services.
I
think
this
is
more.
B
I
have
to
say
my
connection
must
be
bad
because
you
guys
are
kind
of
going
in
and
out
a
little
bit,
but
I
mean
I
the
thing
about
pod
networks
and
connecting
the
services
from
what
you're
talking
about
like
I
I'm
focused
on
the
traffic
that
goes
in
and
out
of
kubernetes
sure
traffic
within
kubernetes
I.
Don't
care
about
right
and
the
Gateway
API
is
about
traffic,
that's
going
in
and
out
of
kubernetes.
D
This
will
still
go
over
the
to
call
it
the
cluster
Network
right.
Anything
you
push
in
here
is
going
to
go
over
the
cluster
Network,
the
what
we
will
call
the
default
pod
network.
B
D
But
sort
of
if
you
talk
Gateway,
but
what
you've
shown
before
was
multi-networking
right.
When
you
have
a
number
of
networks
that
you
attach
as
secondary
networks
to
pods
using
maltes,
two
very
different
use
cases.
A
A
I
I
would
see
the
same
thing
by
the
way,
because
I
would
love
because
I
I
I
we
are
hearing
in
the
width
of
that
Gateway
API
as
well,
and
I
would
love
to
have
eventually,
as
a
next
step
after
services
and
network
policies
to
Gateway
support
multi-networking
as
well.
Right
and
like
is
that
the
concept
of
this
like
select
single
Network,
so
your
gateway
is
on
a
specific
network.
Is
that
yeah.
B
So,
essentially,
you
would
instantiate
a
Gateway
that
would
be
your
gateway
to
a
particular
Network,
yeah
and
then
different.
You
know,
different
namespaces
would
have
the
ability
to
either
request
routes
from
it
or
not,
right
and
so
again
for
your
radio,
Access
Network,
like
an
ANF,
would
be
allowed
to
request
routes,
whereas
an
SMS.
F
B
Route
well
as
in
the
routes
in
the
Gateway
API,
so
there
are
okay,
TCP,
UDP
and
then
L7
routes
and
we're
introducing
the
concept
of
an
egress
route,
and
this
is
where
it
becomes
really
important
right,
because
again,
you
wanna
historically
right.
One
of
the
problems
with
with
kubernetes
from
my
perspective
is
that
you
know
traffic
in
ingressing
and
traffic
out.
B
Egressing
are
controlled,
two
different
ways
and
have
you
know
different
pathing
and-
and
you
know,
controls
so
by
by
putting
it
through
a
Gateway
so
that
you
can
have
both
your
egressing
and
your
ingressing
path
come
in.
You
know
you
can
have
more
control
over
it,
so
the
cluster
operator
and
so
forth.
The
Telco
can
have
more
control.
F
Yeah
I
want
to
point
out
that
I
have
in
the
Sig
networks,
meetings
pushed
for
a
more
like
wider
use
of
services,
but
they
have
pushed
back
and
said
that
the
gateways
are
the
future
of
the
incoming
traffic
to
the
kubernetes
cluster.
So
I
believe
this
is
precisely
the
right
way
to
go
and
basically
for
Ingress
traffic
incoming
traffic
to
the
cluster.
We
should
focus
on
the
Gateway
API
foreign.
A
Yeah
I
think
keep
in
mind.
That's
something
that
we
have
on
our
roadmap
to
to
kind
of,
tackle
and
and
I
put
it
as
a
service
support
for
this.
And
if
we
decide
that
no,
let's
not
touch
service
anymore
and
let's
just
completely
use
gateways
for
multi-networking
perfectly
fine.
But
let's
discuss
that
at
the
time.
Antonio.
C
Yeah,
the
problem
with
Services
is
that
services
start
with
one
problem
and
right
now
it's
serving
more
problems
than
expected
and
is
an
API,
that's
inconsistent.
So
that's
why
Gateway
is
a
better
fit
to
this
new
development
because
first,
is
is
a
is
thought
for
doing
what
we
want
to
do:
select
and
import
and
select
some
back
ends
and
Brew
the
traffic.
To
that
you
know.
Services
selling
points
is
an
API
that
work
together
so.
