►
From YouTube: Istio Networking WG meeting - 2018-07-19
Description
Agenda:
- Gateways and Gateway Types
- Remove Pilot’s sidecar
- Network operator’s guide
- Envoy leak fixed
- Prow issue
A
A
Yeah,
exactly
and
more
precisely
the
the
certificate
generation
and
loading
and
I
think
Osteen
will
give
us
a
better
overview
and
understanding
of
what
the
problem
is,
and
this
ties
also
a
bit
with
how
we
deployed
the
kubernetes
ingress.
So
there
are
the
two.
The
two
items
here
are
a
bit
related,
so
the
question
is:
if
we
deploy
kubernetes
ingress
by
default
or
not
and
may
be
costing,
you
want
to
talk
about
issues
with
the
certificate
managers
and
the
dependency
on
ingress
that
you
discovered
yeah.
B
I,
don't
know
if
everyone
has
context
here,
we
added.
We
definitely
want
to
have
some
way
for
people
to
be
able
to
use
that
man
at
me
protocol
to
get
certificates
automatically
and
to
do
that.
The
solutions
that
is
typically
used
and
seems
to
be
the
hormone
ease
is
a
certain
eager
package
which
is
implementation
of
the
protocol
and
is
able
to
get
certificates
automatically.
B
B
We
do
have
a
minute
work
before
it
worked
in
in
in
0-8.
It
worked
in
in
0-7
with
with
these
two
interests,
the
problem
I'm.
Having
is
that
right
now
it
doesn't
work
with
Gateway.
So
if
you
are
using
ingress
resources
and
the
ingress
gateway,
you
can
use
a
supply
bread
and
you
can
get
certificates
for
that
particular
domain.
B
B
Dms
challenge
again:
it
requires
that
the
user
has
DNA
control
over
the
entire
DNS
of
an
organisation.
So,
if
you
are,
if
you
are,
if
you
are
typically
people
who
test
our
or
who
developed
for
East,
you
are
not
that
we
know
of
the
entire
domain
and
they
do
not
have
control
over
the
entire
DNS.
I
mean
you
you
it.
You
know
it's.
A
B
All
put,
and
the
question
is,
if
you
use
a
DNS
shown,
is
it
that
means
a
DNS
admin
I
mean
who
controls
the
entire
data
center
organization
will
give
you
the
password
and
credentials
or
permission,
and
then
you
will
give
this.
You
possess
purchase
permission
into
the
into
the
set
manager
and
set
manager.
Will
that
point
have
access
the
entire
DNS
and
then
you
will
be
able
to
create
a
txt
record
respond
to
the
challenge.
It's.
C
Like
it's
like
the
that
there
has
to
be
a
DNS
entry
pointing
to
the
ingress
right,
so
there
has
to
be
something
and
the
same
guy
who
are
the
same
people
who
do
that
could
actually
configure
a
certain
manager
that
has
the
credentials,
because
it's
not
the
user
that
needs
to
have
them,
but
I
just
wanted
to
throw
in
that
there's.
Another
means
of
challenges
for
in
there,
I.
B
Think
for
most
enterprises,
that's
not
a
problem
because
they
would
just
not
use
that
manager
means
they
won't
have
a
not
means
that
when
use
will
buy
a
certificate
from
you
know
from
a
normal
provider,
that's
what
most
companies
are
doing
anyway.
So
it's
not
really
a
major
problems
that
it's
it's
just
for
easier
of
development
work
for
developers
who
are
you
know
it
won't
seem
to
go
to
the
administer
an
IP
address
in
in
the
DNS
server.
B
D
D
B
In
most
segment,
most
companies
that
that's
what
will
happen-
I
mean
hey
I,
don't
expect
any
large
enterprise
to
to
bother
and
to
enable
this
option,
but
for
small
developers
who
are
just
doing
the
proof
of
concept
of
doing
some
sort
of
you
know,
testing
or
whatever
it
is
common
I
mean
many.
Many
small
enterprises
are
relying
on
Acme
because
it's
a
free
certificate
and-
and
you
know
it's
very
easy
to
get-
you
don't
have
to
pay
for
it.
So
you
don't
have
to
to
go
through
tough
pain.
B
It's
not
only
the
money,
but
also
that
the
difficulty
in
the
steps
required
to
get
the
certificates
from
our
dorm
provider.
So
anyway,
that's
a
story.
