►
From YouTube: Fireside Chat: SMI and the Future of Service Mesh - Lee Calcote, Bridget Kromhout, & Alex Williams
Description
With the rise of Kubernetes, service meshes have evolved as a way to improve the reliability, security, and efficiency of service-to-service communications. According to the most recent CNCF survey, 18% of organizations are using a service mesh in production, with an additional 47% vetting a mesh for use in production. But adopting a service mesh can still be challenging for many organizations.
Join Lee Calcote and Bridget Kromhout from the Service Mesh Interface (SMI) project, a Sandbox project of CNCF, to learn how they are creating a standard for service meshes on Kubernetes. We’ll dive into how SMI is making it easier for organizations to adopt a mesh and what the future of service mesh looks like.
A
It's
great
to
see
my
colleagues
here
who
I
haven't
seen
in
so
long,
usually
I'll
see
y'all
passing
in
a
hall
somewhere
in
a
city
closed
near
and
far,
but
today
we
have
this
beautiful,
pink
backdrop
and
you
know
and
look
and
like
it's
like
the
Rolling
Hills
is
so
it's
how
nice
so
listen
we're
here
to
talk
about
Sur
as
much
I'm
here
with
Lee
Cal
coat
and
Bridget
come
out.
How
are
you
both.
A
Well,
good
good
good,
so
we're
talking
about
service
match
today
and
we're
talking
about
the
service
mesh
interface
in
particular.
I've
had
the
chance
to
interview
several
people
about
service
matches
over
the
past
few
years,
and
one
of
the
things
that
I
really
learned
out
of
it
is
and
I
think
maybe
we'll
just
briefly
go
over
service
match
and
I'll
give
you
kind
of
how
I've
been
hearing
it
summed
up,
and
then
we
can
get
your
perspective
and
then
I'd
like
to
get
into
I
really
like
to
get
into
the
customer
journey.
A
If
we
can,
you
know
how
it
is
the
customers
fair,
because
they're
the
ones
who
are
watching
this
and
is
there
as
much
as
the
customers
is
also
there
that
end
users,
which
are
people
like
yourselves
as
well
so
I
heard
serve
as
much
as
describe
recently,
is
offering
a
platform
for
simply
simplifying
microservices
communication.
It,
you
know
just
some
very
basic
terms:
it
handles
a
lot
of
complicated
pieces
around
microservices
communicating
which,
with
things
through
like
enforcing
policies,
managing
certificates,
you
know
servicing
issues
relevant
telemetry.
A
So
that
can
you
so
you
can
really
understand
what's
happening
in
your
cluster
and
those
problems,
as
we
all
know,
become
much
more
magnified
with
microservices
and
which
are,
you
know,
essentially
just
chaos
in
some
respects
right
so
service.
Baron
is
this
way
to
just
take
that
burden
away
from
the
developers
and
move
it
into
that
infrastructure
layer
and
so
there's,
basically
decoupling
the
teams
in
a
way
enabling
them
to
be
sick
to
success,
be
successful
at
the
same
time.
So
teams
are
meaning,
like
you
know
the
developers
in
the
operations
team
Bridget.
A
C
We
look
at
the
landscape
where
we
go.
Oh,
no,
it's
not
simple
and
our
organizations
generally,
even
if
they
have
a
few
people,
aren't
simple
and
so
being
able
to
properly
segment
who
changes.
What
who
has
access
to
what
and
who
has
responsibility
for
what
allows
us
to
kind
of
limit
our
blast
radius
for
anything,
we're
working
on
and
hopefully
not
have
too
much
context
baked
into
our
specific
application
and
too
much
spillover
between
roles
and
responsibilities.
B
Both
of
you
hit
it
on
the
head
actually
I
think
there
were
any
number
of
words
that
people
most
well
that
a
lot
of
times.
Actually
people
won't
use
when
they
describe
a
service
mush,
but
Alex
actually
mentioned
a
number
of
them,
and
what
Bridget
did
reinforced
I
really
liked
as
well,
and
that
is
that
you
can
even
like
any
piece
of
technology.
You
can
speak
to
it
about
in
hard
terms,
hard
technical
terms
about
like
what
features
and
functions
and
capabilities
does
it
have.
B
But
much
of
what
the
two
of
you
and
reinforced
was
actually
maybe
maybe
sort
of
the
true
power
of
a
service
mesh
which
is
which
we
mentioned
then
being
a
platform.
They
provide
a
few
things
around
observability
around
helping
take
the
blinders
off
around
helping
you
get
control
over
that
chaos
and
actually,
in
a
very
literal
way,
for
how
some
use
the
service
mash,
but
also
the
the
decoupling
around
teams
and
understanding
there
are
in
any
any,
even
relatively
small
sized
environment.
