►
From YouTube: Network Service Mesh WG Meeting - 2018-08-24
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
B
B
A
B
Someone
has
their
their
mic
on
and
do
some
background
voices
cool
anyways.
So
we
have
the
open
the
cloud
native,
never
conscious,
seminar,
that's
next
Tuesday
during
the
afternoon
in
Vancouver.
If
you
are
going
to
be
there,
you
haven't
registered
now's
the
time
to
do
it.
You
can
go
on
to
the
open
source
summit's
website
and
they
have
a
way
for
you
to
add
a
registration.
Even
if
you've,
even
if
you've
already
gone
to
the
process
of
registering
you,
can
you
can
update
the
registration
a
couple,
a
couple
other
things
on
that?
B
C
I'll
speak
on
their
behalf.
I've
got
my
mute
button
on
the
ready,
so
we're
going
to
do
an
overview
of
the
cross
cloud
CI
and
how
we've
implemented
the
own
app
project
to
the
dashboard
and
yeah
that'll,
be
just
a
quick
20
minutes
on
that
and
I.
Don't
know
that
we're
going
to
talk
too
much
about
our
progress
on
the
CNF
s--,
so
it'll
be
mostly
about
cross
cloud
testing
system
and
dashboard.
D
For
20
minutes
on
network
service
mesh
at
that
seminar,
so
I
will
be
getting
some
slides
together.
It
turns
out
to
be
remarkably
hard
to
condense
things
into
20
minutes.
So
that's
always
interesting.
The
other
thing
that
I'm
kind
of
bouncing
around
is
the
idea
of
possibly
doing
and
I
said
happy
hour
at
a
nearby
bar
and
I
still
need
to
go.
Take
a
look
and
see
if
there
are
some
conflicting
receptions
that
may
be
happening,
Tuesday,
night
or
whatnot.
But
you
know,
if
not
we'll,
probably
go
ahead
and
give
up.
D
You
know
ultimately
mention
it
as
part
of
the
slide
deck,
we'll
probably
get
it
up
on
the
landing
page.
We're
going
to
talk
about
it'll,
hear
it
a
little
bit
et
cetera,
because
I
think
it'd
be
good
to
get
people
together
just
to
visit
and
talk,
especially
since
people
seem
very
excited
and
have
a
lot
of
questions
and
20
minute.
Talk
at
a
seminar
does
not
really
mesh
well
with
that.
B
Okay,
so
yeah
it'll
be
good
to
what
to
work
out.
Yeah
I'm
I'm,
pretty
sure
that
the
the
conference
is
scheduled,
something
for
every
night,
but
my
understanding
is
that
they
limited
the
end
of
the
invites
based
on
various
criteria.
So
like
I,
think
Thursday
was
the
partner
and
speaker
recept,
end
and
speaker.
Reception,
Tuesday
I
think
was
a
diversity
dinner.
B
B
D
B
Yeah
I'm
I'm
happy
to
see
to
have
to
cancel
for
next
weekend.
I,
don't
think
said
well,
I
think
what's
gonna
end
up
happening
is
that
there
will
be
people
are
interested
in
network
service,
mesh
I
think
if
we
hold
a
meeting
on
Friday,
even
if
everyone
was
not
on
an
airplane
and
we
had
the
time
to
take
it
I
don't
think
we
have
enough
time
to
prepare
to
help
to
help
bond
board
and
get
people
on
to
get
people
on.
B
E
I,
just
I
actually
stumbled
upon
that
I
wasn't
even
looking
for
that.
I
forget
what
I
was
doing,
that.
Let
me
down
that
path,
but
when
I
saw
that
I
thought
that's
pretty
interesting
and-
and
you
know
in
my
quest
to
automate
all
the
things
I
thought
you
know-
let's
automate
this
as
well
and
and
it
actually.
C
E
B
E
Hold
on
one
sec,
let
me
just
verify
that,
oh
it,
you
know
what
it
isn't,
but,
but
in
fact,
as
soon
as
this
meeting
is
done,
I'll
push
a
PR
that
does
that
because
I
I
added
it
directly
to
the
Travis
yeah
mo,
but
then
I
went
and
added
it
to
a
bunch
of
other
legato
projects
as
well,
and
for
those
I
put
it
in
the
make
file,
so
I
will
I
will
I
will
push
it.