I
mean
it's
it's
a
narrow
key.
Is
it
simplifying
for
my
second
particular
set
of
users?
We
have
a
solution
which
is
user,
ingress
controller,
give
the
certificate
and
then
switch
to
switch
to
the
virtual
service.
That's.
A
B
B
A
A
B
It
so
if
off
by
default
by
I,
don't
know
if
people
are
aware
right
now
we
ship
with
with
ingress
disabled
by
default.
So
as
a
user,
we
need
to
explicitly
say,
increase,
but
enable
so
they
will
do
that.
So
you
get
the
rule
balance
and
an
IP
address
for
the
legacy.
Ingress
was
a
kubernetes,
the
user
managers
they
get
a
certificate
certificate
is
in
a
secret,
everyone
is
happy,
and
then
they
switch
the
domain
to
veteran
service,
which
is
a
normal
process.
F
B
Using
the
ingress
is
not
a
problem
in
the
east,
your
ingress
they're,
still
using
history,
they're
still
getting
metrics
what
they
do
not
get.
This
is
traffic
split,
another
feature
that
they
give
to
well,
if
our
service-
even
that
is
not
a
problem,
because
because
many
policies
apply
again,
it's
a
choice.
I
mean
it's,
it's
not
really
the
end
of
the
world
if
they
keep
using
ingress.
Ok,.
A
B
A
B
There
is
a
patch
of
trees
and
their
patches
in
progress,
but
they
are
not
yet
up
streams.
Also,
so
we
know
we
can
get
it
with
gateway
working,
but
we
need
to
work
it
only
to
it
and
it's
not
going
to
happen
anymore,
zero.
The
side
effect
as
the
other
product
we
want
to
discuss
is
how
do
we
want
to
do
to
manage
this?
Transition
between
between
legacy,
ingress,
kubernetes,
ingress
and
gateway
I
mean
the
correct
mechanism.
Is
we
have
this
two
IP
addresses
and
user
start
with
immigrants
what
they
have
existing
interests.
B
B
B
A
B
A
B
A
B
B
We're
you,
you
know
the
other
option.
Again
we
discuss
which
we
demand
and
we
went
back
and
forth,
is
if
we
see
that
the
particular
domain
has
both
beautiful
service
and
English,
we
can
just
take
literal
service
and
give
it
priority.
I
know
I
believe
may
implement
it
by
time
and
it's
always
resource
tech
priority,
but
we
can
switch
to
in
this
case
to
the
truss
service.
I.
Think
it's
a
bit
too
late
for
1-0
to
make
this
change
so,
but.
E
All
right,
right,
I
know
it's
not
necessary
this.
It's
going
to
create
more
confusion.
I
mean,
like
you,
saw
the
ingress
box
that
we
haven't
this.
It's
always
a
can
of
worms
with
respect
to
this
interesting
and
people
will
start
expecting
like
you
know
this
and
that
to
work,
the
thing
is
I'm,
just
really
really
leery
of
like
enabling
any
kind
of
module
like
this
close
to
the
release
cycle.
Basically,
members.
E
Had
any
user
testing
so
far
and
of
all
the
tests,
maybe
you've
had
for
the
last
two
or
two
months
and
two
and
a
half
months
from
zero
point.
Eight,
yes,
I
mean
like
you
know:
we've
been
shipping
with
to
whatever
gateways
but
to
load
balance
right,
these
I
guess,
but
I
guess
yeah
and
that's
why
like
story.
So
this
is
safer
option
to
the
extent
that
if
somebody
wants
to
basically
like
you
know,
hey
I
only
have
one
god
balancer.
E
E
Step,
but
it's
right,
it's
clearer
in
terms
of
message
there
as
operationally.
It's
still
like
very,
very,
like
separate
entities
that
you
only
get
one
behavior.
If
you
use
that
if
you
get
uses
this
use
this
and
we
don't
have
to
worry
about,
like
you
know-
we've
not
seen
any
of
the
other
same
bugs
that
come
from
the
ingress
today,
I'm
trying
to
merge
this
I'm
trying
to
have
which.
B
E
You
said
virtual
service
with
the
domain
specified.
Why
is
is
not
working
and
then
explain
to
them
that
not
till
you
delete
this,
it
won't
work
and
then
they'll
be
like,
but
how
do
I
get?
