►
From YouTube: CNCF Telecom User Group Meeting - 2020-08-03
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Most
of
you
know
me.
I
was
running
cncf
for
the
previous
four
years
and
stepped
down,
although
I
still
am
leading
a
lot
of
our
or
involved
in
a
lot
of
our
telecom
efforts.
I've
now
gone
and
launched
a
new
effort
called
linux
foundation,
public
health,
which
some
of
you
are
actually
involved
in
and
definitely
appreciate
that
support
we
just
launched
two
weeks
ago
prior
to
since
our
last
meeting.
A
So
I
will
go
ahead
and
put
the
url
in
here,
just
as
a
little
plug
feel
free
to
take
a
look
at
some
things
that
we're
doing
we're
very
pleased.
On
friday,
canada
just
launched
their
national
app
based
on
the
code-based
covet
shield,
which
we
host
and
ireland
has
another
app
called
covet
green
based
on
code
based
on
a
code
base,
we're
hosting
so
we're
we're
expecting
that
to
now
come
to
many
u.s
states
and
a
number
of
other
countries
as
well.
A
But
that's
my
side,
gig
or
or
this
is
depending
on
your
point
of
view.
So
let's
turn
it
to
telecoms.
I
I
did
want
to
mention
to
welcome
tall
for
their
own
from
red
hat
here.
He
and
I
have
been
speaking
about
tosca
and
cnfs
and
such
for
I
think,
a
year
now,
and
so
I'm
very
interested
to
hear
his
presentation
about
torandot
and
appreciate
his
provocatively
titled.
A
Is
it
time
to
stop
using
helm
for
packaging
instead
standardize
on
tosca,
which
hopefully
will
get
everybody's
attention
here,
if
not
switched
over
by
the
end
of
the
call?
But
I
think
we
should
kick
it
off
with
taylor
if
you
could
take
us
through
for
a
few
minutes
taylor,
particularly
the
updates
around
where
the
cnf
conformance
beta
testing,
how
that
progress
has
been
going
and
then
also
collaboration
with
cmtt,
I
think,
are
going
to
be
the
the
key
initial
topics.
B
Sure,
thanks
stan
okay,
let's
see
I
kind
of
had
it
a
little
bit
backwards
on
that
for
the
collaboration
first,
but
let
me
bring
up
whichever
you
prefer
to
it's:
okay,.
B
Slides
so
it
might
be
yeah,
I
guess
I'll
start
with
the
the
community
side.
I.
C
B
Specifically,
it's
probably
interesting
just
to
jump
right
in
to
this
one,
so
there's
a
lot
of
different
initiatives
and
communities
and
probably
could
put
in
the
turin
dot
right
here
as
far
as
implementations,
but
there's
so
many.
So
this
is
just
a
small
snapshot.
Looking
at
collaborations
now,
specifically,
this
slide
deck
and
I'll
go
back
over.
B
It
is
about
collaborations
around
the
workload
and
where
requirements
are
coming
from
and
how
we're
driving
those
and
stuff
and
there's
been
a
lot
of
work
on
the
within
the
cnt
rc2
work
stream,
and
then
how
does
that
tie
in
with
ovp's
badging?
And
then
how
does
that
tie
in
with
all
the
conformance
testing
from
the
cnf
cloud
native
principles?
Aspect
that
cncf
is
doing
in
the
cnf
conformance
program,
so
we
got
together,
and
this
is
just
a
slide
deck
to
oliver
smith.
B
Frederick's
cots
has
also
been
working
on
this
with
us
and
myself
to
try
to
intro
an
effort
that
we've
started
around
the
workload
specifically
so
that
we
can
try
to
group
the
different
efforts
to
source
as
much
of
the
requirements
as
possible,
while
still
giving
autonomy
for
all
the
different
groups,
because
if,
if,
if
you
look
at
some
of
these,
the
rc2
is
going
to
be
using
opnfv
testing
and
cnf
conformance.
It's
also
leveraging
upstream
end-to-end.
So
it's
has
a
lot
of
upstreams.
B
If
you
look
at
all
of
these,
so
we're
one
of
the
upstreams
that
we're
trying
to
provide-
and
we
want
to
try
to
help
with
this
and
we're
trying
to
do
make
upstream
sources
available
for
all
the
different
communities
like
the
cnf
testbed
can
be
used
as
an
upstream
for
the
ri2
testing.
Just
like
some
of
the
other
testing
pieces,
and
this
particular
effort
is
around
the
workload.
B
B
We
have
a
large
spreadsheet
of
covering
a
lot
of
the
platform
requirements
that
we
thought
would
be
good
from
cloud
native
principles
that
a
platform
could
use
and
we're
trying
to
focus
on
a
lot
of
that
from
the
workload
as
a
group
and
looking
at
how
ovp2
could
use
it
for
as
one
of
the
sources
and
the
platform-as-a-service
initiative,
x-juvella
and
other
ones
that
that
could
use
it.
So
these
are
some
of
the
groups
that
we've
been
reaching
out
to
and
I'm
trying
to
reach
out
to
everyone.
