►
From YouTube: Network Service Mesh WG meeting - 2019-01-08
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
C
A
So
quick
reminder
to
everyone,
especially
new
folks,
on
the
call
this
call
is
recorded
from
time
it
starts
and
then
what
it
ends
that
is
automatically
uploaded
to
YouTube,
so
do
realize
that
you
are
being
recorded
and
what
you
say
here
will
be
public
yep
cool.
That
said,
here's
the
link
meeting
minutes
and,
if
could
start,
adding
themselves
to
the
meeting
minutes.
That
would
be
useful.
A
B
A
C
A
For
folks
who
are
just
arriving,
we
usually
allow
a
few
minutes
for
folks
to
turn
up
usually
takes
till
about
5:00.
After
all,
so
please
keep
in
mind
that
the
meeting
is
recorded
from
the
time
that
it
starts
and
is
automatically
posted
to
YouTube
afterwards,
so
everything
you
say
is
being
recorded
and
will
be
public.
If
folks
could
please
go
to
the
meeting
minutes
that
I
just
stuck
in
the
chat
right
now,
I
stuck
them
in
the
chat
all
but
I
just
stuck
in
the
chat,
the
you
know,
we
keep
a
list
of
attendees.
A
A
B
A
B
D
C
B
E
A
B
D
B
A
Is
frankly,
this
is
another
option
we
can
take,
which
is
that
you
can
specify
the
box
conditional
on
the
provider
so
that,
basically,
if
your
provider
is
the
first,
then
you
use
oh
whatever,
and
that
actually
makes
us
quite
a
bit
stronger,
because
the
the
broader
set
of
environments
that
people
are
working
in
with
that
work,
surface
mesh,
the
less
likely
we
are
to
have
something
that
works
in
one
environment
and
not
another
I
was.
E
E
Libvirt
to
work,
but
another
option,
rather
than
using
vagrant,
is
actually
to
use
terraform
directly
against
lid,
which
is
probably
the
better
approach
to
be
perfectly
honest,
but
it
does
mean
that
you're
not
using
a
box.
You
need
an
image
rather
than
a
vagrant
box,
but
the
problem
with
favoring
boxes
is
they're,
always
tied
to
a
provider
or
so
that
actually
works
better
in
some
regards
yeah.
A
E
It
ended
up
with
another
layer
of
indirection
which,
as
we
know,
solve
all
problems,
but
it
was.
We
have
I'm
going
to
blame
Frederick
here
because
his
hand
his
footprints
were
all
over
it.
The
make
files
are
a
bit
funky
because
they're
not
really
made
files
they're
just
glorified
script
files
in
the
form
of
the
make
file,
but
they're
they're
nested
in
such
a
way
that
if
you
change
out
the
provider,
you
just
basically
change
it
loads.
The
different
more
uses
a
different
set
of
make
targets.
F
E
It
certainly
part
way:
I
can't
say
I'm,
proud
of
that
particular
piece
of
work
yet,
but
the
theory
works
you
just
use
terraform
to
do.
I
think
I
was
doing,
is
using
terraform
to
create
virtual
machines
as
well
as
get
them
configured,
and
that
was
making
steady
progress
and
it
also
theoretically
means
I
think
the
long-term
aim
would
be.
That
would
be
a
good
one,
is
to
use
terraform
consistently
for
everything,
rather
than
doing
vagrant
one
way
and
another
things
a
different
way,
but
it
gave
some
possibilities.
E
We
can
but
seriously
it
was
like
three
weeks
ago
when
I
did
this
and
I
can't
remember
what
the
hell
I
did.
So
the
actual
answer
here
is
for
me
to
stick
that
on
github,
with
no
promise
that
it
works.
So
at
least
somebody
can
see
whether
or
not
I'm
doing
the
right
thing
or
whether
or
not
there
might
be
a
better
approach.
There.
A
C
E
E
D
A
D
D
D
D
E
D
B
A
D
Yeah,
so
one
of
the
things
that
I
have
been
taking
a
look
at
is:
how
can
we?
How
can
we
take
a
look
at
something
like
envoy
and
see
if
we
can
can
chain
envoy
into
a
Indian
as
a
network
service,
and
so
this
has
a
variety
of
very
powerful
immigration,
so
we
can
do
which
may
help
with
multi
cloud
or
multi
cluster
use
cases,
and
also
so
potentially
may
also
solve
an
issue
that
they
currently
have
with
on
land.