How
do
I
know
if
it
is
working
I,
don't
want
to
simply
delete
it
and
then
put,
and
then
you
know,
then,
at
that
time
somebody
is
already
using
it
in
production
and
then
we
would
have
to
write
some
or
to
do
the
smudging
and
so
on
so
food
we
just
takes
us
back,
do
whatever
we
went
missing.
We.
B
Were
finding
I
mean
give
us
the
timing,
I
think
I
think
that's
a
second
structure.
I
want
to
mention
to
confess
that
my
recent
purchase
I
should
say
we're
so
right
now
in
hell.
There
is
an
option
to
actually
enables
this
kind
of
mode
where
we
have.
The
gateways
is
operating
as
both
ingress,
and
so,
if
a
user
really
has
this
need,
they
have
a
way
and
there
was
always
a
way,
because
a
generated
resources
can
specify
any
gateways
they
choose.
B
I
mean
you
can
and
that's
a
very
good
thing,
because
you
can
point
the
ingress
pool.
You
can
have
ten
five
gateways
and
you
can
choose
which
one
will
handle
the
ingress
resources.
Okay,
so
you
may
have
an
internal
robots
that
can
have
you
know
any
any
kind
of
auction
that
you
want,
but
it's
complicated
and
it's
only
for
this
user
sense,
a
default
which
is
kind
of
the
subject
of
discussion
here,
I
think
going
with
what
sure
I'm
saying,
which
is
the
safest
option.
B
G
A
B
B
A
E
B
E
E
E
Doing
that
in
one
week
will
will
probably
end
up
causing
us
more
trouble,
because
then
we
will
have
to
look
at
what
happens
in
one
point,
one
where
you
have
to
support
the
old
one
and
the
new
one
yeah,
and
that's
not
something
we
want
to
take
on,
especially
with
this
third-party
component.
Where
you
know,
we
don't
know
how
the
code
works
and
all
the
other.
B
B
E
B
E
E
E
I
think
that
can
be
very
easily
tackle
with
the
documentation
is
like
we
have
to
have
a
specific
section
which
says:
if
you
are
migrating
from
hold
ingress,
then
this
is
what
you
have
to
do,
where
we
tell
them
like
run
this
helm
command
and
basically
actually
generate
the
one,
but
the
ingress
gateway
yourself
and
you
know,
have
the
other
one
in
hand,
but
don't
enable
at
the
end
because
make
sure
everything
works.
First
and
stuff
is
all
running,
and
then
you
turn
on
the
other
gateway
without
a
load
balancer
itself,
you.
B
E
And
the
rules
and
then
slowly
switch
off
traffic
from
one
to
the
other,
but
it's
just
not
one
week
update
for
the
load,
balancer
IP,
but
that
way
that
could
that's
part
of
that
operational
book.
Yeah
Bokke,
a
sorry
book
that
we
wanted
to
write
where
we
should
tell
you
tell
people.
This
is
how
you
should
migrate.
You
know
in
production
when
you're
doing
this
thing,
which
is
which
we
have
tried
anyway,
it's
not
either
way.
E
B
E
I
B
B
H
B
J
B
K
E
E
B
B
I
I
E
E
E
D
E
We
do
the
emptiness
within
Pilate
itself
and
I
can
add
another
data
point,
which
is
that
when
I
try
to
deploy
the
minimal
version
of
his
tier,
which
has
one
which
has
no
let's
say,
makes
a
nothing
else
except
for
just
pilot,
and
maybe
the
Citadel,
if
need
be,
the
pilot
sidecar
was
constantly
crashing
because
it
has
hard-coded
mix
and
dependencies.
It's
into
F
expects
to
go
talk
to
the
mixer,
to
send
telemetry
and
so
on
so
forth,
and
that.
B
E
B
E
The
small
PR
data-
actually
it
may
be
a
harem
chat
or
something
which,
which
shows
how
to
customize
it
to
just
launch
I
mean
if
this
turns
up
all
plugins,
not
even
Citadel.
It's
just
only
pilot
and
like
one
gateway,
if
you
want
to
and
by
using
the
command
line
option
in
Pilot
to
turn
off
all
the
plugins
and
so
on
and
so
forth,
and
even
that
mode.
This
thing
just
keeps
spinning
and
breaks
literally
breaks
crashes,
big
splashes
and,
unlike
spins
yeah.