B
That's
here
on
the
topical
music
group
and
reached
out
to
some
others
as
well,
and
what
we
have
is
a
spreadsheet
for
restarting
as
far
as
collaborating
on
it.
That's
this
it's
linked
in
here
and
I'll
I'll
put
this
in
the
notes
here
in
a
minute,
but
we
have
a
spreadsheet
that
we're
building
out
all
the
requirements
and
then
saying:
where
do
they
the
source?
If
they
come
from
or
any
type
of
ideas?
B
And
then
is
this
a
principle?
Is
it
some
type
of
functional
requirement
if
we
look
at
like
distinct
network
connections,
but
we
have
a
walkthrough
of
what
all
of
this
is
about,
and
then
the
communities
and
and
where
we're
referencing
everything
the
end
result
we're
hoping
is
a
set
of
requirements
that
then
can
be
used
back
in
those
upstream
projects
to
to
drive
things
and
then
any
tests
that
we
can
say
are
already
covering.
B
So
some
of
these
we
may
say,
oh
this
is
already
covered
by
it
could
be
close
enough
on
the
integration
between
platform
and
workload
that
we
may
say.
Oh
there's,
a
kubernetes
conformance
test
that
covers
that
piece
or
another
one
may
be
the
scene
of
conformance
covers
it
or
whatever
it
may
be,
but
the
end
result
is
to
be
able
to
use
in
in
the
different
projects
and
allow
each
one
of
those
to
go
off
and
use
them
remix
and
modify.
B
That's
not
it
cnf
conformance
itself.
We've
been
working
on
the
first
set
of
of
platform
test
based
on
input
from
many
sources,
including
the
reference
architecture,
requirements
from
ctt
on
the
platform,
but
we've
also
been
talking
directly
with
operators
and
and
cnf
developers
like
so
that
could
be
vendors
that
could
be
developers
on
open
source
projects
like
network
service
mesh
together.
B
B
So
these
are
just
to
quickly
reverse
back,
so
we
here's
the
categories
and
the
first
set
of
workload
tests
and
the
same
categories
that
we're
using
microservice
compatibility
all
the
way
through
stuff,
like
hardware
resilience,
we're
doing
chaos
testing.
These
are
the
same
type
of
categories
that
we're
using
on
workload,
we're
also
using
on
platform,
and
this
is
that
first
set
based
on
feedback
that
we're
trying
to
implement,
and
we
this
is
partly
who's.
B
Currently
active
we'd,
be
happy
to
have
other
folks
that
would
like
to
contribute
directly
to
writing
tests
or,
if
you
have
ideas
for
new
tests,
but
this
is
the
first
set
that
we're
looking
at.
We
have
some
that
were
about
chaos
and
stuff
within
the
containers
and
pods
themselves
and
ensuring
that
the
workloads
continue
we're
doing
the
same
sort
of
stuff
on
the
platform.
B
So
the
idea
here
is
to
have
a
set
that
anyone
could
use
whether
you're
testing.
If
you're
focused
on
the
platform,
you
can
run
the
platform
test.
If
you
want
the
workload,
you
can
run
the
workload
test
or
you
can
run
the
whole
suite
if
you
have,
if
you
desire,
and
then
every
individual
test
can
be
run
by
itself,
as
well
as
the
entire
categories
and
that's
across
platform
or
workload.
B
B
Many
of
the
things
can
be
done
with
kind
of
course,
as
that's
going
to
have
some
limits
once
we're
in
the
platform,
it
seems
to
work
fine
for
workload,
but
the
platform's
going
to
have
more
limits
and
the
the
test
suite
itself
primarily
just
needs
a
a
cube
config
to
target.
So
we
would
like
to
have
other
platforms
available,
and
so
what
we're
going
to
produce
is
something
similar
to
our
cnf
examples.
We
have
a
cnf
examples,
documentation
in
the
scene
of
conformance
what
we'd
like
is
some
platform
examples.
B
So
if
anyone
knows
of
any
platforms
that
are
available,
that
would
be
good
targets
to
to
run
tests
against
then
we'd
be
happy
to
add
those
to
the
list
of
ones
to
check
out
and
and
then
of
course,
it's
the
whole
suite.
The
whole
test
suite
is
based
on
the
kubernetes
conformance
program
and
specifically
in
that
you
can
download
and
run
it
yourself.
So
there's
an
extensive
docs
around
that
and
for
those
that
did
not
see
or
you
weren't
here
last
time
we
got
through
with
a
beta
test
round.
B
D
Yeah
hi
taylor:
this
is
sudeep
batra.
D
Yeah,
actually
this
is
first
time
for
me
joining
this
meeting.
So
I
see
there's
a
lot
of
information.
Is
it
possible
to
share
the
link
of
the
slides
that
you
have
covered?
It
would
be
a
good
reference.
B
E
Hey
taylor,
all
right
yeah,
so
maybe
I
can
just
give
a
quick
overview.