So
it's
in
our
Meg
file
as
well.
So
when
you
build
it
does
automatically.
E
B
So
if
we,
if
we
can
add
a
link
to
the
agenda
in
Kyle's,
we
can
add
a
link
to
the
to
the
agenda
about
that.
That'll
that'll
be
good,
and
that
leaves
us
to
the
last
part
which
is
building
the
NSM
website
using
hugo,
and
now
we
actually
have
an
agenda
item
for
that,
so
we'll
hold
off
for
that
one
for
the
agenda.
It
might
actually
be
next.
B
It
is
next,
so
this
jump
directly
into
the
into
the
OSS
preparations.
So
we
have
a
huge
amount
of
work.
That's
gone
into
the
landing
page,
so
if
you
haven't
seen
it
yet
go
to
the
network
service
mesh
on
IO,
it's
it's
it's
linked
on
the
agenda,
so
you
can
click
on
that
and
and
take
a
look
at
it
and
it
would
be
really
really
helpful
like.
E
It
definitely
is
I
was
super
happy
with
this,
because
because,
basically,
you
know
basically
getting
this
deployed
to
a
global
CDN
basically
took
me
a
few
hours
which
is
you
know.
2018
is
amazing
in
certain
ways,
because
because
that
was
awesome
and
I
love
the
fact
that
it's
that
it's
Hugo
and
we're
seeing
it
and
then
it
just
automatically
deploys
to
never
buy
once
we
merge
them
back
so
so
super
cool.
It.
D
B
D
So
I
think
probably
a
good
review
on
the
concepts
would
be
very
helpful
and
the
more
people
who
try
the
getting
started,
the
better
the
getting
started
is
a
little
bit
sparse
right
now
right
now
it
just
tells
you
how
to
deploy
a
network
service
mesh
to
your
existing
kubernetes
cluster.
We
need
to
get
through
and
sort
of.
Also,
you
show
you
how
to
pry
out
the
test
case,
but
the
other
that
that's
but
that's
kind
of
where
we
stand
right
now.
D
A
It's
I
I
started
that,
and
there
are
there
is
some
fuzziness
there's
some
steps
that
are
not
absolutely
clear,
so
I
I'd
be
willing
to
to
try
to
see
if
I
can
write
that,
just
as
I
try
to
figure
it
out
and
actually
do
it,
because
I
want
to
write
a
more
complex
data,
plane,
NSC
and
I
thought
by
starting
with
Serge's
and
trying
to
make
that
work.
It
would
give
me
a
start
to
writing
another
one.
D
B
We
have
a
limited,
limited
time
because
choose
Tuesdays
I
think
this
is
the
goal.
So
if
we
were
to
start
up
a
Google
Doc
and
what
we
can
do
is
is
you
know
we
start
without
lying,
what
we
what
we
have
and
if
somebody
does
some
work
on
it
and
gets
some
progress,
stick
the
commands
in
and
inside
of
the
Google
Doc
and
then
the
next
person
who
hops
on
can
reek
and
look
at
that
and
then
fill
out
the
next
parts
or
clarify,
and
that
way
that
we
we
don't.
A
B
D
Because
I'm
a
poke
at
this
problem
as
well
so
yeah
this
is
this
is
something
I
would
love
to
see
if
we
can
get
pulled
together
or
something
very
clear
and
simple
for
Tuesday,
because
that
will
be
a
lot
of
people's
first
exposure
and
I.
Think
the
existing
QuickStart
is
very
clear
and
simple
as
far
as
it
goes,
it
just
doesn't
go
very
far
right.
It's
the
run
this
one
command
and
congratulations.
You
are
now
network
service
mesh,
enabled
in
your
cluster,
okay
and
now
what
yeah.
F
B
D
I
think
we
may
have
just
covered
that
I
mean
I,
think
we're
getting
a
pretty
good
set
of
stuff
for
the
documentation.
There's
actually
one
thing
that
you
could
use
an
opinion
on
so
when
I
put
together
the
documentation
page
right
now
it
loads
up
concepts
first.
