►
From YouTube: Network Service Mesh Meeting - 2019-09-17
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
B
A
D
C
D
A
A
If
they're,
if
they
turn
out
not
to
be
optional,
then
we
probably
want
to
get
an
issue
open
to
make
sure
that
we
fix
whatever
bug
is
making
them
mandatory.
Now
we
can
tend
to
use
them
in
our
examples,
because
most
interesting
examples
involve
something
that
looked
has
a
source
IP,
at
least,
but
it's
certainly
not
the
only
kind
of
examples.
A
Totally
yeah
no
I
mean
I
think
the
trick.
There,
though,
is
in
that
use
case.
You
have
a
situation
where
the
final
CNF
in
the
chain
is
providing
a
source
IP
and
you'd
like
that
to
actually
be
the
source
IP
for
the
ient
and
the
intermediate
pass-throughs.
Don't
have
any
opinion
on
this
at
all,
in
fact,
they're
a
good
example
of
not
having
an
IP
on
the
interfaces
because
they
just
passed
it
straight
through.
E
A
D
D
D
A
D
D
I
mean
we
spoke
last
time
that
this
is
the
near-term
solution
for
bigger
problem,
that
I
think
that
we
have,
and
it
is
that
we
have
too
many
things
just
happening
in
the
same
in
the
same
tree,
the
actual
results
you
know
in
in
a
lot
of
dependencies
because
everyone
likes
the
independence
cylinder
it's
where
we
are,
of
course,
so,
of
course,
the
the
I
I
can
agree
that
this
is
a
short-term
solution.
Okay,
near-term
solution-
and
my
only
question
is:
how
do
we?
A
It
also
essentially
allows
us
to
stage
ourselves
for
breaking
the
repo
up
right,
because
the
the
work
that
you
have
to
do
to
successfully
break
the
repo
up
is
going
to
intrinsically
involve
introducing
separate,
go
module
dependencies
and
the
interdependencies
among
them
the
basis
of
this.
Let
us
actually
see
the
shape
of
the
world.
If
we
were
to
break
things
to
the
separate
repos
and
from
there
we
can
sort
of
make
decisions
that
are
more
intelligent
by
doing
it
in
a
stepwise
fashion.
A
Cool,
so
is
there
any
other
discussion
that
happened
from
the
a
dream
meeting
or
stuff
that
you
wanted
to
bring
up?
That
came
out
of
that,
because
I
I'm
super
excited
about
the
multi
module.
Repo
and
I'm
also
super
excited
that
we're
thinking
through
like
how
to
properly
handle
this
Taron
feeding.
D
A
E
A
E
A
E
F
A
E
C
C
C
A
A
C
A
C
We
have
open
source
summit
in
Lyon
France.
With
a
talk
accepted.
We
have
cube
Khan
and
native
Khan
in
North
America
coming
up
with
NSM
Khan.
We
have
a
significant
number
of
talks
submitted
now
and
so
we're
pretty
pretty
happy
with
that.
So
we
will.
We
will
work
to
get
those
talks
worked
out
and
published.
C
A
C
C
A
So
basically,
fuzzing
involves
taking
your
code
and
running
essentially
trying
to
walk
the
tree
if
possible,
inputs
into
it.
It's
a
technique
that
people
use
to
detect
crashes
and
security
issues,
and
we've
recently
had
some
folks
show
up
in
the
community
who
have
quite
kindly
volunteered
to
fuzz
our
code
for
us
and
they've
started
the
process
of
fuzzing
it.
A
This
will
eventually
turn
into
PRS
with
fuzzing
unit
tests,
but
for
the
moment
they've
been
reporting
back
on
problems
that
they've
found
through
the
fuzzing,
and
these
end
up
being
almost
ideal
starter
problems
right
because
essentially,
what
you
get
is
a
panic
that
tells
you
exactly
where
the
helots
upset
right.
So
the
literally
says
there
is
a
problem.
The
problem
is
here:
this
is
the
kind
of
problem
and
that
makes
them
sort
of
ideal
places
to
start,
and
then
we've
also
got
comments
that
point
to
unit
tests
for
these
cases.