D
This
deal
at
the
moment,
which
is
that
you
need
to
have
a
privilege
to
make
container
start
up
an
a
that
and
be
tables
mangling,
which
may
provide
an
opportunity
for
treating
all
it
has.
It
ends
up
waiting
to
why
it
doesn't
happen
immediately.
So
there's
so
there's
an
opportunity
there
that
we
have
and
there's
also
some
interesting
use
cases
like
what
happens
if
your
end
away
dies
or
what
happens
if
we
decide
to
look
great
so
there's
a
lot
of
really
nice
features
that
we
could
that
we
could
add
into
that
space.
D
E
H
H
D
Some
links
in
the
chat
when
you're
looking
at
this
is,
is
it's
not
about
like
trying
to
compete
with
them
it?
There
may
actually
be
an
integration
point
where,
where
the
techniques
that
they're
using
and
the
technique
that
we
were
using
with
our
powers
combined,
you
know
you
know
they're
definitely
I
mean.
H
E
He
will
almost
certainly
have
a
better
understanding
the
problems
and
we
do
grabbing
wealth.
All
right.
John
knows
what
he's
talking
about,
but
I,
don't
and
and
Tim's,
basically
and
at
the
coalface
he
he's
working
on
the
stuff.
So
he
probably
understands
these
things
better
still.
So,
regardless
of
whether
he
has
any
inclination
to
change
his
approach,
it
would
be
nicer
if
we
understood
the
problem
in
its
full
detail,
which
is
almost
certainly
happy
to
explain
to
us.
So
there's.
D
D
C
Sorry
double
muted
I
was
just
curious
and
unfortunately,
it
overlaps
with
OH&S,
but
I
went
to
Paris
last
year
and,
to
be
honest,
the
in
a
fee,
MPLS
Sdn,
World
Congress,
is
probably
one
of
the
best
conventions.
I've
ever
been
to
that
a
lot
of
really
really
smart
people
from
amia
in
the
u.s.
there
and
I
was
just
kind
of
curious.
If
NSM
is
going
to
start
trying
to
push
into
that,
I
mean
it's
very
Sdn
and
NFV
focused,
so
I
feel
like
an
SM
is
probably
a
good
fit
for
that
conference.
C
I'd
have
to
do
a
lot
of
homework.
First
I,
don't
want
to
go
there
and
look
like
a
goober
in
front
of
the
entire
world
and
put
in
SM
back
so
I'm
I
could
talk
to
you
offline
ed
about
potentially
doing
that,
but
I
would
have
to
do
quite
a
lot
of
homework.
I've
just
started
standing
this
stuff
up
in
my
lab
with
them
another
homegrown
kubernetes
project
that
we're
building
in-house
trying
to
make
them
work
together.
So
what
we
kissed
that
tereska.
E
E
It
for
that
audience
particularly,
but
but
this
is
probably
true
of
ons-
it's
nice-
to
talk
about
the
technology
and
how
we
put
it
together,
but
it's
far
more
useful
to
show
it
doing
things
that
are
useful,
and
you
know
in
in
that
regard.
We
should
probably
work
out
what
would
be
a
useful
demonstration
for
that
kind
of
audio.
A
C
Yeah
with
it
being,
you
know,
obviously
MPLS
being
one
of
the
big.
You
know
categories
that
it
covers
it's
very,
very
geared
towards
service
providers
like
charter
and
do
what
to
tell
common
them.
Like
I,
see
a
lot
of
my
SP
peers
there.
It's
you
said
it's
probably
the
best
conference
I've
ever
been
to
no
offense
to
the
Cisco
life
craft.
A
I
There's
also
the
container
world
that
happens
every
year
in
Santa,
Clara,
Convention
Center,
it's
well
attended,
and
probably
that
would
also
be
in
fact,
I
got
an
invite.
This
year
last
year,
I've
given
a
presentation
on
cloud
net
of
architecture
for
telco,
probably
we
can
present
about
network
so
much
this
year.
A
E
Again,
if
we,
if
we're
putting
this
list
together
and
it's
going
in
the
notes
and
then
what
appeals
to
the
audience
would
be
really
really
useful
information
that
I
think
is
again
right.