So
if
you
just
click
on
the
docs
link,
you'll
be
taken
to
concepts.
I
do
sometimes
wonder
if
maybe
documentation
shouldn't
land
you
in
setup
and
the
reasons
we
shouldn't
switch.
The
order
so
that
set
up
leaves
before
concepts.
I
would
very
much
welcome
opinions
on
that.
I.
D
B
A
G
B
A
F
D
D
D
D
B
B
Also
I
had
one
other
thing:
I
was
on
mute.
So
what,
in
terms
of
the
in
terms
of
the
the
OSS
preparation,
so
I
have
this
little
web
application
that
I've
been
working
on
that
allows
people
to
type
in
questions
and
what
I.
So
this
is
based
on
what
I
proposed
earlier
that
we
have
something
that
we
can
type
questions
in
and
they
just
pop
up
on
the
website.
B
So
it's
not
designed
to
be
permanent.
It's
just
for
for
running
during
events.
So
so
I'm
gonna
finish
up
the
application
today
and
we'll
and
we'll
see
about
finding
a
way
to
to
integrate
it
nicely
so
that
that
it
works
so
I
don't
have
a
link
to
it.
Yet
I
haven't
pushed
it
up
to
github
yet,
but
I
will
I
will
push
it
up
soon.
So,
and
one
of
the
nice
things
about
it
is
that
when
you
do
the
answer
like
it
literally
will
just
show
up
on
everyone
else's
page.
B
B
Let's
move
on
to
the
draft
X
Factor
CNF,
so
I
created
a
I.
Just
look
like
I
posted
the
that
gets
done
here,
so
I'll
post
it
right
now
so,
but
basically
I
created
a
a
guest
that
has
information
on
what
I'm
thinking
of
for
the
X
Factor
for
the
expector
CNF
s--,
and
you
have
the
link
here.
I
apologize,
we're
not
happening
up
right
now.
B
There
we
go
and
so
I'm
starting
with
with
this,
and
the
idea
is
not
to
say
this
is
this
is
what
one
of
these
things
like?
How
do
you
do
one
of
these?
So
the
idea
is
to
try
to
set
the
agenda
as
to
what
I
think
the
community
should
should
rally
around
and
specifically
I
so
I.
So
there's
a
diversity
of
bullet
points.
B
You
can
see
it
I
think
the
toe
factor
apps
are
aiming
more
towards
making
it
easy
for
developers
to
to
build
the
applications
to
maximize
the
portability
of
the
of
the
applications
to
be
able
to
deploy
on
modern
clouds.
Minimize
the
divergence
between
development
tests
and
products
then
to
provide
horizontal
scaling
and
I
think
that
we
can
start
with
those
properties
as
well.
B
You
know,
and
and
and
learn
from
them
and
they're
very
high
level.
I
put
some
stuff.
There
were
a
little
bit
more
that
were
a
little
bit
more
specific
on
the
CNF,
so
things
like
no
vnf
should
have
specific
kernel
modules,
and
so,
if
you're
right,
if
you're
rewriting
of
the
NFL,
rename
that
to
see
enough
so
as
you're
moving
from
vnf
to
see
em
f's
many
VMs
have
specific
kernel
modules
and
if
you
have
to
deploy
a
kernel,
module
you're
breaking
out
of
the
isolation.
B
So
that's
just
as
an
example
and
so
to
try
to
work
out
like
how
do
we?
How
do
we
move
people
from
vnf
style
thinking
to
do
CMS
out
thinking
another
challenge
of
running
into
right
now
and
if
I
think
the
Volk
people
are
definitely
on?
This
is
like
even
defining
what
a
CNF
means
like
what
is
a
what
is
a
CNF
and
trying
to
try
to
define
and
say
this
is
this?
Is
a
CNF?
This
isn't
a
CNF
I
think
this
is
going
to
continue
to
evolve
for
a
while.
B
F
D
No
I
think
you're,
actually
I,
think
you're
right
in
the
global
sense,
but
from
the
perspective
there
are
perspectives
from
which
they
are
endpoints,
because
from
the
perspective
of
so
from
the
perspective
of
some
consumer
that
wants
to
connect
to
a
CNF
that
network
service
is
an
in-point
right.