B
Nope,
that's
it
I
just
wanted
to
raise
the
question
and
see
you
know
if
we
should
have
further
discussions
or
start.
You
know,
I
I,
think
regular
communication
or
something
yeah.
A
We
I'm
seeing
this
that
I'm
hoping
that
I'm
seeing
this
as
we
are
aligned,
considering
what
you
want
to
do
and
in
our
phase
three,
if
you
look
at
our
cap,
the
current
for
the
requirements
for
multi-networking
basically
talks
about
integration
with
with
Services
an
external
additional
Services
right
and
basically
we
can
include
Gateway.
Currently
we
I'm
mentioning
only
services
and
network
policies
and,
as
I
said,
maybe
we
doesn't
have
to
be
exactly
Services.
It
can
be
through
Gateway,
which
we,
where
we
can
achieve
the
same
thing.
A
Definitely
then
that's
the
is
way
out
at
least
one
year
off
from
what
we
want
to
when
we
want
to
talk
about
this,
so
we
are
very
early
on
on
the
spot
networks
journey
to
kind
of
currently
even
start
talking
about
integration
with
other
services,
all
right,
so
that's
where
we
at
and
with,
if
only
we
want
to,
but
not
now.
Now
we
want
to
kind
of
touch
on
the
basics
and
if
you
want
to
be
part
of
that,
that's
here,
you're
welcome,
but
I.
A
A
And
just
what's,
I'm
not
sure
whether
you
want
to
be
part
of
this
discussion
or
not
how
much
your
time
you
have
but
I
would
I
I
would
do
appreciate
if
you
would
participate
and
kind
of
keep
us
in
check
in
in
whatever
we
decide
here
in
in
the
beginning,
so
that
we
can
address
your
stuff
in
the
future
right.
So
definitely
that's.
B
Gonna
put
in
a
gap
for
egress,
and
one
of
the
things
I'm
going
to
put
in
is
specifically
for
multiple
Networks,
and
so
maybe
you
know,
maybe
we'll
just
have
to
do
it
in
a
way
that
it
can.
We
can
Circle
back
when
you
get
to
stage
three
or
something
and
figure
out
how
to
align
them.
F
B
Yeah
I'll
I'll
I'll
paste
some
links
in
the
agenda
yeah.
C
A
So
thanks
Philip
for
the
for
the
links
and,
if
anything,
just
copy
them
into
the
chat,
meeting,
chat
and
I
can
I
can
paste
them
here
or
you
should
have
access
edit
access
here
as
well.
Next
topic
cubecon
next
week,
should
we
cancel
next
week?
A
Or
maybe
I
will
ask
otherwise
it's
something
to
Antonio
initially
talked
asked:
is
anyone
going
to
be
on
site
that
wants
to
meet
in
person?
I'm
I'm
I
will
not
be
there,
I'm,
not
sure
anyone
else
going
to
be
on
site
and
willing
to
drive
the
multi-network
in
person.
We
visit
Amsterdam
no.
A
So
I
think
I
I
would
so
maybe
let's
do
this.
Let's
cancel
it
anyway.
Next
week
we
will
cancel
for
sake
of
even
attending
the
any
presentation
and
and
then
we
can
continue
tweaks.
A
Okay,
any
other
comments
on
the
kubecon.
A
No
all
right
all
right.
Let's,
let's
continue
any
other
topics
before
I
can
start
kick
off.
The
discussion
on
cap
I
think
we
missed
that
last
week.
Any
other
topics
I
want
to
bring
up.
A
G
B
G
Hard
to
look
for
maybe
we
should
put
the
all
the
information
into
the
one
resource.
I
think
the
Google
Doc
seems
to
be
the
one
yeah.
A
Let
me
just
thank
you
for
reminding
that
Tom
yeah
I
have
committed
this.
A
So
please
take
a
look.
This
is
my
personal
branch
and
this
is
a
KK.
By
the
way,
this
is
a
kubernetes
repository,
Fork
I'm,
not
sure
whether
this
is
helpful
to
import
things.
If
you
want
to
build
some
client
controller
around
this.
This
is
the
initial
and
the
initial
version
of
this
of
at
least
what
what
was
valid
back
then
so
look
at
it.