B
B
Well,
it's
not
it's
not
there
yet,
but
the
first
chapter
will
I
will
say:
it'll
speak
to
how
it
is
that
whether
you're
a
developer
product
owner
and
operator
an
architect
how
a
network
network
operator
how
how
the
service
nation
is
very
much
so
empowering
of
each
of
those
personas
and
I.
Think
it's
easy
to
overlook
that,
and
just
it's
sort
of
assumed
that
the
the
the
vast
majority
of
the
value
of
a
mesh
is
oriented
towards
the
operator
which
there's
lots
of
value
for
them.
A
C
So
I
would
say
the
best
way
to
understand
the
SMI
project,
and
this
one
was
fun
when
we
were
writing
the
proposal,
to
put
it
in
the
CNC
ifs,
a
sandbox
project,
and
it
turns
out
that
it's
a
spec.
So
it's
like
a
how
many
users
are
there?
Well,
it's
a
spec.
Many
folks
have
implemented
it,
but
I.
The
spec
is
so
that
all
the
people
who
want
to
use
things
like
traffic
shaping
or
a
canary
releases
without
having
to
build
the
application,
you
know
build
their
application
to
have
canary
in
it.
C
People
who
want
to
use
these
things
often
say.
Well:
okay,
cool
I
need
a
service
mesh.
To
give
me
these
kind
of
capabilities,
oh
wow,
there
are
lots
of
service
meshes
out
there.
What's
the
right
one
for
either
me
to
use
as
an
end
user
or
what's
the
right
one
for
me
to
target
as
application
developer
and
the
surface
mess
service.
C
Mesh
interface
say
that
three
times
fast,
a
project
just
allows
there
to
be
interoperability
between
meshes
such
that
both
end
users
and
application
developers
don't
have
to
make
a
lot
of
decisions
up
front
and
set.
Instead,
they
can
evolve
their
use
of
service
mash
as
their
needs
change,
without
worrying
that
they
lock
themselves
into
a
choice.
That'll
make
it
really
hard
for
them
to
do
anything
six
months
down
the
road
so.
B
A
C
Obviously,
but
there
many
of
the
meshes
out
there
have
implemented
at
least
this
and
then
some
more
and
B,
some
more
will
vary
and
this
spec
continues
to
grow.
So
because
this
is
an
open
source
project
in
an
open
source
spec,
you
can
actually
join.
Take
a
look
on
github
see
the
open
issues
of
the
development
of
the
spec.
As
it
goes.
We
have
open
community
meetings
every
two
weeks,
I
mean
much
as
you
would
expect
from
any
scenes.
C
You
have
project
and
it's
I
would
say
that
this
is
much
I,
don't
think,
there's
any
spec
out
there.
That
is
finished,
but
this
is
definitely
stable
and
it
is
being
implemented
and
adapters
created
even
for
meshes
that
don't
implement
it.
There
are
open
source
adapters
to
use
those
messes
messes
meshes
Wow.
That
word
is
remembered.
A
A
You
could
have
a
PhD
in
networking
and
still
be
studying
kind
of
how
these,
how
these
meshes
integrate
with
a
network
right
so
like
it's
very,
very
complicated,
and
so
it
was
described
to
me
as
the
people
who
are
using
it.
It's
they
have
to
just
start
and
learn,
and
it
takes
time
it
takes
a
lot
of
time
to
learn
how
these
servers
mesh
is
actually
work
and
how
to
use
them.
A
C
You're
absolutely
right,
Alex
I
would
say
that,
like
any
technology,
my
advice
is
always
instead
of
try.
This
awesome
new
technology,
maybe
it'll
solve
problems,
for
you
is
what
problems
are
you
trying
to
solve,
because
there
are
a
lot
of
companies
out
there,
large
and
small,
that
provide
support
for
work
in
this
space.
Companies
like
soullow,
do
have
some
really
great
service
mesh
hub
like
unified
dashboards,
stuff,
that's
out
there,
the
flagger,
progressive
delivery
operator
and
obviously
the
fine
folks
that
link.
Are
you
doing
an
ultra-light
service
mesh
there
are.
A
C
Look
at
something
like
solos
offering
and
see
if
or
look
at
your
cloud
provider
whatever
they
might
offer
and
see
if
some
of
the
boxes
that
they
tick
are
the
problem
you're
trying
to
solve,
because
it's
I
can't
give
you
a
generalized
answers
to
how
did
you
make
technology
choices?