E
So
cmtt
is
if
you're
not
familiar
with
the
effort,
is
designing
reference
architectures
of
how
kubernetes
should
be
run
in
the
telco
industry,
and
that
was
a
little
bit
what
you
saw
on
the
slide
deck
that
taylor
was
presenting
earlier
and
as
a
part
of
that,
so
I'm
bill
logan
and
I'm
the
work
stream
lead
for
the
reference
conformance
based
on
kubernetes,
and
one
thing
that
we're
trying
to
do
is
to
have
a
survey
of
how
kubernetes
is
being
used
right
now
in
telecommunications,
and
so
we're
trying
to
survey
the
industry
to
see
and
kind
of
like
track,
both
the
progress
of
the
industry
and
the
progress
of
our
work
against
the
industry.
E
E
Of
what
the
actual
progress
is,
and
so
the
goal
of
this
survey
is
to
have
feedback
from
operators
and
vendors
of
what
their
actual
usage
of
kubernetes
is
right
now.
So,
if
anybody's
on
this
call,
I'd
be
it'd
be
great.
If
you
could
help
us
by
filling
out
the
survey
and
then
we
hope
to
repeat
this
survey
on
an
annual
basis
and
provide
the
results
back
to
the
community.
E
So
we
have
kind
of
a
barometer
of
what
the
actual
usage
of
kubernetes
and
other
cloud
native
technologies
in
the
telecommunications
spaces,
and
this
will
be
used
to
help
inform
the
work
that
we're
doing
in
cntt
in
both
the
reference
architecture
and
the
reference
conformance
work
stream.
So
please
help
us
out
by
filling
it
out
and
passing
it
on
to
other
operators
that
you
work
with.
E
B
B
B
C
C
All
right
excellent,
so
I
wonder
you
know
I
actually
don't
know
how
long
this
meeting
usually
goes.
How
long
do
you
think
I
can
take
or
who's
who's
the
organ.
C
Yeah,
that's
that
sounds
pretty
good.
So,
first
of
all,
thank
you.
Thank
you
again
for
the
introduction
that
was
exactly
right.
You
caught
on
the
provocative
statement
pretty
well.
My
intent
was
to
provoke
there
but
and
yeah.
Also
I
want
to
apologize
for
you.
Some
of
you
don't
know
me
and
I
just
dropped
in
the
meeting
and
parachuted
in
with
a
presentation,
but
it's
a
conversation.
I've
been
having
with
a
few
of
you
for
a
while.
C
I
usually
like
to
have
my
ducks
in
a
row
before
I
present
something,
but
I
think
the
ducks
are
are
aligning.
So
it's
a
good
time.
I
want
to
start
actually
by
asking
you
know
there
are
26
people
here
right
now.
I
think
I
think
I
can
ask
this.
Do
any
of
you
have
any
opinion
about
tosca
or
have
you
worked
with
tosca?
What's
your
impression
of
what
tosca
is,
I
realize
some
of
you
never
heard
of
it,
but
anybody
want
to
pipe
in
with
some
non-judgmental
for
me
impression.
C
Okay,
everybody's
shy,
so
you
know:
task
has
been
around
since
2014.
It's
it's
a
language
that
is
supposed
to
be
used
for
cloud
orchestration.
That
was
the
intent
started
off
in
xml,
moved
very
quickly
to
yaml,
and
it's
now
with
version
1.3
and
a
group
of
us
are
working
on
tosca
2.0.
C
I
think
you
know
when
I
asked
people
about
tasks.
A
lot
of
the
impression
is
they
remember
it
from
it
was
it
was
supposed
to
be
something
like
what
heat
templates
are
in
openstack.
That
is
a
declarative
language
that
you
can
describe
your
resources
and
it
has
topological
knowledge.
So
there
there
it's
basically
a
graph
database,
so
you
have
nodes
and
you
have
edges
between
the
nodes,
right,
vertices
and
edges,
and
but
people
remember
that
it
came
with
all
these
basic
types.
C
It
was
called
the
simple
profile
that
were
meant
to
describe
this
abstract
cloud
infrastructure
and
there
were
things
like
a
compute
node
and
a
network
relationship,
a
network
node
and
relationships
and
data
types
etc.
C
I
think
that
that
left
a
really
bad
taste
in
people's
mouths,
and
when
I
asked
people
about
tosca
today
they
either
say
it's.
I
heard
it
described
as
a
necessary
evil,
so
people
who
work
with
onap
and
etsy
mano,
if
you
know
etsy
mano,
embrace
tosca
as
a
descriptor
for
network
functions,
so
people
who
need
to
work
with
it
find
it
a
necessary
evil.
C
My
opinion
is
that
if
taska
didn't
exist,
we
would
have
to
invent
it.
There
have
been
similar
efforts
to
use
the
yang
language
to
to
start
enhancing
it
and
describe
things
but
pascal's
been
around
for
a
while
and
and
is
designed
to
do
exactly
that
and
part
of
what
the
team
working
on
faster
2.0
right
now
is
really
trying
to
remove
that
kind
of
craft
that
we
had
before.
So
one
of
the
first
things
we
did
was
actually
remove.