We
know
it
could
come
them
talking
about.
Seo
would
be
a
fantastic
thing.
We
know
that
ons
if
we
start
talking
service
provider,
that
would
that
would
definitely
kind
of
them
start
them.
A
E
D
E
E
E
D
Thank
you,
so
the
main
topic
that
I
want
to
start
with
is
the
is
about
the
repo
location.
So,
right
now
the
repo
was
stored
within
legato
and
legato
has
been
absolutely
fantastic
and
very
gracious
with
letting
us
use
their
space
to
get
started.
But
as
we
look
at
our
direction
moving
forward,
there's
a
couple
things
coming
up:
we're
keeping
it
in
the
got
to
may
not
be
the
best
place.
So
first
one
is:
it
is
an
independent
project.
D
It's
not
part
of
legato
which
may
cause
some
issues
when
we
start
to
approach
groups
like
C
and
CF
and
to
become
a
member
and
simultaneously
when
we
there's
been
discussions
about
splitting
up
the
repo
into
multiple
repos
and
having
a
basically
flooding
of
the
legato
namespace
is.
Is
that
good
for
so
I
owned
the
network
service
mesh
group
in
in
github?
D
And
so
my
recommendation
is
that
we
move
the
bean
every
service,
mesh
repo
to
the
network
service,
mesh
group
and
add
all
the
appropriate
credentials
and
so
on,
and
the
main
thing
we
will
have
to
look
at
in
this
particular
path
is
going
to
be
the
circle
CI
integration
and
make
sure
that
all
our
web
hooks
work
and
so
on.
And
besides
that
it
should
be
a
relatively
straightforward
move.
So
anyways
we'll
leave
the
topic
open
so
for
comments
or
questions
right
now,.
A
Their
opinions,
often
when
people
express
opinions,
you'll
get
a
sense
of
consensus.
If
there
is
some
agreement
we
might
fall
back
to
voting,
they
die.
I
tend
to
I,
tend
to
prefer
consensus
as
a
way
of
through
sorting
out
these
questions.
The
next
thing
is
to
sort
of
ask
okay.
Well,
what
would
be
involved
right
so
I'd
be
curious
in
hearing
a
broader
set
of
thoughts
and
opinions
from
folks
on
the
notion
of
doing
this
and
I
know.
D
E
D
D
You
know
make
sure
that
everything
is
is
working
and
that
makes
make
sure
the
community
is
not
having
any
problems
committing
to
it
or
doing
poor
request
and
so
on,
and
then
we
and
then
we
step
forward
and
I
think
we
also
need
to
have
some
documentation
up
on
the
page
as
to
how
to
how
to
how
to
update
your
git
repo
to
point
to
the
new.
Actually,
it
should
do
forwarding
automatically,
but
it's
probably
some
good
idea
to
to
point
to
the
network
service
mesh
repo
directly.
D
E
E
D
D
D
Okay,
so
in
in
that
particular
scenario,
yeah,
that's
let's
go
and
start
putting
together
some.
Let's
start,
they
start
putting
together
some
information
on
this
well
I'll
reach
out
to
get
up
today
and
I'll
I'll
ask
them
about
the
NSM
thing,
borrowing,
barring
any.
If
they
say
no,
then,
like
I
said,
like
I,
said:
I
think
we
should
move
the
legato
site
first
and
we
may
need
we
may
need
to
ask
actually
Edie
who
owns
that
is
the
I
forget
the
name
of
was
identifying
who
owns
the
account
that
that
does
the
public
I.
A
Believe
if
memory
serves
that
I
currently
own
the
domain
name,
so
I
will
reach
out
to
Kyle
and
see
what
we
can
do
about
hitting
credentials.
The
net
worst
case
scenario.
We
just
create
a
new
net
with
the
account
and
share
the
credentials
a
little
bit
more
appropriately,
that
from
there
and
move
the
domain
over
to
it.
D
B
Okay,
so
sometime
before
Capcom,
North
America,
like
in
the
beginning
of
December
I,
started
trying
to
extract
the
common
code
from
all
the
examples
that
we
have
and
try
to
put
it
in
a
kind
of
an
SDK
tree.