Wouldn't
that
be
amazing?
That
would
be
a
for.
A
You,
yes,
you
can
Haley,
so
perhaps
what
we
can
talk
about
are
some
of
the
you
know
how
you
break
down
serve
as
much
as
like
people
are
using
them
for
traffic
management,
they're
using
them
for
security
and
they're,
using
them
for
observability,
so
in
traffic
management
it
may
just
you
know,
you
know.
Sre
teams
particularly,
is
spending
more
time
on
just
traffic
management
within
the
cluster
itself.
So
perhaps
we
could
start
there
as
an
example.
A
B
Slowly
and
and
once
you
become
comfortable
with
a
particular
capability
like
the
one
you
were,
you
were
identifying
for
observability
or
4s
are
used,
for
example,
or
traffic
management
press
are
used.
Then
there's
it
could
be
pretty
scary.
It's
kind
of
an
interesting
thing
that
networking
goes
deep
and
wide
and
that
actually
service
meshes
are
a
little
bit
of
a
next-gen
I'm
Sdn
I'm,
except
you
know,
sort
of
built,
much
more
for
cloud
native,
much
more
for
developers
and
and
api's
and
to
be
able
to
have
control
over
the
network
in
that
way.
B
So
it's
an
interesting
consideration
that,
like
you'd,
said,
networking
and
and
hence
service
messages,
can
take
a
while
to
digest
but
there,
but
if
developer,
if
you
will
I
guess,
is
ever
gonna
really
feel
comfortable
with
fine
grained
traffic
control.
It's
probably
in
this
context.
This
is
networking
made
for
developers
and
operators
being.
C
Absolutely
so
I
think
one
of
the
things
people
try
to
do
when
they
are
attempting
to
measure
C
and
understand
what's
happening
in
their
systems.
Is
they
want
to
do
a
B
testing
or
they
want
to
do
canary
releases,
or
they
just
want
to
look
at
traffic
metrics
and
feed
them
into
some
other
tool
that
they
have?
It's
like
the
traffic's
traffic
metrics
API
is
one
that
you
can
plug
into
your
observe
observability
platformer,
tooling
of
choice.
C
That's
again,
like
you,
don't
necessarily
have
a
lot
of
large
enterprises
that
are
totally
cool
with
individual
J,
random
application,
developer
walking
around
in
the
production
logs,
looking
at
everything
and
so
having
the
ability
to
send
certain
kinds
of
your
traffic
to
different
places
or
see
what
the
percentage
of
completions
are
without
necessarily
inspecting
every
packet
yeah.
Those
are
those
are
kind
of
the
granular
things
that,
once
you
start
actually
implementing
this
for
realsies,
you
start
to
care
about.
A
What
are
some
of
the
other
things
Li
that
you're,
seeing
in
your
own
research
into
your
book,
that
you're
finding
are
challenging
aspects
of
people,
learning
how
to
digest
service
matches
and
and
what
are
they
trying?
You
know
what
you
know,
what
are
the
kind
of
the
maybe
the
kind
of
the
the
more
significant
problems
are
facing
upon
their
journey.
I
would
imagine
network
management
and
understanding.
That
is
a
big
part
of
it.
A.
B
Mesh
adoption
is
a
focus
of
mine,
in
particular
that
I'd
say
just
by
pure
rock
number:
account
that
that
the
majority
of
folks
that
I
speak
to
are
adopting
because
they
want
to
get
to
observability
or
that's
the
piece
of
value
that
that
they
they
need
the
most
and
most
critically
now
is
they
just
don't
I,
don't
know,
what's
going
on
between
their
services
and
what's
going
on
they
now
that
said,
there
are
I
guess
to
put
some
adopters
into
a
box.
I
would
say,
there's
a
lot
of
times.
B
So
it's
secure,
so
the
encryption
is
that
package
of
security
is
like,
and
sometimes
they
just
stop
there,
because
that's
that
was
enough,
for
they
I
mean
they
take
a
maybe
I
shouldn't
say:
stop
with
a
full
stop,
but
they
take
a
pause
in
their
in
their
journey,
become
comfortable
and
then
for
some
of
the
capabilities
of
the
mesh.
People
are
getting
some
of
those
things
elsewhere
there.
It's
not
like
observability
and
monitoring
is
a
new
thing.
B
People
have
other
tools,
meshes
integrate
into
those
existing
tools
and
but
for
some,
particularly
those
that
are
starting
out
new
projects.
Doing
an
in
context
of
a
service
measure
they're
there
or
they're
aware
that
one
will
be
in
the
environment
way.