All
it
knows
is
it
shows
pack
us
back
and
forth
from
the
global
perspective,
you're
absolutely
right.
A
Yeah
I
think
that,
from
my
perspective
and
tell
me
if
I'm
wrong,
I
think
when
we
talk
about
the
NS,
a
network
service,
endpoint
we're
talking
about
a
control,
plane
endpoint.
It's
saying
that
the
final
thing
we
want
configured
happens
there,
but
but
it
could
be
an
endpoint
that,
for
existent,
for
example,
is
a
is
a
firewall
or
a
router.
So
therefore
a
try
packet
so
still
throw
flow
through
it.
Isn't
that
isn't
that
correct
from
so?
In
other
words,
it
would
be
a
control,
plane,
endpoint,
but
not
a
date.
B
Yeah
and
I
have
a
tendency
to
to
also
think
well
not
when
I
was
writing
this
one
things
that
occurred
to
me
was
that
we
want
to
be
very
distinct
as
well,
but
there's
there's
also
a
relationship
to
service
a
service
function,
chaining
on
in
cloud
native
environments
and
so
so
I
think
part
of
part
of
how
I'm
going
to
try
to
approach.
This
is
about
not
just
about
how
do
you
scale
the
C
and
F,
but
how
do
you
scale
the
service
function
chain
itself
and.
F
B
And
as
and
they
and
they
can
grow
independently,
like
maybe
one
is
very
computationally
expensive
and
it
has
to
grow
with
computation.
Maybe
number
two
is
very
data
or
very
throughput
intensive
and
that
has
to
grow
parallel
to
land,
you
more
Nix
and
landing
landing.
One
of
these
on
a
system
that
already
has
where
you
already
have
one
on
it
is
not
gonna
actually
gonna.
Does
it
work?
Yes?
B
A
A
B
I'll,
try
I'll,
try,
answering
and
tell
me
if
I
hit,
if
I
hit
your
question
so
in
terms
of
in
terms
of
endpoints.
Like
yes,
there's,
there's
endpoints.
Yes,
we
have
two
data
flow
through
them,
but
I,
but
I
would
also
argue
that
every
CNS
itself
is
also
is
also
an
end
points
like
we
looked
at
at
least
from
the
these
from
the
point
of
perspective
from
from
the
from
NSM,
maybe
other
SF
service
function,
chaining
and
control
planes
don't
view
it
in
this
particular
way.
B
You
know
if
you
go
through
the
how
our
C
or
D
was
developed.
You
know
we
have
a
source
and
a
destination,
that's
defined
for
each
for
each
node,
and
so
when
you
have
a
firewall
the
chain
the
chain
is
this.
This
is
like
in
the
VPN
gateway
case.
You
have
corporate
Internet
connectivity
and
inside
of
that
inside
of
that
service,
when
you,
when
the
first
connection,
is
if
you're
not
part
of
the
service
you
go
to
the
firewall
and
then
the
second
one
is.
B
If
you
are
part
of
the
service-
and
you
are
the
firewall,
then
you're
connected
to
the
Gateway,
so
the
Gateway
becomes
the
end
point
for
the
firewall
and
this
changed
it
just
so
in
essence,
it's
the
same
pattern
over
and
over
and
over
again
and
and
there's
they
all
act,
each
each
section
in
the
next
change
acts
as
the
endpoint
for
the
for
the
previous
one.
So.
F
G
D
F
D
And
so
that's
I
think
a
little
distinction,
one
of
the
things
that
I
think
we
probably
want
to
also
focus
on
and
I
find
this
very
useful
when
I
think
about
cloud
native
about
CMS.
One
of
the
things
that's
in
the
cloud
native
definition
is
something
about
immutable
infrastructure
and
and
I
think
this
concept
of
immutable
infrastructure,
together
with
the
notion
of
what
is
on
what
side
of
the
infrastructure
is
a
huge,
be
powerful
way
to
think
about
those.
