►
From YouTube: Network Service Mesh Meeting - 2019-05-28
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
A
A
You
okay,
so
let's
go
ahead
and
get
started
then
so
we
have
recurring
this
call
who
cleared
the
domain?
The
domain
network
service
mesh
meeting.
We
have
the
NSM
documents,
call
which
occurs
weekly
every
Wednesday
at
the
same
time
as
this
mating,
and
we
also
have
the
NSM
use
case
call
which
occurs
every
second,
fourth
and
fifth
at
every
Monday
there,
every
second,
fourth
and
fifth
Monday
at
this
time
as
well.
A
A
A
So
that
is
a
that
was
a
big
discussion
that
we're
going
to
end
up
having
with
the
industry
over
at
the
jury
minimum
the
next
few
months
and
not
longer.
So
it's
definitely
good
to
make
sure
that
people
here
participate
with
with
those
discussions
and
those
discussions
is
my
guest,
so
probably
in
the
patenting
within
the
cnc.
If
telecom
user
group
and
likely
a
few
other
venues
as
well,
we
will
post
those
as
they
if
they
come
up.
A
Okay,
we
will
do
that
for
the
retrospective,
then
so
a
couple
things
coming
up
as
well,
that
are
not
on
the
current
agenda.
There
is
a
elephant
event
that
is
that
I
was
made
aware
of
that
at
the
very
minimum.
Tanner
will
be
there,
which
is,
which
is
I,
think
it
was
an
elephant
DDF.
So
if
there's
anyone
who
wants
to
so,
this
will
be
held
and
see
actually
kind
of
details
on
this
in
stock.
Oh
that's
right!
A
So
if
there's
anyone
who
wants
to
effectively
represent
the
NSM
in
this
particular
event
and
and
talk
with
people
and
build
up
some
awareness,
that'd
be
that'd,
be
fantastic.
I
am
meeting
with
with.
We
do
have
some
people
in
some
of
the
elephant
community,
so
we
have
like
people
like
Prem
who
do
work
with
odl
and
we're
also
having
conversations
with
a
couple
people
from
tungsten
fabric
to
work
out
how
to
bring
some
of
their
network
services
into
into
NSM.
I
will
discuss
more
where
that
as
as
I
learn.
A
A
A
So
if
it's
it's
the
Thursday
and
Friday
before
ons
Europe,
which
I
believe
occurs
on
Monday,
Tuesday
and
Wednesday,
the
call
for
papers
on
that
is
on
May
31st.
So
if
anyone
wants
to
talk
about
DVD
k
and
then
assembler,
that
would
be
a
good
venue
to
put
in
a
talk.
But
the
deadline
is
literally
in
a
couple
of
days,
so
today's
the
28th,
so
the
deadline
is
probably
depending
on
timezone,
Wednesday
or
Thursday.
A
C
No,
no
I
I
just
wanted
to
make
a
small
remark
about
the
coop
con
North
America
I
mean
we
still
have
time
for
planning,
etc,
etc.
But
my
impression
from
the
last
from
on
Cuba
on
you
is
that
these
two
two
sessions
of
35
minutes
are
not
enough.
I
mean
once
people
when
you
open
the
for
for
questions.
People
start
asking
a
lot
and
it's
good
to
have
the
discussion,
so
my
suggestion
would
be
if
we
have
to
choose
again
between
on
85
minutes,
not
if
I'm
not
wrong
and
to
35
minutes.
D
C
B
A
Yeah
I
think
that's
a
good
observation.
A
the
one
thing
that
I
with
the
85
minute,
one
is
I,
think
they'll
be
a
bit
more
preparation
upfront
because
more
likely,
we
they
may
ask
for
some
type
of
a
hands-on
thing.
So
we
should
make
sure
that
we
get
the
details
on
that,
but
I
I
actually
strongly
agree
with
you
as
well
like
35
minutes.
It's
like
crunch
time
and
we
end
up
with
the
one
of
the
other
thing
that
makes
it
problematic.
Is
there
were
a
lot
of
people
in
the
deep
dive?
A
B
C
B
Really
good
the
videos
and
slides
are
already
up
at
the
event.
Page
I
have
been
tracking
Google
Analytics
against
the
network
service
measures
that
I
okay,
so
we
had
feet
hundred
seven
last
week
come
through,
which
is
a
big
jump
for
us,
and
then
we
also
had
you
know.
I've
also
been
I
was
also
using
QR
codes
in
order
to
track
by
different
campaigns
attached
to
QR
codes.