E
C
For
costa
and
yeah,
you
know
what
I
I
think
the
best
way
for
me.
I
don't
I
didn't
plan
I
do
have
slides,
but
I
thought
for
this
group.
It
would
be
better
not
to
go
by
slides,
but
just
to
keep
it
an
open
conversation,
so
do
feel
free
to
interrupt
me
at
any
time.
So
I
will
share
my
screen
if
I
can
and
show
you
how
it
works,
how
tosca
on
kubernetes
works
specifically
because
there
are
versions
of
tosca
using
openstack
and
using
bpmn
kind
of
middleware
solutions.
C
A
A
It
was
challenging
to
try
and
use
a
heat
template
to
describe
a
cnf
on
kubernetes,
and
that
tosca
was
not
that
and
so
was
more
portable,
but
had
potentially
the
opposite
problem
of
being
so
general
purpose
that
it
didn't
necessarily
describe
it,
and
I
guess
I'm
I'm
particularly
curious
or
concerned
about
well,
if
you
eliminate
the
primitives
what
is
left-
and
I
do
think
that
just
some
examples
would
be-
would
be
the
best
way
of
trying
to
address
that.
C
Yeah
I'll
show
you
I'll
say
briefly
that
tosca
does
have
some
killer
features
that
kubernetes
crts
don't
have
kubernetes
crds,
keep
getting
better
right
right
now,
there's
like
the
open
api
kind
of
data
validation
right.
If
you
look
at
the
history
of
kubernetes,
we
started
off
with
specs.
We
added
statuses,
you
know
status
sub-resources
and
we
added
data
validation.
C
You
know,
kubernetes
manifests,
are
slowly
growing
up,
they're,
still
very
very
far
from
what
you
get
with
tosca
with
hospital.
You
have
a
very,
very
rich
data
validation
system.
It's
very
strictly
type
language
object
oriented,
so
you
can
inherit
types
and
also
you
have
the
topology
which
you
have
relationships
between
types.
You
know
when
you
build
things
like
service
meshes
and
things
and
kubernetes,
you
know
we
we
build
them,
but
they're,
not
kubernetes
itself
doesn't
know
about
the
the
relationships
right.
We
have
selectors.
C
We
have
other
ways
to
do
it
in
kubernetes
and
another
killer
feature
feature.
Is
service
composition?
What
I'm
going
to
show
you
defining
things
that
are
multi-cluster?
You
can
replace
parts
on
demand.
So
if
you
have
a
network
service
and
you
need
a
certain
network
function,
it
can
be
taken
from
the
inventory
and
replace
with
something
else.
Once
you're
object
oriented
you
have
contracts
right
with
the
base
type.
So
if
you
expect
a
certain
base
type
to
exist,
you
know
subclasses
could
implement
it
in
different
ways.
C
I
I
kind
of
wish
kubernetes
was
built
on
tosca
from
the
start.
I
think
we're
slowly
moving
towards
it.
That's
why
I
say
that
if
it
didn't
exist,
we
would
invent
it
we're
slowly,
recreating
the
wheels
of
what
cost
is
in
kubernetes
right
now.
As
far
as
I
can
tell
that's,
not
a
problem,
you
know
we
can
fill
in
the
back
fill
in
the
gap.
Kubernetes
is
an
extensible
platform
and
I'll
show
you
an
operator
today
that
can
bring
pasta.
So
I
might.
C
All
right,
so
I
hope
it's
all
big
enough,
so
I
sent
on
the
agenda.
You
can
see
a
link
to
this
too
so
yeah,
instead
of
slide,
where
you
can
read
the
readme
page,
I
think
it
goes
through
a
lot
of
what
it
is,
but
it's
really.
It
is
just
an
operator,
it's
a
kubernetes
operator,
and
but
what
it's
meant
to
do
is
consume
tosca
packages
and
orchestrate
them
on
kubernetes
turn
them
into
kubernetes
workloads
and
what's
interesting
about
the
workflow
is
they
are
integrated?
It
uses
annotations,
so
it's
kind
of.
C
If
you
use
helm,
you
know
you
deploy
something
with
helm.
You
can
delete
that
whole
deployment
with
help
right,
so
the
language
in
task
is
called
service
templates.
So
what
you
deploy
with
toska
is
a
service
right,
but
it
can
be
an
application.
It
can
be
any
other
workload.
C
What
I'm
trying
to
show
you
today
is
the
example
of
a
telco
workout
specifically.
So
yes,
I
work
specifically.
So
I'm
quick
introduction.
My
name
is
tal.
I'm
from
red,
hat's
office,
cto
office,
and
specifically
the
networking
group
right.
So
this
is
telco
is,
is
behind
all
this
really
telco
is
the
example
of
kubernetes
workloads
that
I
think
are
some
of
the
most
complex,
because
they're
always
very
heterogeneous.
You
always
deal
with
black
boxes
right
that
come
from
various
vendors
that
have
their
own
vertical
stacks.
C
You
deal
with
v
and
fms,
sometimes
entire
orchestration
systems
that
come
with
the
products.