D
So,
for
example,
for
cloud
native,
the
kernel
is
part
of
the
infrastructure
and
therefore
the
kernel
is
immutable
right,
and
so
that's
why
your
CNF
can't
have
a
kernel
module,
but
there
there
may
be
other
kinds
of
things
as
we
explore
here
that
fall
out.
Naturally,
from
this,
in
terms
of
you
know,
immutable
infrastructure
and
where
the
line
of
demarcation
and
infrastructure
for
CNX
yeah.
B
And
that's
a
really
great
point
and
I.
Think
configuring,
the
kernel
through
the
device
plug-in
API
that
in
a
controlled
manner,
is
acceptable
but
saying
we're
gonna
we're
going
to
shove
a
kernel
module
in
that's
going
to
affect
others.
There
see
NFS
like
you
could
still.
You
can
there's
still
use
cases
like
these
are
sort
of
heuristics.
If
you
absolutely
need
to
do
it,
no
one's
going
to
stop
you
from
doing
it,
but
you
lose.
You
lose
something
for
it.
B
Definitely
agree,
and
so
that's
you
know,
and
and
so
that's
not
saying
that
was
like
us-
it's
like
a
set
of
guidelines
and
in
heuristics,
but
I
think
that
yeah
I
think
that's
a
that's.
A
really
great
way
to
look
at
it
as
well
is
to
try
to
define
the
the
borders
and
see
yeah
one
of
the
things
that
I
was
that
I.
That
I
realized
was
the
orders
of
magnitudes
more
complexity.
That's
here
compared
to
the
web
application
version.
B
So
when
12
factors
apps
comes
out
and
Adam
was
building
it
out,
I
mean
he
had
seen,
probably
10,000
or
100,000
different
deployments,
and
they
probably
worked
with
with
customers
to
help
resolve
scalability
issues
with
them
and
again
how
to
configure
and
manage
them.
And
one
of
the
things
I
think
we're
going
to
run
into
here
is
that
the
diversity
of
the
types
of
deployments
that
we're
going
to
see
is
gonna
be,
is
going
to
be
significantly.
B
B
This
is
not
being
like,
like
Tuesday
is
not
like
the
the
end
point
for
this,
like
I
love
to
be
able
to
shop
it
around
on
Tuesday
and
get
more
people
involved,
but
I
think
this
is
actually
something
that
is
gonna,
be
like
ongoing.
So
as
we
continue
to
work
with
network
service
mesh
and
we
people
are
gonna,
ask
us
for
guidance,
we
can
give
them
as
we
can
give
them
these.
B
The
point
I
want
to
try
to
drive
with
to
try
to
drive
with
people
is
that
they
need
to
start
looking
at
their
their
cloud
native
infrastructure
as
as
having
specific
properties
that
if
they
follow
certain
patterns,
they
will
gain
the
benefits
about
native,
and
so,
which
means
don't
don't
rely
on
scaling
up.
Don't
rely
on
kernel
modules.
B
Do
rely
on
on
scaling,
horizontal,
do
rely
on
being
very
explicit
and
declarative
in
your
configurations
and
defining
your
capabilities
and
how
you
can
communicate
how
you,
what
payloads
you
accept
and
yeah-
and
you
know,
and
also
from
the
operational
perspective
like
it's
very
one
of
the
things
that
they
have
in
the
12
factor
apps.
That
is
a
discussions
I've
had
with
with
with
others,
was
about
it's
like
were
to
keep
logging.
B
You
treat
your
logs
as
events
streams,
and
so,
whenever
you
have
a
logs
you,
you
inject
it
into
an
event
stream
that
you
have
something
like
fluent
Dior
or
flu,
or
something
similar,
that's
capable
of
capturing
those
those
logs
and
aggregating
together,
and
then
do
you
have
another
tool
that
can
be
used
to
to
gain
insight
on
your
entire
distributed
system.
Your.
C
B
Cluster
and
so
like
these
are
these
are
patterns
that,
if
you're
working
primarily
with
a
single
on
this
iam
system,
then
you
won't
be
familiar
with
these
type
of
patterns.
But
if
you
worked
on
distributed
systems,
they're
extremely
common
and
so
there's
also
an
opportunity
to
help
as
we
move
towards
more
more
distributed
paths
to
help
bring
some
of
these
well
known
concepts
from
other
from
from
other
areas.