B
C
Yeah,
so
one
thing
that
I
wanted
I
I
have
put
here
the
link
to
the
telco
using
Telecom
user
group,
kickoff
presentation
and
the
video
is
on
YouTube
already
he
posted
it
into
the
2g
channel
on
CN
CF
just
for
reference.
If
anyone
is
interested
but
I
actually
I
liked
a
lot.
This
slide
I'm
still
about
to
watch
the
video
and
to
see
what
actually
was
discussed
here.
I
think
my
observation
from
from
from
this
slide
here
and
my
conclusion,
for
this
is
okay.
We're
part
of
this
wide
landscape.
C
I
mean
there
are
a
lot
of
things
there.
So
what
I
would
really
and
I
have
some
something
in
the
end
of
the
world
group.
Call
if
you
manage
to
get
to
it,
like
their
old
maps,
think
that
I'm
trying
to
bring
for
a
couple
of
course,
my
my
main
observation
here
would
be
we're
part
of
a
wide
landscape.
We
should
try
to
start
interoperate
with
some
of
these
guys
around
us
to
just
have
our
roots
there.
You
know,
connections,
strong
connections
not
be
just.
We
are
another
bullet
in
the
list.
A
E
C
You
can
do
yes,
you
can
literally
do
whatever
you
feel
like
doing
that's.
Why,
in
one
of
the
slides,
I
was
pointing
to
our
examples,
repo
we're
actually
the
only
one
of
the
examples
is
based
on
TVP.
Everything
else
is
pure,
like
given
more
simpler,
okay,
not
one
to
toot
example.
Sorry,
obviously,
but
others
are
more
or
less
simpler,
simple
examples,
but
they
are
not
straightly
bound
to
VP.
Now
the
reference
to
VP
was
because
we
already
have
the
ACL
filtering
there.
We
have
the
cross
connects
as
a
concept
out
of
the
box.
C
C
We
should
start
calling
this
data
plan,
because
I
think
that
this
is
something
that
actually
is
the
confusion,
because
in
our
glossary
we
moved
away
from
this
on
the
data
plane
is
essentially
the
chain
of
CNS
that
run
on
top.
So
we
call
this
kind
of
four
the
elements
or
forwarding
plane
or
something
along
those
lines
are.
A
One
of
them
is
the
network
services
as
I
described,
which
I
think
it's
probably
the
highest
value
area
for
them,
but
also
put
the
potential
of
getting
them
as
a
as
a
data
plane
as
well
and
getting
them
to
implement
the
data,
API
and
I
think
if
we,
if
they
were
to
implement
the
data
play
an
API
they'll,
give
us
a
really
good
feedback
on
where
on
where
the
API
is
slacking.
So
we
can
get
something.
That's
not
that's.
We
don't
just
say
is
generic,
but
it's
probably
generic
and
simultaneously
that
they
can.
A
C
In
partner
we're
also
looking
even,
we
started
at
the
conference
looking
at
how
we
can
implement
a
very
simple
kernel
based
and
actually
I
found
an
issue
today,
just
just
for
reference
just
to
have
it
there,
just
a
simple
Linux,
kernel-based
forwarding
just
yeah.
We
have
something
as
a
baseline.
That
says
this
is
what
you
can
do
and
if
you
want
advanced
features,
use
some
of
the
other
guys.
F
A
F
A
Sounds
to
me
that
they
were
interested
in
doing
a
something
that
was
too
fabric
related
as
the
as
a
kubernetes
as
a
data
plane
that
that
sits
within
the
kubernetes
part.
Okay,
I
I'm
also
going
to
ask
them
to
look
at
something
that
is
similar
to
to
the
odl
path,
because
one
of
our
goals
is
to
be
able
to
integrate
multiple,
multiple
groups
and
there's
there's
areas
where
audio
absolutely
shines,
especially
with
the
Netcom
support
and
there's
other
areas
where
it
tungsten
fabric
really
shines
that
are
complementary
and
so
I'm.
A
What
I
want
to
do
is
I
want
to
try
to
make
sure
that
we
find
what
areas
are
complementary
to
MSM
and
the
community
so
that
we
can
get
good
alignment
and,
in
the
long
run
sure
them
I
mean
there'll,
be
some
healthy
competition
in
certain
areas.
We
want
to
find
out
in
that
early
amendment
in
for
for
the
areas
where
they
don't
cover
each
other.
Okay,.