I
didn't
know
that
it's
an
SDK
until
he
told
me
that
it
is
yeah,
so
I
have
a
PR.
There
I
put
some
work
there.
It
already
reviewed
it.
Thank
you.
We
will
try
to
adhere
to
his
comments.
Of
course.
For
the
time
being,
it
was
mostly
about
just
extracting
the
common
parts.
A
B
Yeah
I
mean
I
would
try
to
do
this
cause
as
possible
to
whatever
you're
proposing
I
mean
I'm
still
trying
to
evaluate
how
how
it
could
look
like
I
already
changed
some
namings
there
and
refactor
it
it's
a
little
bit.
No,
but
yeah
I
mean
the
idea.
Is
that
so
we
end
up
with
a
easy
way
to
create
Services
Inc?
Oh
okay,
not
only
services,
network
services
and
ten
points,
but
also
clients,
and
this
should
be
really
too
trivial.
B
So
whatever
Fred
was
showing
on
the
stage
Capcom,
also
on
VPP
conference,
there
should
be
really
trivial,
I
mean
it
should
be
really
easy
to
do
it
after
the
SDK
gets,
gets
worst
I
strongly
suggest
and
would
use
any
command.
So
if,
if
anyone
can
anyone
interested
to
please
just
see
the
pr
is
sick
so
for
the
poor
request,
I
probably
need
to
change
his
naming,
because
it's
what
it's
not
really
great.
B
E
F
E
Here
so
I'm
going
to
apologize,
if
you
haven't
put
documentation
in
there
for
everything
you'd
like
to
document
and
at
least
make
sure
there's
some
placeholder
documents,
just
read
these
or
in
you
know,
installs
or
how
to
use
documents.
They
don't
have
to
contain
anything.
But
if
they're
there
then
at
least
everybody's
clear
that
they're
missing.
If
they're,
not
there,
people
don't
notice
that
until
too
late
and
then
anyone.
F
B
E
As
I
say,
it's
mostly
it's
not
so
important
that
you
write
the
documentation,
it's
more
important,
that
you
point
out
what's
missing,
so
that
the
next
person
and
who
comes
along
to
edit,
the
files
can
knows
what
they
need
to
add.
If
you
don't
stick
any
sort
of
clues
that
there's
missing
documentation
there,
then
anyone
who's
doing
a
bit
code.
Editing
won't
won't
add
things.
So
so
all
I'm
saying
is
leave
leave
space
for
it.
You
know
make
it
obvious
that
it's
missing
look.
B
B
My
selector
unit
tests
yesterday
I
will
try
to
what
more
I
think
that
we
need
this,
and
actually
my
idea
was
to
propose
to
focus
our
main
efforts
there
for
a
week,
two
or
even
a
month,
and
not
kind
of
focus
on
new
things
before
we
get
some
at
least
some
unit
test
coverage,
because
if
you
run
make
tests
today,
you
will
see
that
it's
kind
of
not
really
existing
I
mean
yeah
just
to
fill
them,
but
yeah.
The
more
we
get.
B
A
Looking
to
get
involved,
basically,
writing
testing
writing.
Documentation
are
always
good
sort
of
first
efforts
as
you're
getting
involved.
You
learn
a
hell
of
a
lot
about
a
system
when
you
sit
down
to
try
and
write
some
unit
tests
for
pieces
of
it
because
you're
the
first
step
with
writing
unit
tests
is
of
what
actually
is
supposed
to
be
happening
here.
E
F
Noticed
that
there
is
a
open
issue,
Isabelle
trying
to
do
some
documentation
for
the
ICMP
examples
for
cos,
I'm
a
newcomer.
So
it
will
be
really
really
looking
forward
for
those
kind
of
documentation
to
try
to
get
some
hands-on.
Any
examples
to
that
I
can
try
to
follow
up
and
to
catch
up
the
rhythm.
F
D
Yeah,
you
know,
take
it
a
step
further
as
well.
If
you're
a
CNF
provider,
then
having
something
as
well
on
the
other
side,
where
you
can,
you
can
effectively
like
if
your
CNF
only
works
with
kernel
interfaces,
and
you
specify
that
as
a
config,
then
that's
what
you
get
or
if
you
support
mif,
or
you
support
your
host
user
or
so
on.