So.
B
B
C
G
Now
they
don't
have
any
nx5
Nick's
that
are
available
for
other
folks.
They
do
have
some
systems
within
till
next,
so
that's
an
option
working
on
getting
access
to
more
of
those
building
out
the
specs.
That
would
be
useful
for
the
various
project
scene,
a
scene
of
comparison
and
an
innocent
and
and
then
there's
been
a
lot
of
discussions.
Edie
is
aware
of
on
the
CNX
for
support
with
VPP
I
think
a
lot
of
that's
coming
forward.
So
if
we
have
that,
then
I
think
we'll
be
able
to
do
more
testing
yeah.
D
See
nx4
from
from
what
Michael
was
saying
if
I
understood
correctly,
cx4
should
work
with
VPP
18
with
EDD
1804.
There
is
some
thing
odd
about
the
PDK
drivers,
the
more
recent
DP
decay
drivers
for
Mellanox
things
in
general.
That's
currently
being
discussed
and
sorted
out
so
but
I
do
believe
see
an
explore
should
work
with
VPP,
but
not
with
the
latest
1807,
because
it's
using
the
latest
DP
D
K
and
the
latest
DP
D
K
drivers
appear
to
be
broken,
but
it
does
work
with
1804.
D
G
Yeah
yeah
he's.
We
talked
a
lot
of
this
week
about
it,
and
so
once
I'm
gonna
clean
up
on
the
comparison
code
is
theirs
and
we're
gonna
update
some
of
the
docs.
That's
on
the
CNS
CNCs
CNS
project,
we're
gonna
all
that's
available.
What
can
roll
something
in
with
maybe
a
shorter
dock
for
the
innocent
project.
B
G
Think
right
now,
most
of
its
waiting
on
Mike
to
finish
his
testing
and
then,
if
we
hear
anything
back
from
Mellanox
on
any
specific
items
or
working
out
where
the
the
problem
is
on
the
DPP
code,
where
we're
having
to
go
specific
versions.
So
that's
kind
of
outside
of
our
hands
right
now,
right
now,
I
think
we're
good,
no
help
needed.
Okay,.
G
B
D
D
B
And
just
a
little
note
that
if
you
look
at
the
at
at
the
narratives,
you
see
that
there's
a
little
bit
of
a
I
guess
you'd
call
a
pneumonic
with
it,
so
it
was
like
Sarah
and
a
secure
internet
and
they
have
Hannah
and
the
hardware
Nix
and
so
on.
So
it
makes
a
little
bit
easier
to
to
to
remember
so
when
someone's
talking
about
Sarah,
you
know
what
use
case
it
is
so
it's
talking
about
Hannah.
You
know
what
use
keisters
exactly.
D
D
You
know
we
have
our
character,
Hannah
she's,
writing
in
communities
out
to
be
deployed
on
her
on-prem
to
rid
of
these
cluster,
and
one
of
the
pods
needs
some
hardware
deck
now
so
sort
of
set
the
stage
people
tend
to
think
about
Hardware
Nick's
the
same
way
they
think
about
GPUs
and
they're.
So
not
right.
So
some
of
the
nodes
in
Hannah's
cluster
has
special
Hardware.
Next
is
some
dump.
It's
also
the
case
that
not
all
the
special
Hardware
Nick's
are
the
same
right.
D
So
not
all
of
them
connect
to
the
same
network,
for
example
right
they
connect
to
a
bunch
of
different
networks,
not
all
the
special
hardware.
Nics
that
connect
the
same
network
have
the
same
speed.
So,
for
example,
some
are
ten
gigs
several
hundred
gig
there
might
be
some
40
gigs
whatever,
so
you
think,
okay
great,
so
we've
got
what
hardware
NIST
networking
this
speed
Ness
and
then,
of
course,
you
also
realize
that
not
all
the
hardware
mix
are
treated
the
same
by
the
network.
D
They
connect
to
either
right,
so
some
Nix
connected
to
what
the
network
may
have
ACLs
applied
to
them.