A
I'm
gonna
have
a
talk
with
one
of
their
with
one
of
the
people
from
juniper
this
week
and
we're
gonna
give
me
a
better
understanding,
because
I
really
don't
understand
on
some
fabric,
if
you
like,
and
but
once
once
he
I
have
a
conversation
with
them.
Then
I'll
then
I'll
be
in
a
call
with
him
next
week
to
try
to
to
work
out
that
alignment.
So.
F
F
A
A
C
B
B
C
B
B
A
C
C
A
B
So
but
we
are
we're
trucking
right
along
in
terms
of
our
followers
we're
up
to
221
now,
which
is
good
and
a
lot
of
tweeting
happening
and
there's
a
lot
of
things
being
retweeted,
and
you
know
we're
all-
is
going
really
well.
A
A
C
C
B
G
C
So
from
my
point
of
view,
it's
a
much
much
much
better
situation.
Then
I
mean
at
some
point.
We
had
a
lot
of
problems
with
the
public
court,
specifically
leaking
resources,
etc,
etc.
So
I
would
say
that
the
CI
in
general
is
in
much
better
shape
and
then
the
test
themself
also
are
behaving
not
that
bad,
so
I
am
running
kind
of
a
parallel
CI
just
on
a
circle
CI.
For
my
examples:
rep,
okay,
my
it's
not
mine,
which
there.
B
C
C
If
you
just
want
to
deploy
it
on
a
plain
cluster
and
deploy
some
of
the
examples
on
top
it
works,
more
is
stable.
Now
there
is
some
more
strange
issue
where
the
order
link
of
the
deployment
matters,
but
okay
I
will
try
to
figure
out
the
integration
test
and
send
it
for
more
investigation,
but
I
mean.
D
C
Am
more
more
or
less
I
can
say
that
I'm
happy
with
whoever
I
see
from
the
core
components.
We
have
just
integrated
the
latest
release
of
EPP,
which
was
released
yesterday.
Two
point:
one
point:
one
which
should
help
us
have
a
better
ipv6
support,
amongst
other
things,
so
I
think
that
we
are
moving
into
the
right
direction.
C
B
C
G
C
Think
that
we
have
a
lot
of
the
things
done
now.
I
mean
the
downstate
there's
a
lot.
We
have
something
seems
light
which
are
not
that
much
mostly
my
from
the
things
that
he
we
have
fault
lines
already
in
the
channel
and
here
on.
The
left.
I
think
that
we
just
need
to
go
again,
one
by
Y
and
figure
out,
what's
Stuart
Inc,
and
what
because
I
think
that
some
of
the
things
are
yeah
more
or
less
figured
out
already.
If.
A
C
E
B
B
E
C
So
yeah,
my
my
at
least
feeling
here
is
that
for
this
first
release
we
just
need,
if
you
can
just
say:
okay,
if
you
deploy
this
on
a
clean
cluster
around
whatever
workload
you
want
to
run
and
then,
if
you
want
to
run
it
again,
you
have
to
clean
up
everything
and
not
rely
on
things.
Getting
up
and
down
deploying
things
and
stuff
I
mean
it's
the
very
first
release.
It's
our
fights.
B
Okay,
you
shouldn't
you
shouldn't
have
to
tear
down
the
NSF
infrastructure
every
time,
because
then
you
can't
have
multiple
workloads
and
multiple
network
service
and
points,
and
things
going
successfully
right,
though
it
has
to
at
least
be
the
case
running
in
a
long-term
and
whoa
and
redeploy
various
things
on
top
of
it
and
still
have
stuff
work.
Well,.
C
We
all
know
that
there
are.
There
are
things
like
corner
cases
where,
if
I
deploy
something
and
then
while
deleting
it,
I
deployed
the
other
thing
and
there's
this
intermediate
state
where
things
get
wrong-
and
you
know
all
these
kinds
of
strange
things
happening
because
to
be
frank
in
our
CI
weekly,
the
environment,
every
time
I
mean
after
each
test.
So.
B
D
B
We
should
be
able
to
keep
the
infrastructure
up
and
deploy
these
things
and
have
stuff
work
and
not
be
dangling
things,
so
that
may
be
something
to
look
at
after
the
release
in
terms
of
the
integration
testing.
It
may
be
something
you
may
want
to
think
about.
Andre
su
I
know
you're
thinking,
deep
thoughts
about
other
creation
testing
going
forward,
because
we
do
need
to
make
sure
we
function
properly.