There
is
a
lot
of
code
changes
because
of
a
generation.
I
just
run
mic
update
and
it
just
generated
a
bunch
of
changes.
Look
at
the
comments.
A
It's
split,
I
I
managed
to
split
the
changes
that
I
did
step
by
step
so
that
you
can
see
only
the
changes
that
I
did
so.
The
first
comment
is
just
adding
the
the
types
then.
The
other
comments
is
some
validation,
some
storage
stuff,
and
the
last
comment
is
about
like
the
whole
generated
code,
so
I
just
added
the
type
and
just
Auto
generated
the
code,
so
that
was
the
only
thing.
I
tried
with
this
I'm,
not
sure
I
haven't
started
even
the
controller,
so
I'm
not
sure
how
valid
that
part
is.
A
But
this
is
a
start,
and
if
you
want
to
do
something
more
about
it,
create
PRS
against
my
Branch,
you
I
think
you
should
be
able
to
do
that.
If
you
want
to
change
something
over
there
as
we
go
with
that
cap,
I
will
try
to
keep
updating
the
the
type
the
the
the
object
as
we
go,
but
this
is
something
that
you
can
start
off
of
and
do
something
about
this.
If
you
need
to
I
will
add
this
as
well
a
good
point
to
okay.
A
Not
sure
how
useful
this
is,
but
hopefully
someone
can
build
some
some
some
demo
controller
against
that
all
right,
let
me
switch
to
now
discussion
around
the
cap.
Is
that
the
one?
Yes,
that's,
the
one
I
should
stop.
A
Had
a
discussion
with
some
folks
and
I
have
a
proposal
for
this,
so
where
we
ended
last
week
last
week
we
ended
how
do
I
create
the
default
pod
network
based
on
current
parameters
or
on
current
existing
stuff
in
the
cluster,
so
the
case
would
be
when
I
create
a
cluster
and
then
the
other
case
would
be
when
I'm
just
migrating
from
a
version
X
to
Version
Y,
where
y
has
pod
Network
right
in
it.
So
that
was
the
problem.
What
we
were
kind
of
facing
so
to
address
this.
A
What
I
I
want
to
propose
and
and
bear
with
me
and
and
see
what
how
that
would
work
so
first
of
all,
is
I
want
to
slightly
change
how
we
are
behaving
the
ipam
of
fields.
A
Last
week
or
till
last
week
the
iPhone
4
had
kind
of
three
options:
it
had
a
external
kubernetes
or
not
specified
the
last
not
specified
meant
that
I,
don't
I,
don't
want
to
have
any
IPS.
So
what
I
want
to
change
this
to
a
slightly
different
I.
D
A
Yeah,
so
what
does
it
mean
when
it's
not
specified
when
it
was
not
specified?
It
meant
that
kubernetes
will
not
report
that
IP
for
you
or
will.
D
Not
require
that
IP
for
you,
so
you
will
not
get
an
end
point
exactly.
A
You
will
not
get
an
endpoint
then,
so
what
I
want
to
do
is
slightly
change.
That
I
want
to
introduce
a
explicit
flag
for
the
not
specified.
So
basically
we
would
have
external
kubernetes
and
none
that
will
represent
this.
So
basically
none.
Probably
the
name
has
to
be
changed,
but
still
this
field
would
represent
exactly
that
third
case
when
we
had
last
week.
So
if
you
want,
if
you
don't
care
about-
and
you
want
to
explicitly
call
that
out,
I
don't
want
to
have
any
IPS
on
this
interface.
A
A
So,
if
I
don't
specify
it,
you
still
is
not
expected,
is
undefined
what
I'm
I'm
trying
to
call
it
out
the
configuration
is
undefined,
but
why
I
want
to
do
this
and
I
will
I
will
give
me
a
second
I
will
explain
why
the
other
aspect,
if
this
is
not
is
is
not
specified,
is
I
would
like
those
two
fields
to
behave
like
the
pot
Siders
on
node.
What
does
it
mean?
Is
that
how
do
that?
A
How
those
fields
behave
in
in
now
today
is
that
they
can
be
mutated
Once
in
a
Lifetime
right.