So
you
end
up
having
to
stitch
together
a
lot
of
things
that
are
very
different
and
it's
often
very
hard
right.
You
have
big
projects
like
onap,
trying
to
herd
the
cats
so
to
speak,
but
a
lingo
of
franca
like
tosca
is
a
great
way,
I
think,
to
model
the
various
parts
you
have
and
able
to
delegate
to
those
orchestration
other
orchestration
systems,
if
you
need
to
via
kubernetes
operators
or
sometimes
just
api,
calls
to
an
orchestrator
running
elsewhere.
C
If
that's,
what
needs
to
be
the
case?
So
the
example
I'm
going
to
show
you
today
is
a
multi-cluster
network
service,
specifically
telephony,
so
there's
a
central
cluster
there's
an
edge
cluster.
C
Each
one
is
running
the
asterix
pbx.
One
of
them
is
running
it
in
as
a
vnf
using
a
keyboard.
The
other
is
the
smaller
edge.
Cluster
is
going
to
run
it
as
a
container.
The
cnf
multis
is
being
used
to
create
a
data
plane
between
them.
Actually
that
goes
through
a
pnf,
an
external
router,
so
so
that
can
also
be
orchestrated,
basically
using
netpof
and
yeah.
So
in
the
end
we
have
the
control
plane.
That
is
the
standard
of
barnettis.
C
Then
we
have
the
data
plane,
which
is
the
second
interface
I'm
putting
one
in
quotation
mark,
because
I'm
running
this
on
one
machine
right
now
for
a
demo,
and
you
can
have
sip
phones
that
connect
to
each
of
them
and
can
call
each
other,
and
that
goes
through
the
sip
trunk
that
goes
over
the
lan.
So
let's
see
how
that's
implemented
so
I'll
look
at
the
csar
packages,
so
each
one
of
these
things
is
csar,
so
csar
dot
csar
stands
for
cloud
service
archive,
and
that
is
a
task
definition.
It's
very
very
light.
C
It's
basically
a
zip
file
that
has
that.
Has
this
manifest
it's
inspired
a
little
bit
by
java
jars.
That's
not
too
interesting!
What's
interesting
is
this
yaml
file?
That
is
the
actual
possible
plastic
definition,
so
I
want
to
start
from
the
top
off
top
down.
This
is
the
network
service
that
I
want
to
deploy
as
a
whole
and
yeah.
You
know,
if
you,
you
know
you're
yaml.
C
Some
of
this
would
be
familiar,
but
then
you
know
this
is
this
is
the
task
language,
so
we
have
a
topology
template,
which
is
everything
this
whole
service,
and
then
we
have
a
bunch
of
node
templates,
and
this
is
this
looks
very
abstract.
Each
have
a
type
so
this
this
is
a
pbx
type,
which
is
a
base
type,
and
there
are
other
things
that
inherit
from
it,
and
here
I'm
defining
very
generally
what
I
want.
I
want
certain
capabilities.
I
want
to
create
a
trunk.
C
Obviously,
you
would
probably
want
to
connect
this
deployment
to
a
real
introduction
to
a
real
database,
but
this
is
an
easy
way
to
generate
them,
and
then
I
have
requirements
that
create
the
topology.
So
this
needs
to
connect
to
a
data
plane
which
is
its
own
node
and
also
there's
a
trunk
that
runs
on
that
and
it
wants
to
connect
to
the
edge.
So
this
is
the
central
pbx,
here's
the
edge
pbx,
it's
very,
very
similar,
it's
just
the
opposite
way.
C
In
no
case
did
I
say
that
this
should
be
a
vnf,
or
this
is
a
cnf.
Yet
how
I
do
that
is
by
policy,
so
policies
are
are
ways
to
give
hints
to
the
orchestrator
to
tell
it
what
I
want
to
do
in
this
case.
This
is
a
provisioning
policy
and
I
can
tell
it
which
site
I
wanted
to
run
so
there
are
these
function
calls,
and
I
can
also
give
it
substitution
input.
So
this
is
a
little
bit
technical
here
for
tosca,
but
the
way
the
task
it
does
service
composition
is
by
substitution.
C
So
these
nodes
can
be
substituted
by
entire
workloads
that
implement
that
same
pbx.
So
in
this
case
I
want
I'm
going
to
substitute
it
with
a
vnf
that
is
much
more
complex.
It
has
a
database,
it
uses
kubert
in
this
case
it's
a
self-contained
cnf,
but
I
decide
which
one
it's
going
to
be
according
to
policy
and
there's
an
inventory
and
I'll
show
you
how
I
register
things
an
inventory
here.
These
can
replace
so
I'll
move
a
little
bit
quickly.
I
don't
want
to
take
too
much
time
on
this.
E
C
Is
defined
using
so
here
I
import
various
kind
of
profiles
into
the
taskbar,
and
one
of
them
is
the
kubernetes
profile
and
the
kubernetes
profile
will
look
a
lot
like
kubernetes,
manifest
that
you
already
know
so.
There's
the
kubernetes
metadata
there's
the
deployment
udp
here
is
defined
as
a
service.