A
So
you
can
go
and
do
the
reproduction
book.
Moon's
is
the
one
who's
doing
the
fuzzing,
they're
they're
awesome,
and
so,
if
you
are
part
of
the
network
service
mission,
community
and
you're,
looking
for
a
way
to
get
started
in
contributing
to
network
service
mesh,
these
are
pretty
straightforward
problems.
A
A
A
A
Is
that
if
you
run
extensive
CI
and
you
run
it
a
lot,
you
shake
out
all
of
your
highs
and
bugs,
and
so
as
the
code
of
all's
and
you
do
fix,
comes
in
and
G
it
past
C
on
it
got
in
occasionally
it
had
a
heisenbugs,
and
so
it
shakes
out
in
stiff
in
CI
stability
a
little
bit
going
forward,
but
because
there's
so
much
CI
being
run
all
the
time
you
do
catch
them.
So,
overall
it
makes
the
code
harder.
C
C
A
A
A
The
journal
could
take
a
look
at
it
effectively.
What's
going
on
here
is
the
the
refactor
of
the
SDK,
particularly
providing
things
like
internal
traceability,
the
things
in
provide.
You
know
it
worked
out
well,
and
so
we
started
looking
at
other
components
of
the
system
that
could
be
Frank
refactor
this
way.
So,
for
example,
we
do
expect
people
to
be
building
just
like
we
have
the
SDK
for
network
service
in
points,
because
we
expect
people
to
be
rolling
network
service
in
place.
A
We
do
expect,
as
we
evolve
towards
multi
data
plane,
for
people
to
be
building
data
planes
of
various
sorts,
particularly
in
the
support
of
hardware
NICs,
but
that's
not
the
only
set
of
things
people
might
do
it
for,
and
so
we
wanted
to
make
that
easier
as
part
of
the
system,
so
SDK
style,
refactor,
the
network
service
manager.
Aundrey,
do
you
want
to
say
a
few
words
yeah.
B
B
Establishing
to
network
service
in
an
asset
manager
to
small
pieces,
with
Janus
of
smaller
network
service
operations,
all
office
implement
our
network
service
requests
protocol
and
we
do
it
for
both
local
to
remote
and
remote
to
local.
So
you
have
two
chains
of
operations
at
the
moment:
I'm
working
to
simplify
the
healing
stuff.
We
have
at
the
moment
to
have
short
chain
so
all
of
his
here,
every
user
small
pieces
of
God.
It
will
be
easier
to
miss
and
easier
to
maintain.
A
And
part
of
the
idea
here
also
is
the
there's
people
we
expect
people
to
be
writing
network
service
managers
for
other
environments
like
VM
environments
and
bare
metal
environments,
and
so
this
will
hopefully
make
that
process
easier,
and
then
it
also,
we
expect
to
start
seeing
more
proxy
network
service
managers.
Things
like
we've
talked
for
a
while
about
the
create
proxy
network
service
manager,
and
the
idea
in
both
cases
is
just
have
reusable
SDK
components.
A
You
can
chain
together
for
most
of
what
a
network
service
manager
runs,
and
then
you
can
just
write
the
one
little
piece
that
is
different
for
your
particular
Network
Service
Manager,
so
maybe
your
network
service
manager,
instead
of
turning
to
a
you,
know
an
SM
date
forward
or
which
is
kind
of
criminally
specific.
Maybe
it
turns
to
some
Sdn
controller
and
asked
it
to
do
something.
You
know
that
kind
of
thing
cool,
so
we've
already
talked
about
the
multi
module
repost
off.
F
A
A
It's
an
ARP
entry
problem,
so
yes,
yeah
any
kind
of
healing
where
a
client
gets
automatically
reconnected
to
a
new
network
service
endpoint
so
that
it
keeps
continued
service
without
Ethernet
context.
There's
going
to
be
a
different
Ethernet
address
on
the
end
video
pre
connected
to
and
therefore
your
arpan
trees
will
be
mismatched
and
it's
not
a
fatal
error.