D
Like
all
of
this,
all
of
these
preliminary
things
to
set
up
the
initial
connection
and
shared
memory
and
all
that
kind
of
stuff
having
something
that
deals
with
all
that,
so
you
can
just
focus
purely
on
your
on
your
logic.
I
think
it's
going
to
be
very
powerful.
So
that's
we're
trying
to
make
we're
trying
to
make
that
as
easy
as
easy
as
possible.
B
A
First,
with
the
existing
favourite
pieces
right,
so
he
submitted
a
patch
577
trying
to
get
bigger
and
working
with
Lewbert
in
the
course
of
this,
there's
been
sort
of
a
lot
of
back-and-forth
trying
to
make
this
work
and
not
break
other
things.
So,
for
example,
one
of
the
things
he
was
that
to
make
this
work
he
needed
to
use
sshfs
as
the
HFS
doesn't
work
with
the
vmware
provider
for
vagrant.
So
there
was
a
lot
of
sort
of
dancing
around
these
things.
A
He
got
all
this
working
with
a
Debbie,
9
box
and
then
I
sort
of
suggested.
Could
we
make
this
work
with
1804
box?
He
was
sort
of
poking
around
that
you
know
and
I
think
probably
for
the
immediate
set
of
needs
which
is
trying
to
prevent
myth
you
from
having
to
rebase
this
stupid
patch.
Every
time
he
moves.
We
can
probably
just
make
the
Box
choice
dependent
on
the
provider
and
that
will
probably
work
now.
A
There's
a
broader
conversation
here
about
what
a
smarter
medium
to
longer
term
plan
would
be,
which
I
think
gets
into
the
set
of
stuff
that
was
through
the
ian
was
chatting
about
against
that
are
on
terraform
and
bloody,
bloody
blah
and
I.
Think
that's
all
a
good
idea.
I
think
doing
that
to
the
exclusion
of
solving
that's
using
be
a
problem
would
be
an
example
of
letting
the
perfect.
E
So
I
mean
I
when
I
was
looking
at
this,
then
there
are
certain
things
that
they've
been
quite
good
at,
including
making
sure
that
you're,
basically
using
your
local
copy
of
code
in
in
all
of
your
favor
and
boxers,
hence
sshfs
among
other
things,
but
it's
the
problem
you're
going
to
run
into
there.
Is
that
vagrant?
Quite
well,
it's
two
things:
it's
very
opinionated
and
it's
quite
limited
tariff,
and
and
honestly
it's
providers
are
very
weird,
as
in
no
to
providers
look
very
similar.
E
So
you
end
up
rewriting
a
lot
code,
just
to
kind
of
get
the
next
provider
to
work,
which
is
I.
Think
where
a
chunk
of
where
the
the
current
change
is
going
and
I'm,
absolutely
not
saying
we
should
stop
doing
that.
We
should
start
working
on
terraform
instead
I'm
saying
that
I
think
a
bit
a
development
work
on
terraform
might
obsolete
this
in
the
longer
term,
and
that
would
be
worth
doing
so.
E
So
that's
what
I'm
saying
why
I'm
saying
we
should
focus
on
the
terraform
stuff
as
well,
which
would
be
quite
nice,
because
we've
also
got
the
packet
net
code,
which
is
separate
again.
It's
a
third
back-end
with
its
own
set
of
scripts
and
I'm,
wondering
whether
that
could
be
terraformed
at
some
point
in
the
future
requirement.
D
So
the
own
problem,
or
perhaps
perhaps
there's
a
library
out
there,
which
solves
this,
but
there's
no
terraform,
kubernetes
installer,
so
I
have
to
install
kubernetes
using
a
set
of
scripts
in
order
to
in
order
to
make
it
work.
So
if
terraform
had
something
that
would
install
a
cluster
for
you,
that
may
be
a
lot
easier
to
to
just
use
terraform
for
the
entire
lot
yeah.
E
But
you
I
mean
again
you've
already
done
that
so
and
you've
done
it
in
terraform
as
well,
so
so
one
platform
or
another.
So
the
question
is:
can
we
basically
recycle
that
for
every
platform,
because
once
you
brought
you
know
the
thing
about
terraform,
that's
specific
to
terraform
is
bringing
the
virtual
machines
or
the
house
SERP
initially,
but
once
they're
up,
then
every
single
bit
of
terraform
should
be
the
same
for
all
platforms.