So
basically,
if
I
do
not
have
anything
set
on
that
field,
I
can
set
it.
So
that's
something
that
I
would
like
to
change
in
in
behavior
of
this
object,
and
this
will
allow
us
to
fully
support
what
I'm
thinking
the
way
to
migrate
and
introduce
this
object
to
existing
and
and
to
the
existing
clusters.
A
G
Yeah
I'm
I'm
still
unclear
that
the
difference
between
the
noun
and
then
the
no
icon.
A
It
still
behaves
like
none
where
I
sorry,
I
I
will
not
specify
that,
but
over
time,
if
I
want
to
change
that,
I
had
that
that
ability-
and
basically
that
leads
me-
let
me
let
me
show
you
an
example
of
how
would
that
behave,
because
what
I
want
to
do
next
is
the
defaultpod
network
would
be
able
to
be
created
in
two
ways
in
automatic
way
or
in
a
manual
way.
A
So,
first,
let's,
let's
go
on
the
market,
maybe
go
through
the
automatic
way,
because
what
does
automatic
may
win
way
means
basically
what
it
will
mean
that
a
KCM
will
be
the
one
that
will
create
the
default
pod
Network,
that's
what
I
envisioned
there
will
be
a
controller
in
it
that
will
handle
this
whole
thing.
A
We're
gonna
have
to
have
a
controller
for
pod
Network
anyway,
so
I
think
it
should
land
in
KCM,
but
that's
something
to
we
can
discuss,
but
right
now,
I'm,
I'm
thinking
of
KCM
and
for
the
sake
of
discussion,
let's,
let's
say
it's
KCM
and
we
need
to
rely
on
existing
flags
that
KCM
has
today
right
and
those
flags
are
allocate
notesiders
and
I'm,
referring
here
to
existing
flags
that
that
KCM
has
and
the
other
flag
is
cluster
side.
Those
are
the
two
flags
and
what
do
they
do?
A
The
first
one
identifies
whether
I
use
the
built-in
ipam
controller
in
KCM,
okay
and
then
the
other
flag
is
used
to
say
what
IP
ciders.
This
iPhone
control
has
to
leverage
right.
So
basically,
this
guy
as
well
identifies
whether
I
do
a
single
stack
or
dual
stack
all
right.
Those
are
the
existing
kind
of
behavior
of
those
two
flags.
Tomorrow
you
have
a
hand
or
or
that's
from.
C
Okay,
so
this
topic
is
hot
these
days
with
a
with
a
cluster
cider
and
the
and
the
no
type,
and
so
to
give
some
content.
There
is
some
historical
reasons
why
kubernetes
has
an
iPhone
controller
on
the
on
the
controller
manager,
and
this
adds
the
the
what's
the
name:
no
respectful
sided
people,
so
they
implementation
can
consume
that
field.
But
the
discussions
from
the
last
day
is
that
this
is
an.
C
A
C
C
A
A
So
no,
but
then
okay
I
hear
you
so
then
that
has
to
be
standardized
right
and
then
this
can
standardize
this
right,
because
and
and
hopefully
I
can
handle
all
the
other
cases,
but
you're
right.
This
is
this
is
something
that
might
be
tricky.
So
let
me
maybe
finish
on
what
I'm
thinking
that
I
hear
you.
So
basically,
what
you're
saying
is
that
post
either
can
be
set,
but
the
whole
ipam
is
done
by
someone
else,
but
in
that
use
case
Antonio,
who
is
setting
the
Pod
cider?
A
Is
it
still
I
pump
controller
KCM
or
is
it
done
by
or
is
it
done
by
external
as
well.
F
C
A
Yeah
and
that's
perfectly
fine
because,
as
you
said,
those
are
fields
that
you
can
your
implementation
can
leverage
and
that
then,
that
we
just
enable
who
is
doing
that
right
is
either
doing
it.
Is
it
all
fully
owned
by
ipam
controller
or
not
right,
whether
I
mean
the
fields
are
there
for,
if
I'm
not
using
it,
then
your
implementation
can
do
whatever
we
did,
and
this
is
what
what
I'm
saying
here
as
well
right.