B
B
E
L
I
E
B
E
Things
but
V
being
experts
and
design
the
control
plane.
We
can
add
this
for
performance
for
reliability
in
whole,
bunch
of
other
things
for
our
site,
because
it's
like.
Yes,
we
implement
all
the
same
things
that
you
do
so
you
talk
to
your
the
same,
using
the
same
as
to
your
auth
and
you
get
the
same
matrix
and
comes
through
mixer
and
so
on
and
so
forth.
If
I.
B
Cannot
do
a
small
point
here:
I
mean
besties,
because
many
people
raise
up
the
issue
of
eating
our
own
dog
food
and
I
want
to
make
it
clear
that
mixer
team
is
working
on
direct
api
and
that's
also
don't
forget
missing
that
we
recommend
to
user.
So
if
you
have
G
RPC
server,
you
can
integrate
that
acute
mixer,
yeah
I,
don't
even
you
are
not,
and
second,
oh,
but.
L
E
B
Also,
we
want
to
show
people
that
we
can
integrate
directly
with
with
the
static
Avenger
and
the
way
that
pilot
is
out
indicating
and
in
descending
order.
Dication
is
something
that
we
recommend
for
people
who
do
direct
integration
with
Easter
and
with
the
CA,
and
if
we
want
to
show
that
the
sidecar
issues
we
can
put
one
on
Prometheus,
you
can
photograph
Anna,
which
actually
do
need,
and
they
don't
yet
have
a
solution
for
for
this
kind
of
stuff.
Yeah.
I
B
M
B
A
B
E
B
B
A
B
Here's
a
fix
we
can
discuss,
you
bring
it
with
back,
but
it's
we
have
a
PR
that
his
radius
that
fixes
a
bug.
If
he
has
an
alternative
here,
we
can
discuss
it
later.
I
A
E
I
A
So
bother
gateway
types,
so
we
have
right
now
we
have
a
gateway,
API
definition
and
the
gateway
can
fulfill
multiple
conceptual
roles.
So
we
can
have
an
external
facing
gateway.
We
can
have
a
gateway,
that's
in
a
way
external
to
the
cluster,
but
internal
to
the
mesh.
We
can
have
the
external
gateway,
and
yesterday
there
was
a
discussion
to
to
pre,
create
a
gateway
to
connect
to
it
all
the
Easter.
A
N
A
It
to
some
sort
of
an
internal
issue
gateway,
so
we
have
all
these
types
and
we
don't
have
anything
any
document
to
actually
explain
our
thinking
and
how
people
could
actually
use
the
gateway
between
the
now
between,
besides
the
logical
and
well
understood,
meaning
for
ingress
right
and
I,
like
I
can
write
something
like
that.
Maybe
we
should
put
it
on
history
on
if
people
agree,
but.
B
A
A
That's
the
thing
we
refer
to
a
thing
that
does
not
exist,
really
it's
not
configured
but
and
it's
a
research
thing
because
the
user
I
come
and
I
read
all
this
and
I.
Don't
understand
like
there
is
a
reference
in
the
blog
that
Frank
wrote
but
I
think
it's
still
not
very
clear.
I've
been
talking
with
people
I've
been
getting
a
lot
of
questions
about
this
various.
E
E
The
default
gateway
that,
like
that,
you
use
to
confuse
write
rules
for
all
the
side:
cars
inside
the
mesh.
So
when
you
write
a
virtual
service-
and
you
want
that,
like
you
know,
model
its
name
to
appear
to
all
the
other
side-
cars
inside
the
matter,
they
can
refer
to
it.
You
don't
have
to
target
every
side,
car
and
the
mesh
and
writer
so
space
you.
We
could
simply
like
it's
same
as
writing,
a
rule
which
is
which
applies
to
all
side.
Cars.
I
E
N
B
E
N
A
N
K
E
E
At
that
point,
this
is
gonna
be
contort.
The
gateway
specification
to
like
you
know,
make
the
whole
thing
as
an
option
and
then
say
this
is
mesh,
and
at
that
point
it's
no
longer
a
gateway
specification
is
just
like
you
know,
namesake,
because
it
has
nothing
configuration
that
you
can
tweak
and
configure
and
change
its
are
going
to
go
and
say
mesh
KP
I'm
only
going
to
accept
these
internal
service
registries
and
everything
else
is
gone.
I,
don't
think
so,
because
that's
not
how
pilot
just
to.