So
that
was
the
thing
that
I
thought
was
exciting
that
we've
you
write
it
once
and
every
time
someone
fixes
it.
E
E
An
ideal
world,
that's
what
I'd
like
to
see.
Yeah
I
mean
I'm,
not
sure
how
feasible
that
is
for
vagrant
itself.
Annoying
because
I
know
that's
where
we
start
and
well
most
people
have,
but
let's
just
see,
what's
possible,
I
mean
theoretically
vagrant
you
can
SSH
into
a
vagrant
box,
so
you
should
be
able
to
make
the
terraform
work
and
I'd
be
amazed
if
someone
haven't
tried
testing
their
terraform
scripts
against
a
current
boxes.
So
there's
a
bit
of
Investigation,
hey.
C
Ian
I'd
be
willing
to
explore
this
with
you,
we're
still
kind
of
battling
legal
internally,
but
your
favorite
guy,
Brandon
and
I
have
been
working
on
some
stuff
exactly
in
this
space,
with
like
around
terraform
and
deploying
kubernetes
and
things
so
I,
don't
know
when
we're
gonna
be
allowed
to
share
it,
we're
trying
to
hope
we
get
it
in
this
first
quarter,
but
if
we
do
get
the
green
light
to
open
source,
that
I
think
we
could
leverage
a
lot
of
that
codes.
If.
E
You
want
a
conversation
offline,
then
we
can
see
it
rather
than
sharing
your
code.
You
can't
least
tell
me
what
your
thoughts
are
and
maybe
I
can
steal
your
thoughts,
which
are
a
little
bit
less
than
legally
encumbered.
Shall
we
say,
but
we
we
can
figure
something
out.
One
way
or
the
other
I
mean
it
just
feels
to
me:
I
mean
I'm.
Firstly,
I'm
sure
people
have
destroyed
keeping,
as
you
said,
terraform,
but
secondly,
quite
honestly
for
the
purposes
we
have
then
there's
nothing
fundamentally
wrong
with
what
Frederick's
doing
so.
E
I,
don't
think
we're
I
think
this
is
in
reaching
distance
I
just
don't
know
quite
how
long
it
would
take
to
get
it
to
the
point
that
you'd
want
to
develop
that
way.
But
if
we
could
get
this
running
on,
you
know
AWS,
to
take
one
example
and
find
you
well
I
won't
say
you
can't
have
SR
iov
in
AWS,
but
it's
not
exactly
easy
to
work
with
and
it
gets
bloody
expensive
very
quickly.
E
E
Post,
what
I've
got,
which
is
frankly
embarrassing,
but
with
the
heading
off
in
the
right
end
and
I,
think
it's
more
theory
than
practice.
I,
don't
know
how
far
tested
I
got
but
I'll
stick
it
up
on
my
github
and,
and
you
know,
for
the
purposes
of
making
it
visible
to
people
I'll
pull
request
in
and
then
you
can
see
where
I
was
going
with
it.
E
E
G
E
Fine,
because
that's
only
one
little
bit
of
the
code
that
I
have
to
change
I,
just
love
to
see
your
examples,
because
we
should
be
well
two
things
right,
I'm,
not
very
well
educated
and
terraform.
So
reading
law
form
helps
me
understand
how
to
write
good
terraform,
but
but
the
other
one
is
that
it's
there's,
probably
things
ideas
that
we
can
reuse
so
for
what
we're
doing
so
awesome.
D
B
A
Even
just
collecting
the
list
of
things
that
we
could
be
working
towards
is
a
good
first
start,
and
then
we
sort
out
what
are
the
orders
and
what
we
want
to
try
and
achieve
by
when
so,
for
example,
you
know
some
of
the
things
that
have
gone
by
I
know:
we've
got
you
know,
folks,
working
right
now
on
the
robustness
story.
In
other
words,
me.
E
A
Back
correctly,
when
various
components
of
the
system,
or
at
least
as
correctly
as
can
be
done
and
part
of
this
ends
up
being
some
of
the
stuff
with
auto
healing,
because
it
turns
out
that
auto
healing
just
sort
of
falls
out
when
you
make
this
system
more
robust
or
you
would
actually
almost
have
to
do
effort
not
to
and
then
so
that's
one
piece
and
I
think
that's
that's
kind
of
critical.