So
let
me
go
go
through
the
examples
right
so
based
on
those
two
fields
in
KCM.
A
This
is
how
the
the
default
pod
Network
in
automatic
way,
will
be
created.
So
basically,
this
will
be
used
for,
like
migration
cases
or
I,
create
a
cluster
and
I.
Don't
care
about
multi-networking
right.
So,
if
I
have
this
field
set,
this
field
set
will
ensure
that
the
ipam,
4
and
6
will
get
a
kubernetes
value,
one
of
them,
and
then
the
cluster
cider
field
will
Define,
which
of
them
I
set
and
basically
which
of
them.
A
I
said
right,
because
cluster
cider
defines
whether
I
do
dual
stack
or
single
stack
so
and
here
I
have
I
have
here
an
exact
examples
right
so
allocate
cider
is
true,
and
then
my
my
cluster
cider
is
a
V4
and
V6.
So
I
would
create
a
pod
Network,
which
is
default
with
with
both
set
as
kubernetes
right
then.
A
The
other
case
is
I,
have
again
a
true
allocate,
but
just
with
a
single
stack,
so
I
will
set
kubernetes
for
one,
but
none
for
the
other,
because
I
don't
want
any
IPS
on
the
other
on
the
other
family
stack
right
and
then
last
example,
where
I
have
allocated
as
a
false
means:
I,
don't
I
I'm,
not
doing
anything
support
network
when
it's
in
automated
way
it
will
be
created
in
in
this
spec
in
this
way
where
a
spec
is
empty,
what
it
can
happen
next
right
now
nobody's
expecting
anything.
A
C
Yeah,
the
one
is
through
the
the
the
the
flow
I
mean.
What
what
is
first,
what
is
that
so
I'm
I,
an
operator
I,
want
to
configure
my
question:
what
do
I
configure?
First,
the
controller
manager
flights.
A
B
G
A
And
then,
and
and
basically
what
I
want
to
say,
the
kind
of
put
more
time
on
this
case
right.
This
is
what
I
was
referring.
This
is
what
Antonio
was
referring
to
the
node
spec
pod
cider
field,
when
it's
not
set-
and
this
is
what
exactly
I
want
to
kind
of
enable
the
same
same
model
for
the
ipam
field
when
it's
not
set,
you
can
set
it
after
the
after
the
fact.
A
C
A
E
E
A
E
E
A
A
If
your
your
implementation
is
driven
by
cni
CRI
will
finally
support
pod
Network
and
then
it
has
to
be
changed
and
supported
in
such
a
way
today,
as
you
today,
let's
say:
I
do
I
deploy
this
today.
No,
neither
Cena
is
going
to
care
about
this
because
that's
the
point.
We
have
to
be
Backward
Compatible,
so
this
pod
Network
for
someone
who
doesn't
Implement
multi-networking,
will
just
be
there,
but
it
will
be
just
as
a
background.
A
They
don't
care
about
multi-networking
the
way
it
is
it's
fine
for
them
and
that's
where
it
ends
so
cni
still
works.
Cubelet
just
looks
at
okay,
the
default
code
network
is
there
but
cni
or
neither
nobody
is
looking
at
that
object,
and
it's
just
there.
So
you
will
have
just
one
additional
object
in
your
cluster
with
this
change
and
you
don't
care
about
it.
Cni
doesn't
change.
I
still
look
at
the
first.
A
E
A
E
A
G
Okay,
let
me
take
the
example
of
the
single
stack
ipv
for
realizations,
so
so
the
if
we
creating
the
key
bottom
the
cluster
and
then
install
it
in
the
cube
funnel
at
that
time,
the
cluster
cider
is
configured
I
mean
that
at
that
time
the
crust,
a
cider
and
then
the
the
frontal
same
cider
ranges,
and
at
that
time
the
when
the
airport
is
launched,
then
the
frontal
adding
is
the
allocated
the
Pod,
IP
and
then
assigned
to
be
a
physical
interface
notifications.
G
At
that
time,
so
they
are
as
the
the
definition
of
the
iPhone
4
and
the
iPhone
6
the
kubernetes
if
the
kubernetes
is
specified
at
that
time.