So
this
is
a
a
load.
Balancer
just
inherits
the
kubernetes
service
and
it's
of
the
load
balancer
type,
so
we
have
ports
we
have
ingress
attributes
so
attribute
is
what
we
call
in
tosca.
C
That's
basically
the
status
sub
resource
in
kubernetes.
You
know
runtime
attributes
that
change
properties
or
the
things
that
tend
to
not
change.
So
a
lot
of
this
you
know,
if
you
know,
kubernetes
manifests
it's
almost
the
same.
You
see
the
difference
here.
There's
some
task
of
function
calls,
but
this
is
a
way
to
to
package
together
and
yeah.
Excuse
me
this.
This
is
not
a
really
great
implementation
of
rtp
tunneling,
but
this
is
for
the
demo.
I
think
this
will
work.
What's
interesting.
C
This
is
installed
and
running,
there's
cloud
native
configuration,
that
is
the
service,
can
orchestrate
itself
using
a
a
script
in
this
case
so
and
all
these
artifacts
that
are
attached
to
it.
So
if
I
look
back
to
the
csr
file,
you'll
see
it
created
this
artifact
sub
directories
here,
and
there
is
actually.
This
is
interesting.
There's
there's
a
tar
bottle
here
that
I
included
of
the
entire
container
image.
Now
I
don't
have
to
do
that.
C
You
know
when
you
create
a
deployment,
you
know
you
give
a
container,
you
give
the
image
and
it
can
just
be
a
url
on
some
docker
repository
in
this
case.
I'm
using
the
get
artifact
function
to
actually
take
take
the
artifact
that
I'm
attaching
here
in
the
in
the
csr
file.
So
I
don't
have
to
do
this,
but
I
can
package
the
cnf
container
image
in
that
zip
file
with
it
and
turon
dot
will
be
able
to
upload
that
automatically
to
its
inventory.
C
F
C
Not
too
too
interesting,
but
it
does
some
takes
various
templates
and
creates
the
configuration
files
for
asterisk
according
to
various
inputs
that
it
gets
and
takes
from
the
from
the
tosca.
C
So
how
does
it
all
work
so
I'll
show
you?
I
have
the
two
clusters
running.
So
here's
the
the
central
cluster
and
this
one
is
the
big
one.
It
has
a
lot
more
memory,
it
has
qbert
the
edge
cluster
is,
doesn't
have
q
vert,
it's
it's
a
bit
smaller,
so
I
will
show
you
first.
I
have
a
little
cheat
sheet
here.
I
will
first
install
turn
dot
on
the
central
workspace.
C
And
the
turn
dot
command
line
is
a
very
you
can
do
everything
with
cube
ctl,
but
the
torn
dot
tool,
just
like
many
other
tools
like
istio,
cto
and
home.
These
command
lines
just
make
it
much
easier
to
work
instead
of
using
the
crgs.
E
C
Oh
sorry,
that's
the
edge,
so
we
have
so
here
it
is.
We
have
the
200.
operator
and
we
have
the
turned
out
inventory
running
separately.
In
this
case.
It's
just
a
docker,
the
default
docker
repository,
but
this
could
be
something
much
more
big.
Any
kind
of
doctor
repository
could
work,
so
it
could
be
quay
harbor,
all
right,
so
that's
installed
here
now
I'm
going
to
register
okay,
there's
the
hello
world
working
door,
we'll
jump
right
to
this,
so
I'm
going
to
register
first
of
all,
we're
going
to
set
a
delegate
okay.
So
the
delegate.
C
Delegate
is
a
way:
oh
okay.
I
want
to
change
to
set
the
default
oops
I'll
work
on
the
central
cluster
and
then
I'm
going
to
register
delegate
what
the
delegate
does
is
connects
the
central
cluster
to
the
edge
cluster.
So
now
the
edge
cluster,
the
central
cluster
knows
how
to
use
kubernetes
client
certificate
to
actually
connect
to
the
edge
cluster,
and
now
I'm
going
to
register
on
the
inventory,
all
the
csars
that
I
that
I
had
I'm
just
going
to
every
time.
This
happens.
C
C
It
has
all
of
them
and
right
now
they
have
no
services
deployed
with
anything
and
now
I'm
actually
going
to
deploy
the
network
service
of
a
single
command
and
what
it
does
is
it
uses
the
template
that
I
already
registered
and
gives
it
an
input.
I
tell
it
which
namespace
I
want
to
work
and
I'm
going
to
do
a
watch.
C
So
what
this
watch
does
it
does
a
service
list
and
we're
going
to
see
see
what
happens
here.
So
we
have.
The
telephony
network
service
was
instantiated
and
immediately
it's
substituted
using
the
policy,
the
the
asterisk
vnf
for
it,
and
we
also
have
a
simple
data
plane
that
was
created
using
malta
cni,
so
it
already.
These
are
the
inputs
that
are
coming
in.
So
these
are
instant.
This
has
already
been
instantiated.
This
is
done
statute,
so
the
the
cube
does
take
a
while
to
start.