Then
do
we
want
to
take
notes
on
some
of
this,
so
robustness
story,
auto
healing?
As
a
part
of
that,
then
we've
got
inter-domain.
A
I
A
A
Yeah,
that's
definitely-
and
this
is
this-
is
basically
this
is
not
so
much
I'm
a
developer.
I
can
debug
things,
it's
more
of
a
I,
have
a
production
system
and
something
went
wrong.
What
the
right
I
know
Jeffery
has
no
idea.
What
that's
like
so
definitely
definitely
open
tracing
is
something
there
and
I
there's
some
work
underway.
A
The
other
thing
that
comes
immediately
to
mind
is
IO
am
because
I
think
we
may
be
able
to
use
IOM
to
do
open
tracing
style
things
at
l2
and
l3.
So
we
could
actually
do
some
of
that,
because
you're
gonna
have,
as
the
as
people
do
what's
gonna
come,
naturally,
which
is
vulcanizing
to
more
more
micro
services,
you're
going
to
want
to
know
where
things
went
so
you've
got
that
for
the
bucket
ability,
deployment
there's
also
security
and
I
tend
to
think
of
this
in
terms
of
who
properly
authenticate
and
authorize.
B
E
A
E
But
that
might
be
physical
Ethernet
and
therefore
you
might
be
right
down
the
down
to
you
know
down
to
the
wire
at
that
point
and
I
think
we've
had
this.
Thank
you
for
saying
that
hardware
support
so
not
just
the
NIC,
though
that's
my
point.
It
can
also
be
the
thing
that
the
Nick's
connected
to,
if
you're,
putting
a
data
plane
together
and
the
data
plane
does
the
excellent
that's
great.
All
you
need
is
layer.
E
3
connectivity
now
is
layer,
3
connectivity,
a
thing
you
have
normally
yes,
but
if
it's
MPLS
you
need
more
than
just
layer
3
connectivity.
So
we
can't
pretend
that
the
hardware
isn't
a
network
service
or
might
not
be
a
network
service
in
its
own
right
that
other
network
services
consume.
We've
got
to
remember
that
that
everything
is
it's
got
features,
and
maybe
it's
got
all
the
features
we're
looking
for,
and
maybe
it
hasn't,
and
maybe
we
should
be
asking
before
we
make
assumptions.
A
A
So
right
now
the
network
service,
when
you
specify
matching
the
network
service,
the
only
verb
that
we
have
is
route
which
basically
says.
Okay.
If
you
match
this,
please
brought
me
to
notification
that
has
Kirk
that
has
labels
and
that's
a
super
powerful,
so
powerful
verb.
There
are
two
other
verbs
that
we
sort
of
hummed
a
few
bars
about
in
PRS,
right
verb,
which
is
basically
saying
okay.
A
E
A
D
E
C
Requests,
sir,
can
instead
of
just
strictly
SRV
six,
could
we
just
look
at
segment
routing
holistically,
because
different
parts
of
my
convoluted
network
are
going
to
implement
SR
in
different
ways?
So
that's
why
I
ask
for
just
straight
MPLS.
Do
is
I
would
be
very
interested
in
not
just
as
our
v6
but
also
SR
MPLS,
if
possible
and.
E
An
MPLS
is
a
multitude
of
sins
as
well,
because
you've
got
a
data
control
plane
that
goes
with
it,
which
varies,
but
it's.
The
point
is
that
we
need
something
which
is
an
abstract
description
of
a
connection
and
the
more
examples
we
can
throw
at
it.
The
more
sure
we
can
be
we've
got
to
the
abstract
and
not.
We
haven't
made
assumptions
about
the
data
plane
that
we're
talking
about
and.
E
D
We're
we're
out
of
time,
but
what
I
recommend
is
we
add
this
to
the
to
the
next
week's
agenda
we
can
discuss
what
this
really
means,
I
think
yeah,
I
think
the
short
answer
is
going
to
be
that
well,
I
personally,
think
it's
a
good
idea
and
with
that,
thank
you
for
attending
the
year's
first
meeting.
The
next
meeting
is
going
to
be
at
the
same
time
week
from
now.
So
thank
you
all
and
catch
the
next
week.