The
this
this
is
the
kubernetes
cares
about
life
location,
but
actually
the
in
the
disallocate,
the
unknown
side
that
you
crossed
the
cider
the
case
the
front
Network,
the
the
Primal
network
interface.
The
cni
will
do
that
not
not
the
kubernetes
actually
so.
A
A
Configuration
if
your
flannel
configuration
is
used
and
your
KCM
is
still
configured
as
such,
then
it
will
be
this
right
and
and
I
want
to
go
to
my
last,
maybe
just
before.
We
continue
because
I
want
to
create
a
another
option.
I
want
to
add
another
flag
to
KCM,
just
just
to
kind
of.
Maybe
that
will
address
what
you're
saying
Tomo
so
so.
G
A
Because
there's
no
other
way
to
do
this,
Tomo
I,
don't
think
there
is
other
way
to
okay.
G
G
So
everything
is
okay,
but
these
both
are
same
skin
I
mean
that
the
at-risk
I
don't
know
the
the
AWS
or
the
gcps,
the
kubernetes,
but
they're
almost
open
source
they
are
kubernetes.
Deployment
should
use
these
stuff.
One
of
the
stuff
I
mean
that
the
classical
side
is
specified
and
then
the
cni
plugin
is
configured
and
then
the
actual
IP
allocation
is
done
by
the,
not
the
kubernetes
sure.
A
G
A
Step
back
because
tomorrow,
you're
mixing
two
things
and
and
keep
in
mind
whatever
we
talked
about
this
ipam
field
is
about
who's.
Doing
maybe
maybe
this
is
a
wrong
wrong.
I
need
to
probably
rephrase
this
because
I
think
we
go.
We
go
on
what
I
want
to
say
a
bit.
Maybe
too
far,
maybe
we
need
to
redefine
what
does
this
I
need
to
spell
out
again?
What
does
this
field
means?
Iphone
4
doesn't
mean
that
kubernetes
is
going
to
do
your
parent
pod
item
management
because
that
doesn't
exist
today.
A
We
don't
have
that
today.
That's
that
doesn't
exist.
That's
iPhone
controller
that
I'm,
referring
to
in
kubernetes
what
it
will
do.
It
was
just
assigned
the
note,
dot,
spec,
dot,
pod
cider
field.
That's
it
that's
the
only
thing
it
does,
and
then
we
have
additional
objects.
That
Antonio
is
working
on
where
you
can
expand,
those
spotsiders
add
more
and
then
there
is
additional
things
to
configure
it.
That's
what
I
want
to
leverage-
and
this
is
the
field
only
for
that,
because
that's
the
only
thing
kubernetes
can
do
today.
A
A
This
field
is
only
to
identify
whether
I
use
the
kubernetes
item,
controller
or
or
not,
and
then
just
seed
it
all
off
to
whatever
external
stuff
you
can
have,
and
that
means
is
what
Antonio
was
referring
to
is
that
the
node.spec.pot
cider
field
is
as
well
set
by
you
and
you
leverage
it
or
you
don't
it's
up
to
you
in
final
case
I'm,
not
sure
whether
flannel
and
that's
a
good
example.
Let's
assume
I'm
not
sure
today,
flannel
can
even
read
that
field.
A
A
So
they
don't
even
look
at
the
node
pods
nodes
back
Port
cider,
so
those
even
will
not
bother
be
caring
about
those
fields,
because
flannel
will
not
even
Leverage
those
two
Fields
because
I
said
whatever
I
said
here,
nobody
is
going
to
care,
except
for,
of
course,
if
I
want
to
be
aligned
with
Q
proxy
at
that
point,
so
that's
a
different
story,
but
then
that's
good
boils
down
to
how
you
create
your
cluster,
but
then
whether
your
planner
is
going
to
use
those
fields
or
not.
That's
that's.
On
the
other
hand,.
D
F
D
I
have
any
other
network.
I
cannot
use
the
kubernetes
iPad
controller.
A
A
B
D
A
D
D
D
A
A
A
Yeah
yeah
sorry
I'm,
trying
to
make
it
step
by
step,
but.