They
can
move
a
lot
faster
when
they
just
trust
that
the
infrastructure
will
provide,
and
that's
actually
because
of
all
those
folks
that
we
spoken
to
you.
It's
actually
why
we
started
the
project
measure
e
the
surface.
A
They're
so
so
Bridget
I
was
seeing
some
stats
that,
were,
you
know,
courting
a
believe
us,
according
to
the
to
the
CN
CF
about
eighteen
there's,
they
did
some
surveys
and
about
eighteen
percent
of
the
people
said
that
they're
using
service
measures.
So
that's
them.
That's
a
pretty.
You
know,
that's
not
quite
20
percent,
but
it's
still,
you
know
it's
still
not
zero
I
mean
is
you're
still
seeing
a
lot
of
people
I
think
comparatively
kubernetes
architectures
are
upwards
of
seventy
percent.
A
You
know
in
their
survey,
are
using
these
are
using
kubernetes
as
a
platform,
but
for
those
eighteen
percent,
you
know
what
are
you
seeing
at
some
of
the
day
two
challenges
that
they're
facing,
because
you
can
start
getting.
You
can
get
started
with
a
service
match,
but
there's
always
day.
Two
then
there's
that
in
day,
three
and
beyond,
what
are
some
of
the
things
that
are
noteworthy
to
you
that
you
can
share
yeah.
C
Absolutely
so,
it
was
actually
recently
chatting
with
Lee,
specifically
around
the
conformance
testing
for
SMI.
This
idea
that
it
would
be
nice
if
you
know
that
what
you're
developing
and
what
you
have
in
production
works
for
some
values
of
works
and
maybe
meet
some
sort
of
minimum
standards.
It's
very
simple
to
deploy
something
from
github
and
be
like
it
works,
and
it's
a
lot
more
of
a
lift
like
you're,
saying
Alex
the
day
to
operability
of
is
this
replicable?
Can
we
deploy
this
in?
You
know
our
eighty
data
centers
or
800
datacenters?
C
Can
we
upgrade
it
according
to
our
schedule?
Can
we
make
sure
that
we
don't
have
the
single
points
of
failure
that
come
from
the
one-off
and
the
experiment
and
instead
get
into
productivity,
because
I
find
service
not
just
really
exciting,
and
so
does
lean,
probably
Alex,
just
you
and
the
vast
majority
of
enterprises,
governments,
educational
institutions
that
are
out
there
would
really
not
prefer
to
spend
all
their
time
on
the
excitement
of
service
match.
C
They
want
to
get
something
done
with
it,
and
so
that's
why
we're
all
work
together
in
the
ecosystem
to
try
to
make
everything
a
lot
more
operable,
and
so
that's
why
I
would
my
recommendation
there
would
be
go.
Look
at
the
vendors
in
the
space.
Look
at
the
open
source
projects
in
the
space.
You
have
very
smart
people.
You
probably
can
implement
a
lot
of
these
things,
but
we
go
a
lot
farther
together.
C
C
A
B
A
B
About
some
things
that
well,
that
I
think
that
I
have
yet
to
be
realized
about
the
power
of
a
mesh
and
its
enablement
of
the
service
owner
the
product
owner.
There's
some
intelligence
that
can
be
built
into
the
data
plane
in
specific
means
that
you
can
get
some
things
with
no
code
change
if
you're
a
product
owner
and
you've
got
a
choice
between
taking
two
development
sprints
and
having
having
developers
implement
what
can
be
done
in
infrastructure.
Maybe
you'll
choose
just
let
the
the
mesh
do
that
for
you.
B
A
C
Of
a
lot
of
organizations,
there
are
a
little
bit
leery
still
of
letting
their
developers
commit
code
to
an
open
source
project
and
I,
say:
okay,
a
great
place
for
you
to
participate
is
testing
validation
and
you
know
documentation,
opening
issues
against
a
spec,
because
and
that
way
it
serves
the
interests
of
your
org.
It's.
A
Such
a
professional
opportunity
isn't
it
it
allows
you
to
meet
so
many
people
and
to
really
open
your
horizons.
I've
always
I've
always
thought
that
would
be
such
a
cool
opportunity
for
someone
who's
new
to
the
space
to
really
get
into
it.
I
want
to
thank
you.
Both
I
I
would
have
loved
to
talk
for
another
hour,
at
least,
but
there's
18
months
at
least
ahead
of
us,
so
I'm
sure
we'll
have
the
opportunity
again
so
Lea
calcio
Lea,
it's
great
to
see
you
again
again
here.
Bridget
kromm
out
always
wonderful
to
talk
with
you.