The
infrastructure
staying
continuously
up
and
workloads
coming
and
going,
but
I
mean
with.
C
Our
first
release
I
think
it's
more
about
proof
of
concept,
people
that
can
deploy
and
see
how
distinct
works
in
practice,
but
at
least
I
don't
expect
that
this
should
work
in
order.
You
know
getting
cup
down
things
and
moving
around
and
etcetera
etcetera
I
mean
we
are
I,
think
we
are
not
verifying
this
at
least
yeah.
B
C
A
C
C
F
C
This
is
something
that
I
was
I
just
wanted
to
bring
up.
I.
Don't
think
that
this
is
something
that
we
should
fix.
It's
just
more
like.
Let's
talk
a
little
bit
like
open
and
like
in
a
common
call
about
the
things
that
we
envision
next,
so,
of
course,
I
have
put
the
first
couple
of
items
here,
so
I
can
start
it.
If
you
let
me
so,
of
course,
the
lease
is
the
first
thing
which
which
we
are
actively
working
there.
C
Then
there
is
the
story
of
the
example,
so
I
try
to
keep
them
in
a
separate
report.
There's
been
discussions
about
renaming
things
in
the
main
repo,
so
I
would,
at
least
from
my
point
of
view,
I
would
like
to
have
the
confirmation
of
the
community
that
we
would
like
to
keep
these
things
separate
maintained
in
a
separate
area
protested
continuously.
It
is
the
way
that
I'm
doing
now
so
I,
don't
know
it
Fred
Andre,
it
tastes
from
the
things
I
generally.
B
That
said,
I
do
realize
there
are
some
times
sure
we
deal
with
care.
I
would
love
a
world
where
you
know.
We've
got
the
imprint
one
repo
and
we've
got
the
examples
of
another,
and
so
you
know
when
we're
doing
the
CI
we're
not
having
to
rebuild
the
examples
every
time,
even
though
they
haven't
really
changed.
C
Okay,
any
other
opinions
here,
so
the
other
thing
that
we
have
that
we
have
discussed
that
cube
con
with.
What's
on
it,
some
of
the
other
guys
I
think
with
Taylor
before
it's
about
the
scenes
you
have
tests
based
in
a
moment.
This
is
something
that
is
bending
for
for
a
while
and
I
am
personally
I
mean
I
would
like
to
get
involved
them
to
his
poke
with.
What's
on
a
lot
about
what
are
the
things
there
are
and
the
plans.
C
A
Thing
that
we're
waiting
for
at
the
moment
was
there
was
a
real
name
of
the
guy
fermenter
you're
like
Michael,
Henderson
or
Pearson,
or
something
somewhere
that
is
working
on
getting
VPP
networking
with
OpenStack
doing
its
thing,
and
they
were
having
some
connectivity
issues
between
between
node
to
node
communication
between
OpenStack
nodes.
So
once
he
had
that
working,
that's.
A
B
D
B
To
just
have
somebody
sit
down
with
you
guys,
the
denver
in
particular
and
I
think
you
also
wants,
and
maybe
we're
interested
in
actually
doing
a
little
bit
of
coding
necessary
for
the
other
C's
that
you
need
and
more
having
a
partner
to
help
you
through
rather
than
having
someone
go,
do
it
for
you
did
I
get
that
correctly.
Yeah.
D
B
B
You
know
that
you're
gonna
use
everywhere,
which
is
just
got
something
that
advertises
a
service
and
consumes
the
same
service,
because
the
basic
topology
you
guys
have
is
a
loop
with
your
V
switch
and
use
that
everything,
including
the
V,
switch
and
then
or
any
auxilary
completely.
You
need
that's
not
currently
being
able
by
a
network
service
mesh
with
the
connection
of
the
interfaces,
just
push
that
directly.
D
C
I
was
actually
thinking
to
take
whatever
we
call
the
firewall
now
and
actually
enhance
it
to
be
completely
config
mark
configurable
to
expose
as
much
as
possible
from
the
BP
agent
capabilities
like
routing
the
cross
connect,
so
something
that
you
can
essentially
configure
through
a
config
map.
So
this
is
something
that
we
call
to
it
Watson,
something
like
the
universe,
so
CNF
like
you
can
you
can
do
routing
or
you
can
actually
I
asked
the
guys
there
from
the
agent.
They
were
kind
thought
the
ACO
based
forwarding.
B
This
is
fine.