D
A
Probably
bigger
picture
would
help
us.
How
were
we
getting
at
man?
That's
that's
a
fair
point
per
yes.
I
will
try
to
maybe
draw
something
out.
Yes
in
this
talk,
Antonio.
C
Yeah
it's
to
clarify
something,
the
way
that
most
people
is
using
the
node,
so
the
ipan
that
we're
talking
here
is
subnet
and
assigned
a
smaller
assignment
to
note
and
the
way
that
people
is
using
now
is
this:
subnet
is
cascaded
to
some
or
cubelet
or
container
D
is
generating
the
the
cni
conflict
by
from
this
sub.
This
is
the
90
of
the
implementations
that
there
is
indeed
that's
why
I'm
saying
both
of
you
were
right,
but
I
mean
just
keep
this
in
mind,
I'm,
going
to
we're
talking
iPhone.
C
This
is
iPhone
as
I
have
a
Sonic
and
I
have
a
smaller
Summits
per
month,
so
it's
assigning
and
submit
to
a
node
to
use
the
usually
the
course
look
alive.
A
That
I
kind
of
addressed
your
concern
here
did
I
change
your
mind
so
so
on.
What
really
this
ipam
field
means
it's
not
for
pair
pod
iPhone.
This
is
not
it
I'm
not
getting
there.
This
is
not
this
filters.
Just
keep
in
mind.
This
field
is
cluster
level.
It's
it's
cluster
level,
not
per
node,
so
I
cannot
do
with
this
one
field.
I
cannot
do
another
parent
pod
item.
That
kind
of
addresses
your
concern.
G
So
I'm
I'm,
really
returning
that
the
full
do
the
real
IP
assignment
to
the
pod
interface
in
case
of
the
kubernetes.
G
A
It
defines
two
things:
it
defines
that
the
IP
is
expected
on
the
specific
interface
and
then
defines
on
Define.
So
so
basically
that
means
that
your
implementation,
your
cni,
has
to
provide
those
IPS,
ipv4
or
6,
or
both
or
just
either
or
just
one,
and
then
additionally
can
I
leverage
the
node
for
my
per
nodes
either
and
that's
it
so.
A
Kubernetes,
basically,
I
can
Leverage
The
ipam
kcn.
So
you
need
to
look
at
the
tomorrow.
You
need
to
read
in
what
kcm's
iPhone
controller
do
today,
because
you
need
to
and
you
need
to
catch
up
on
that
one
looks
like
and
see
what
does
this
guy
do
and
because
a
built-in
mechanics
to
configure
ipam
and
based
on
KCM
and
exactly
what
I
said
here?
Please
read
on
what
kcm's
iPhone
controller
do
today
and
that's
the
only
thing,
and
basically
this
is
only
to
set
the
node.spec.pot
cider
field
and
that's
it.
So
that's
what
kubernetes
means.
A
Otherwise
it
doesn't
do
anything
else.
External
means
you
we
don't
even
do
that
and
it's
all
full
on
you
whatever
you
want
to
do-
and
this
is
what
Antonio
was
saying
before
on
this
field
can
be
said
by
you.
You
don't
even
have
to
care
about
that
IP
because
you
have
other
mechanisms
to
Define
that
range.
So
so
so
that's
the
case,
I
think
we
Antonio
we're
gonna.
Call
it
just
just
we're.
Gonna
probably
can
come
back
to
this
next
week.
G
Could
you
could
you
send
him
in
the
pointer
for
the
the
the
pointer
for
ipam
controller,
sure.
A
A
Know
it's
or
or
that
yeah
let's,
let's
we
can
kind
of
move
to
this
to
select
discussion
all
right,
I
think
then
it
will
be
make
more
sense
of
what
I'm
trying
to
do
here.
If
we
know
more
okay,
I
think
yeah,
that's
apologies
on
my
side.
That's
I
I
always
have
been
more
specific
on.
What
does
this
really
do
all
right
lead
through
try
to
read
through
this
with
the
understanding
of
what
those
ipam
controller
does
today
and
then
maybe
it
will
make
more
sense.
A
Then,
okay,
thanks
everyone
see
you
in
two
weeks.