C
You
know
it's
starting
a
whole
virtual
machine,
but
I
will
show
you
what
we
see
so
for
now.
It's
not
up
yet
it
does
take
a
while,
but
on
the
on
the
edge
side
we
have
the
asterisks.
So,
first
of
all,
toronto
installed
itself
on
the
edge
cluster,
so
I
was
able
to
do
that
and
once
it
was
installed
there,
it
could
delegate
workloads
to
it.
So
we
have
the
cnf
already
running.
C
It's
pods.
Everything
is
already
running
there,
and
here
we
have
the
complicated
udp
service
that
tunnels,
the
the
sip
phones
and
the
inventory
is
where
the
it
was
registers.
Oh,
I
think
I
saw
a
change
here,
so
this
was
instantiated.
C
We
don't
yet
have
the
outputs,
so
it
still
starts
trying
to
configure
itself
and
another
interesting
thing
about
configuration.
They
have
access,
because
this
whole
workload
is
one
whole
thing
when,
when
the
asterisk
vnf
needs
to
be
configured,
it
is
able
to
access
data
from
the
rest
right.
So
it
knows
that
it's
part
of
the
service.
So
if
it
needs
to
connect
to
the
data
plane
and
needs
to
know
what
the
address
is,
it
can
ask
it
directly:
it's
just
there
in
the
data.
C
C
C
That
came
up,
so
you
know
any
any.
You
can
use
the
the
basic
api
server
to
access
any
of
these,
so
everybody.
F
C
To
to
the
rest
of
the
workload,
so
I
want
to
actually
connect
phones
right
now.
I
think
I
am
out
of
time
too,
but
I
can
connect
phones
to
both
of
these
pdxs
and
make
phone
calls.
C
C
Yeah-
nobody
I'm
talking
to
myself
here,
okay,
so,
but
did
it
did
register.
C
So
that
is
what
I
wanted
to
show
you
I
I
was
really
hoping
to
discuss
what
the
implications
of
this
could
be
because
the
yeah,
if
you
think
of
what
I
showed
you
here,
I
think
this
is
very
close
to
the
kind
of
service
composition
that
we're
looking
to
do
with
network
services
right
when
we
look
specifically
in
cloud
native
and
kubernetes
right,
we
want
to
be
able
to
to
compose
services
using
network
functions
registered
on
inventories
using
network
services
registers
on
inventory
too.
C
If
you
notice,
they
were
both
there
that
the
difference
between
the
the
layers
was
not.
They
were
just
implemented
in
different
ways,
but
they
were
all
speaking
pasta
and
distance
right
I'll
mention
too
that
there
is
support
here
for
helm,
so
helm
can
be
attached
as
an
artifact.
A
But
if
you're
not
interested
in
having
a
comparable
service
running
on
openstack,
if
you're
just
focused
on
deploying
a
set
of
service
compositions
onto
kubernetes,
could
you
please
speak
to
how
you
see
the
value
of
trinidad
versus
doing
it
in
helm?.
A
C
So
so
so
I'll
say
this:
the.
C
Helm,
I
think
I'll
show
you.
Actually
there
is
a
fact
about
that.
So
it's
you
know.
This
is
a
new
project.
So
it's
the
fact
is
not
a
real
frequently
asked
question.
These
are
questions
I
ask
myself,
but
I'm
not.
C
C
From
red
hat,
this
is
not
an
official
red
hat
product,
it's
not
on
any
kind
of
road
map.
This
is
something
that
some
guy
at
red
hat,
is
working
on
to
try
to
solve
problems.
So
keep
that
in
mind
too,
but
here
it
is
okay.
So
why
use
tuscan
c-stars
instead
of
package
down
charts,
so
this
this
is
where
I
get
very
provocative.
C
The
fact
that
ham
became
really
standardized,
even
though
there
are
many
efforts
to
create
packaging
formats
for
kubernetes,
I'm
sure
some
of
you
tried
some
of
them
and
we
all
kind
of
aligned
on
helm
to
an
extent
because
it
was
the
most
neutral
helm
just
takes
at
its
core
helm,
takes
ginger-like
templates
and
turns
them
into
kubernetes
manifests,
but
for
the
most
part
too,
helm
is
pretty
dumb
about
it.
It's
just
text
files
and
it
just
do
text
templating.
C
Has
not
served
this
well
text
templating
is,
is
one
of
the
worst
ways
to
generate
yaml.
The
ammo
is
a
language
that
is
based
around
indentation
and
in
any
case,
it's
not
validated
at
all
right
until
you
actually
render
the
text
template
into
yaml.
You
don't
know
if
it's
really
there
now.
The
parallel
effort
in
kubernetes,
as
I
talked
about,
was
to
crds,
have
become
much
smarter
right.
There's
the
last
few
versions
of
kubernetes
had
the
open
api
version.
3
validation
schema
right,
so
you
can
actually
give
schemas
to
things.
C
So
that's
a
big
step
forward,
but
tosca
is
very
strictly
tight
from
the
beginning,
so
the
task
of
code
that
I
showed
you
you
know
if,
if
you
make
any
kind
of
mistake
here
in
terms
of
type
or
you
know
all
the
properties
have
their
own
types,
it
won't
compile.