The
reason
I'm
laughing
is
that
we're
building
we're
building
a
router,
really
a
rudder
and
switch
so
yeah
I
mean
that's
also
true.
I
I've
been
thinking
more
from
the
direction
of
things,
something
relatively
an
ambitious
that
you
know
those
guys
could
do
themselves,
but
but
you,
if
you're
gonna,
go
actually
help
them
I'll
leave
it
in
your
April
hands,
because
your
approach
sounds
fine
as
well
yeah,
because
I
think
at
the
end
of
the
day,
however,
we
get
there
from
the
point
of
view
of
code.
B
The
fundamental
thing
we
need
right
now,
because
it's
the
functionality
we're
currently
using
is
something
where
you
know.
Basically,
the
beasts
which
comes
up
requests,
a
connection.
The
first
thing
it
hits
you
gets
plugged
into
the
first
and
see
n-f-l,
that's
just
a
routable
interface,
because
there
are
going
to
be
some
routes
and
so
forth,
and
we
can
ramp
up
and
ramp
down
and
listen
dice.
The
steps
we
take
to
get
for
tween
a
and
B,
but
it
but
fundamentally
I
think
that's
where
they're
starting.
C
D
I
guess
to
review
also
with
what
Danna
saying
in
the
user
group
to
telco
user
group,
where
he
had
or
volunteers
Lucena
for
saying
what
all
is
deployed:
out-of-band.
Well,
the
B
switch
and
a
bunch
of
other
things
in
tests
that
are
deployed
out-of-band.
The
more
that
we're
using
the
network
service
match
the
more
deployed
in
the
cloud
native
way,
and
we
can.
If
we
can
talk
about
that
in
the
fantastic.
A
B
Yep
and
the
thing
is
I
think
we
can
progressively
chip
away
at
that.
You
know
because,
for
example,
things
like
the
SRBs
REO
vpf
thing
is
probably
a
little
a
little
further
out
for
network
service
mesh
as
a
feature,
but
we
can
you
know
we
can
do.
We
can
do
some
out-of-band
work
and
not
even
terribly
much
out-of-band
work
with
that
until
we
get
there
and
still
use
network
service
mesh
for
plumbing
the
connections
between.
A
A
A
F
Yeah,
so
mesh
3
as
a
project
is
essentially
meant
to
do
because
benchmarking
around
service
meshes.
So
what
bunch
of
us
are
doing
is
trying
to
get
networks
of
this
mesh,
and
the
idea
here
is:
how
do
we
benchmark
because
in
case
of
service
mesh,
they
would
essentially
traffic
where
the
English
controller
and
then
try
to
see
how
it
performs?
What
we're
thinking
was
probably
develop.
A
network
service,
mesh
client
and
then
take
one
example,
see
enough
and
then
try
to
do
a
bit
of
benchmarking
around
it.
A
F
A
So
perhaps
I
was
thinking
more
like
in
to
work
I
like
the
details
like.
If
this
is
a
path
that
we
really
want
to
go,
that
we
really
want
to
go
down,
then
being
able
to
get
the
details
on
exactly
what
we're.
What
are
we
like?
What
do
we
need
to
provide
them?
What
do
they
would
they
want
to
to
provide
us
and
try
to
work
out
that
integration,
yeah.
G
C
Perfect,
so
the
next
thing
that
I
have
put
here
in
the
list
is
this
service
mission
interface,
which
was
the
new
the
new
service
mesh
hype
that
that
was
announced
at
cube
con
now.
At
first
glance,
it
actually
seems
a
little
bit
far
from
what
we
are
doing,
but
on
the
other
side,
if
you
remember
this
slide
that
I
wanted
to
point
out
from
the
GOG
kickoff,
where
actually
we
are
kind
of
a
part
of
a
landscape.
So
this
is
again
its
line
with
what
I
think
that
we
should
try
to
do
so.
C
I
would
like
I
mean,
and
on
our
side
we
are
going
to
dedicate
some
resource
actually
mama.
Here
she
is
going
to
work
to
try
to
figure
out
what
this
pack
is
about.
How
this
aligns
with
us
not
really
the
line
today.
Can
we
apply
to
change
some
things
there?
Can
we
can
we
try
to
figure
out
some
adapters
and,
in
general,
how
this
thing
is
going
and
again?
This
is
all
about
being
part
of
the
wider
landscape
and
try
to
be
I
mean
if
these
are
specs
that
are
going
on.
C
A
D
A
And
splits
there
they're
effectively
the
same
so
I
think
so.