Some
may
have
cost
guarantees
and-
and
so
you
sort
of
start
putting
down
this
long
list
of
attributes,
you
should
have
discover
that
you
can't
just
go
and
numerate
all
the
attributes,
because
network
engineers
are
creative
and
we
have
created
way
more
attributes
than
you
think
we
have
it's
a
from
his
point
of
view,
though
her
situation
is
very
simple
right
from
her
point
of
view.
D
She
has
a
pod
that
has
to
be
scheduled
just
some
place,
that
has
a
hardware
Nick
and
then
she
needs
to
make
injected
into
that
pot.
So
you
know
getting
serious
definition
of
Hell
having
to
figure
out
which
one
where
you
know
how
do
I
find
out
all
the
nicks
in
the
cluster.
How
do
I
find
out
which
hard
work
makes
provide
the
service
I
need?
How
does
the
scheduling
get
handled?
D
You
know,
how
do
we
dot?
We
avoid
collisions
in
terms
of
the
scheduling
and,
of
course,
then
we
have
Ariane
de
our
friendly
neighborhood
spider,
who
comes
and
sort
of,
introduces
herself
talks
a
little
bit
about
network
service
mesh
and
how
it's
sort
of
like
service
smash
only
for
l2
and
l3,
and
then
yes,
the
chests,
had
to
explain
her
problems
right.
So
Hannah's
like
okay,
it's
really
simple,
I'm
gonna
play
my
pod
or
I
got
the
hardware
interface
that
I
need
injected.
D
It's
a
very
simple
problem,
but
she
doesn't
have
to
think
about
which
one
she
doesn't
wanna.
Think
about
what
note
it's
on.
She
doesn't
want
to
have
to
wind
up
with
a
hardware
Nick
that
doesn't
do
what
she
needs
and
she
still
wants
to
give
her
normal
kubernetes
networking
into
her
pot
right.
So
you
know,
obviously
the
first
question
that
everyone
should
be
asking
is
what
about
the
device
plug-in
API,
and
you
know
it
doesn't
really
quite
get
there,
because
you
know
as
Hanna
points
out.
D
You
know
the
device
plugin
doesn't
get
the
information
it
needs
to
handle
the
network
enos
of
Nick's.
The
C&I
plugin
doesn't
get
the
information
it
needs
to
handle
a
heart
greatness
of
mix,
and
so
you
get
this
disconnect
plus
Hanna
does
need
any
old
hardware
Nick.
She
needs
a
hardware
Nick.
That
does
the
things
she
needs
right.
So
just
saying
give
me
a
hardware
neck
or
give
me
a
40,
gig
hardware
Nick
or
give
me
a
hardware
Nick
plugged
into
VLAN
10.
D
None
of
those
actually
describe
what
it
is
they
kind
of
needs,
because
what
Hanna
really
needs
is
connectivity
to
a
network
service
and
in
here
our
friends
at
Panda.
So
we
certainly
have
a
hot
network
service
master
could
help
right.
So
number
service
mesh
thinks
about
things
in
terms-
and
this
is
the
familiar
intro
to
network
services,
there
were
service,
endpoints
and
connections,
but
here's
the
thing
when
you
actually
look
at
this
in
service
situation.
D
What
you
really
have
is
the
node
with
Hana
spot,
and
you
have
the
hardware
net
and
the
hardware
make
is
really
the
l2
and
l3
connection
and
in
in
Sarah's
situation.
The
network
service
endpoint,
is
really
the
port
on
the
top
of
rack
switch
right
and
that's
really.
What
we're
connecting
Sarah
to
via
the
mechanism
of
a
hardware
net
make
sense.
So
far.
D
Cool
and
then
we
dive
into
how
do
I
use
this,
and
this
is
the
very
familiar
I'm
gonna
skim
through,
because
we're
all
very
familiar
with
this
stuff.
You
know
talking
about
defining
a
network
service,
in
this
case
enhanced
corporate
connectivity
and
that
sort
of
represents
the
class
of
hardware
Nick
that
Hannah
needs
and
then
in
the
pod
spec.
You
simply
put
a
resource
request
for
enhanced
corporate
connectivity
into
your
pods
back
and
then,
of
course,
there's
the
obvious
question
about
the
network
enos
here,
and
this
is
where
the
network
service
manager
comes
in.