I
I
M
E
I
I
I
B
A
Definitely
need
something
to
clarify
all
this:
okay,
because
people,
people
are
not
yet
users
in
general,
they
are
used
to
copernicus,
ingress
Ani
and
the
gateway
Plaza
virtual
service
concepts
are
more
powerful
right,
so
they
allow
some
more
more
use
cases
and
we
need
to
have
you
know
we
need
to
show
the
way.
Basically.
I
I
Yeah
and
stuff
like
work
on
blog
content,
for
example,
so
like
in
the
next
little
bit,
I
have
some
gateway
stuff
that
we
can
publish
and
that
kind
of
thing
to
try
and
help
understanding,
because
I
think
that's
really.
The
key
is
like.
We
can
document
a
lot
of
this,
but
a
lot
of
it
is
just
kind
of
best
practices
that
really
don't
quite
belong
in
the
in
the
official
documentation,
but
just
need
to
be
online
somewhere,
and
so
that's
where
I
do
know.
E
I
I
D
E
I
E
A
We
don't
really
need
any,
or
at
least
we
won't
make
any
change
like
my
proposal
was
initially
to
have
a
gateway
configuration
for
the
mesh,
and
that's
obviously
like,
as
you
guys
heard
with
the
hostname,
was
the
fact
we
don't
have
a
port.
Apparently
it's
not
something
we
can
do
right
now,
and
another
thing
was
to
have
a
feel
that
the
scripts
that
you
use
it
over
gate,
with
whether
it's
internal
external
and
so
on,
but
that
also
doesn't
make
too
much
sense.
A
B
B
Now
we
have
an
option
you
have,
and
there
are
two
options:
one
one
that
creates
a
mesh
expansion
using
the
main
gateway
by
exposing
pilots
on
on
on
each
port,
and
there
is
another
option
that
creates
a
separate
gateway
of
type
int
analog
peroxide.
The
antenna
load
balancer
is
matching
what
we
had
in
0.
B
B
B
I
I
B
A
N
B
I
O
I
added
it
so
I
volunteered
to
start
working
on
this
a
little
bit
and
just
I
want
to
get
some
thoughts.
If
anybody
has
it
in
this
meeting.
In
fact,
the
prior
conversation
nationally
hit
on
some
of
the
thoughts
which
is
you
know.
Some
of
these
best
practices
would
be
good
to
have
in
an
operator's
kind,
but
specifically
to
try
to
make
some
progress
on
this.
O
If
you
look
at
content
its
plan
to
replace
it's
like
the
faq
and
the
troubleshooting,
and
we
can
constrain
it
to
sort
of
just
kind
of
replacing
and
we're
getting
a
better
flow
for
the
FAQ
and
the
troubleshooting
sections
right
now,
but
as
a
two
operators
guide,
you
would
want
more
you'd
want
some
configuration
stuff.
You'd
want
more
information
on
how
to
display
the
state
of
things
more
to
be
able
to
sort
of
spot-check
how
I
still
have
the
mesh.
Whatever
is
running,
I
think
the
risk
is
to
try
to
jump
right
into.
O
That
is
that
a
lot
of
that
content
is
already
in
is
the
Ohio
in
other
areas.
You
know
like
in
the
in
the
tasks
and
in
some
of
the
concepts
so
I
just
wanted
to
throw
this
out
there.
If
anybody
had
thought
through
this
or
had
an
idea
and
if
not
I
think
my
approach
will
be
to
sort
of
start,
putting
together
a
flow
that
has
like
configuration
and
state,
but
mostly
just
has
pointers
to
our
other
content.
You
know,
so
those
sections
will
be
not
a
lot.
O
There
won't
be
a
large
volume
to
those
sections,
but
there
will
be
a
section
to
provide
a
flow,
but
there'll
be
a
point.
They'll
like
point
to
the
task
sections
or
the
some
of
the
other
configuration
sections,
but
the
the
both
volume
is
content
would
be.
You
know
to
replace
the
FAQ
and
the
troubleshooting,
so
I
just
wanted
to
get
any
thoughts.
People
had
before
we
launch
into
this
I
think.
N
You're
totally
in
sync,
with
sort
of
the
thinking
anyway,
so
I
think
what
you
said
is
right
that
the
first
thought
was
to
sort
of
take
the
troubleshooting
and
FAQ
stuff
and
rework
some
of
those
things
over
the
clean
or
instruction
them
into
category
and
put
that
into
there.