I
think
it's
as
some.
My
as
it
is
right
now
is
not
very
I'm.
Gonna
say
it's
not
interesting,
because
it
definitely
is,
but
it's
it's
something
that
plays
at
a
level
that's
way
higher
than
then
we
play
at.
However,
we
I
think
we
have
a
couple
options,
so
one
option
would
be
to
try
to
work
out
how
to
enable
these
type
of
things.
So
someone
says
something
like
they
want
to
do
a
split
or
something
we
could.
A
We
could
help
with
some
of
those
with
some
of
those
splits
if
they,
if
they
want,
especially
when
you
started
looking
at
things
like
how
do
you
do
federated
clustering,
because
that
way,
if
you're
trying
to
access
something-
and
you
want
to
say
that
this
thing
could
be
split
across
multiple
clusters-
we
we
could
provide
help
with
that.
Another
thing
as
well
as
we
could
also
contribute
to
the
spec
and
push
things
that
are
that
are
lower
level
and
do
it.
C
A
C
B
A
basic
rule
that
applies
here
it's
the
law
of
no
unnatural
acts,
which
is
to
say,
I,
think
it's
a
really
good
idea.
First
to
engage
but
I
think
we
big
lion
and
Frederick
are
saying
which
is:
let's
not
try
and
commit
unnatural
acts
by
forcing
ourselves
there's
the
HTTP
part
of
the
spec
is
pranaam
death.
C
B
B
F
B
We
should
be
able
to
do
chaos
gorilla
where
we
kill
everything
in
the
Indies.
We
basically
update
restart
the
entire
service
measure,
all
at
once
and
everything
still
works
and
then
the
other
one
we
talked
about
cube.
Con
was
dynamic
rewiring,
particularly
the
packet
capture
example,
and
then
we
also
have
the
hardware
NIC
/a,
sorry
avi,
stuff,
also
that
we
will
eventually
need
to
do
for
the
NFA
use
cases.
A
A
B
G
G
Okay
for
guerrilla
yeah,
okay,
so
general
idea
is
to
have
in
it
container,
to
request
initial
NSE
pass
a
connection
to
the
monitoring
container
as
a
sidecar.
So
we
will
be
able
to
do
fully
recover
from
this
sidecar
for
services
and
the
same
could
be
achieve.
It
use
increased
decay
if
monitoring
can
of
an
SMD.
G
B
Yep,
but
you
know
the
packet
capture
example
by
the
way
for
dynamic
rewiring.
This
was
whether
we
actually
gotten
talking
to
people
at
the
conference,
which
is
if
we
can
do
that
in
a
McCree
wiring.
We
can
wire
in
something
like
a
web
shark
packet
capture
box
dynamically
into
the
chain
of
a
particular
pod.
So
let's
say
you've
got
a
pod
that
it's
got
some
weird
things
going
on.
You
can't
tell
what's
happening,
you
can
dynamically
wire
a
packet
capture
box,
say
before.
E
B
So
you
picture
this
way.
If
you
have
a
pod,
that's
got
a
connection
between
the
pod
and
the
the
workload
to
a
firewall
to
a
VPN
gateway.
The
idea
we
had
dynamic
rewiring
would
be
to
allow
you
to
selectively
for
a
particular
pod.
Why
are
a
packet
capture
box
before
the
firewall
on
the
VPN
gateway
so
that
you
could
actually
capture
and
see?
What's
going
on.
D
E
E
B
Overwhelmed
and
I
need
to
tell
somebody
hey.
This
is
not
working
for
me.
I've
got
too
many
connections.
Could
you
possible
that's
another
interesting
one
to
throw
into
the
mix.
To
think
about.
I
would
throw
that
into
the
same
general
category
as
Auto
healing,
which
is
to
say,
with
Auto
healing
you're
you're,
basically
saying:
look,
okay,
I
no
longer
have
the
connection,
I
need
to
the
thing
I'm
talking
to
it,
so
I
gotta
go
find
something
else
to
do.
B
I
think
it
described
it's
not
quite
the
same
as
auto
healing,
but
it's
in
the
same
flavor
and
that
it's
not
something
the
user
asked
for
it.
So
it's
a
dynamic
attribute
of
the
system
when
I
was
thinking
about
with
that
a
mercury
wiring
was
the
user,
makes
legitimately
different
policy
choice
about
how
they
want
the
wiring
to
look
and
that
we
go
ahead
and
take
existing
workloads
and
rewire.
The
in
response
to
that.