D
So
in
Hannah's
pod
gets,
you
know,
dropped
in
with
its
eminent
container.
It
requests
a
connection.
The
NSM
injects
the
hardware
neck
and
since
the
accept-
and
so
from
Hannah's
pods
point
of
view,
it
just
looks
like
the
interface
appears
and
it
works
very
much
like
we're
used
to
things
working
for
network
service,
mash
and,
of
course,
for
the
scheduling
and
everything
else
that
all
works,
like
normal
device,
plug-in
mechanisms
exactly.
D
D
A
D
Like
it,
I
mean
in
particular
the
one
slide:
I
was
kind
of
really
happy
with
how
it
came
out.
Was
this
slide
because
a
lot
of
people?
It
really
puts
the
fine
point
on
it
that
what
you
really
care
about
is
what
service
is
being
provided
to
you
by
the
network,
you're
plugging
into
because
all
too
many
people,
like
I've,
literally
seen
lots
of
cases
where
people
are
thinking
in
terms
of
well.
Okay,
I
need
a
hard
Riddick,
okay,
but
what
kind
of
heart
Granick
oh
I,
need
a
40-acre
Grimek.
D
D
Not
quite
that
complicated,
yet
literally,
all
we're
doing
is
saying
we're
defining
an
enhanced
corporate
connectivity
and
for
whatever
that
happens
to
be,
and
then
you
know
in
the
particular
pods
some
NICs
will
be
marked
as
providing
enhanced
corporate
connectivity
whatever
that
means
so
Hannah
doesn't
actually
really
want
to
know.
Generally
speaking,
all
the
minutiae
of
what
enhanced
corporate
connectivity
means,
because
that
can
get
to
be
quite
stark.
She
just
wants
to
know.
This
is
the
thing
this
is
the
kind
of
thing
I
connect
to
and
it
gets
the
services
that
I
eat.
A
D
No
they've,
simply
maimed
just
like
network
services
in
general
and
network
service
mesh
are
are
simply
named
services
that
provide
stuff.
If
you
certainly
think
of
it.
This
way,
if
I
have
a
kubernetes
service
right,
III
I
don't
specify
in
the
kubernetes
service
all
the
things
about
the
the
service
that
actually
is
I
basically
tell
you
some
very
basic
things
about
it
and
I,
give
it
a
name
and
and
similarly
for
network
service
mash.
You
know,
I,
don't
have
to
enumerate.
D
A
A
Yeah,
there's
gonna
have
to
be
some
kind
of
ultimately
orchestration
driver.
That
knows
how
are
they
gonna?
Have
these
enormous
flat
space
of
all
these
services,
with
slightly
different
names
and
and
to
help
know
which
one's
the
deploy
them?
Then
someone's
gonna
have
to
map
the
abstract
name
to
the
to
the
details
of
what
makes
that
abstract
name
unique
for
people
like
Anna
who
don't
really
want
to
know.
You
know
we're
all
a
we're
all
in
a
world
where
everybody
has
their
own
domain
and
knowledge,
and
they
don't
want
to
know
about
the
other
person.
A
B
A
B
Little
a
little
thing
on
that
particular
section,
so
we're
since
we're
dealing
with
keys
with
basically
key
value.
As
for
for
looking
up
the
name,
there's
there
there's
nothing
that
stops
people
from
adding
some
organization
through
the
key
structure
like
they
do
in
that
CD.
So
you
can
stick
a
slash
somewhere
and
then
say
this
is
my
corporate
stuff.
This
is
my
VPN
stuff
or
so
on
and
and
maybe
build
out
a
little
bit
of
structure
as
well,
so
I
think
we'll
see.
B
Some
patterns,
like
naturally
evolve
that
that
lit
up
with
us
and
we
require
nope,
ideally
no
code
changes
to
what
we're
doing
to
help
support
these
other
than
maybe
maybe
enumerate,
but
kubernetes
in
that
CD
have
some
techniques
for
enumerating
based
on
that
ford
slash,
so
we
may
even
be
able
to
make
use
of
that.
So.