And
the
other
thing
you
said
was
also
kind
of
the
idea
is
to
make
this
more
of
a
higher
level.
N
Point
you
off
to
other
Docs
like
talk
about
high
level
problems
and
say
you
know
here,
go
and
read
this,
and
that,
rather
than
trying
to
have
like
a
ton
of
documentation
right
here
directly,
so
I
think
that's
kind
of
what
you
said:
you're
leaning
towards
and
that's
that
was
the
current
thinking
in
terms
of
timing.
But
in
terms
of
timing,
though
I
think
really
was
you
know
what
replacing
the
troubleshooting
getting
that
totient
stuff
into
this
I.
Actually.
A
Think
we
need
this
operational
guide
along
with
the
troubleshooting
and
the
FAQ,
because
troubleshooting
is
for
some
very,
like
very
concrete
problems.
One
may
have
right,
while
the
operational
guide
is
more
like
general,
about
how
to
configure
how
to
deploy
how
to
do
stuff.
Troubleshooting
is
when
you
tried
the
operational
guide
and
you
hit
errors,
but.
B
N
N
O
Because
the
current,
if
you
look
at
the
current
troubleshooting,
there
are
really
point
issues
that
people
had
very
specific
point
issues:
it's
not
really
troubleshooting,
it's
not
really
giving
people
a
good
view
of
how
do
I
go
find
a
problem.
It's
just
here
was
a
problem.
If
you
see
it
through
this,
yes.
N
But
why
is
this
here
rather
than
the
FAQ
or
yeah
or
just
let's
move
it
out,
put
it
into
a
more
structured
place?
Here's
how
to
do
so.
I
think.
The
word
troubleshooting
is
you
know,
is
the
only
issue
here,
but
you
know
that
kind
of
information
is
all
going
to
be
captured
in
a
nice
place
that
you
go
and
say
here,
operator,
type
issues
and
and
here's
an
FAQ
that
might
point
you
up
to
this.
In
fact,
it's
I
will.
B
B
B
D
A
O
B
A
N
Asked
actually
I
will
just
want
to
go
back
one
more
quick
question
about
the
previous
topic
about
the
workloads
for
these
new
gateways,
so
the
mesh
expansion
one
it
identifies.
You
know
pod,
but
you're,
saying
there's
no,
there's
not
going
to
be
an
actual,
an
actual
server
of
pods
the
workloads
implemented
and
looking
for
I'm
gonna
run
it
by
default
for
mish
formation.
B
A
B
A
B
I
L
A
L
I
I
B
Let's
regret
no
sorry,
I
went
to
seize.
The
second
use.
Key
is
for
people
who
are
not
comfortable
to
expose
Isleworth
to
the
open
Internet,
and
they
want
to
have
a
private
behind
the
firewall.
At
this
that's
kind
of
Sony
Google
is
exposing
pretty
much
all
services
over
HTTPS
on
the
public
Internet.
But
not
everyone
is
us.
I
N
B
We
table
the
documentation,
I
could
comment
since
the
values
the
camera
license
is
auto-generated
documentation
from
that,
but
probably
we
need
to
up.
We
will
definitely
talk
date.
The
guides
for
for
Ramesh
expansions.
That's
not
on
my
to-do
list.
After
I
finish
with
the
code,
changes
by
Friday
I
think
we
are
trying
to
cut
a
final
build
Friday.
F
B
A
E
It's
done
and
it's
merged
in
I
mean
unwise.
Thing
is
merging,
I
think
it's
only.
The
change
is
to
pilot
to
like
use
the
new
option.
That's
the
only
thing
that
the
halls
yeah
so
I
think
that
lots
of
noise
today
and
then,
if,
if
the
bra
people
can
fix
that
proud
thing,
because
every
time
I
do
a
retest,
it
just
skips
the
entire
test.
Then
I
have
to
do
like
you
know.
B
E
E
F
B
A
B
J
B
I
B
Yeah,
a
few,
if
you
okay,
let
me
put
it
this
way.
If
you
know
what
you're
doing
and
you
do
some
complicated
configuration,
you
can
already
have
this
working
without
that
network
for
few
services.
If
you
have
a
small
number
of
services
in
that
demo,
for
something
and
and
it's
possible
to
do
it
today,
but
it's.