You
would
get
a
very
clear
message
of
you
know
what
what
went
wrong
and
at
what
line.
So
this
is
a
strictly
type
language
strictly
typed
object,
oriented
language,
so
so.
C
A
So
that's
a
I
mean,
that's
that
right,
there
is
a
is
a
clear
distinction.
I'm
curious
if
you
played
with
customize
with
the
k,
yeah.
C
A
Well,
so
I
yeah
so
my
suggestion
to
you
would
be
that
it
would
probably
be
worthwhile
to
flesh
this
out
a
little
bit
more
in
your
faq
and
you
can
treat
it
as
one
of
your
first
initial
external
questions
that
you're
asking,
but.
C
F
I
apologize,
I
got
a
question
as
well,
if
you
don't
mind
so
so
one
of
the
things
that
I've
that
I've
been
looking
at
from
the
enterprise
side.
So
we
know
that
helm
has
significant
drawbacks
and
you
and
you
stay
with
them
very
well.
F
One
of
the
ways
that
we've
gotten
around
some
of
those
problems
is
not
using
helmet
isolation,
but
using
something
like
a
terraform
on
top
in
order
to
to
drive
the
and
crazy
the
helm
templates
and
to
parameterize
them
properly
and
then
deliver
deliver
that,
and
so
in
terms
of
in
in
terms
of
terraform,
with
helm
and
and
truand.helm.
F
Do
you
think
you'd
see
privilege,
or
do
you
think
do
you
think
you
would
see
improvements
by
by
moving
over
to
the
configuration
stuff
to
turn
dot
instead
of
just
having
like
terraform
itself
call
helm
directly.
C
So
terraform
is
obviously
a
very
mature
product,
so
you
get
a
lot
of
other
things
with
it,
but
it's
also
external.
I
think
tornado
is
an
operator
right.
It's
it's
not
running
outside
the
cluster,
it's
a
very
small
executable
when
it
comes
down
to
it,
so
it
just
runs
in
a
very
small
container,
not
a
problem
to
run
it
on
the
edge,
as
you
saw
in
my
demo.
You
know
the
edge
cluster
had
its
own
version
running
so
there's
nothing
external
here.
C
Terraform
will
have
to
be
set
up
as
an
orchestrator
right
outside
it
could
run
in
the
cluster
right.
But
it's
it
is
an
orchestrated
trend
is
not
exactly
an
orchestrator
right.
It
extends
kubernetes
native
orchestration
abilities.
So
I
think
that's
that's
a
more
forward-looking
cloud-native
approach,
I
would
say,
but
of
course
I'm
biased
here.
F
I
I
I
get
an,
I
think:
there's
a
possibility
for
the
three
of
them
to
to
work
together
as
well
like
you
could
have
terraform,
create
and
parameterize
the
initial
tour
and
dot
and
then
use
turn
dot
at
that
point
within
within
a
individual
cluster
as
an
option.
So
I'm
I'm
trying
to
understand
some
of
the
positioning
so
this
this
helps.
C
I
I
think,
you're
exactly
right.
You
know,
there's
a
kind
of
unix
philosophy
going
on
right
here
that
I'm
I'm
not
trying
to
solve
all
the
problems
that
this.
What
toronto
specifically
is
trying
to
do
is
bring
tosca
to
kubernetes.
C
Now,
where
the
tusca
comes
from,
how
you
generate
it,
how
you
design
and
model
everything
is
left
up
to
you
and
an
external
service
to
do
that.
You
know
if,
if
you
know,
products
like
cloudify
cloudify
is
a
whole
orchestrator
based
around
a
tosca
like
dialect
right,
but
it
runs
itself
outside
it
has
a
workflow
engine.
It
can
do
other
things.
200.
has
no
workflow
engine
right.
It
doesn't
try
to
do
workflows
right.
I
think
workflows.
Are
you
know
if
you
want
to
use,
for
example,
argo
workflows?
C
I
have
been
starting
to
work
with
a
profile
for
argo
for
tosca,
so
you
would
be
able
to
define
argo
workflows,
use
using
tasks
as
well
and
just
import
them
into
your
service
template.
If
that's
the
way
you
want
to
go
but
yeah,
this
is
very
kubernetes
native
that
that's
the
intent
and
you
know
two
other.
I
think
the
big
features
that
differentiate
from
helm,
so
one
of
them
is
of
course,
as
I
said,
the
grammat
the
grammar
itself.
This
is
a
strictly
type
object-oriented
language,
but
it's
also.
C
The
service
composition,
I
think
is,
is
another
killer
feature
the
idea
that
you
can
define
types
abstractly
and
have
them
substituted
from
the
inventory
by
policy
right
yeah.
I
I
diverge
here
a
little
bit,
but
if
some
of
you
have
seen
some
of
my
talks
at
ones,
I
talk
a
lot
about
how
the
future
of
orchestration
is
really
policy
oriented
right,
we're
we're
moving
more
and
more
towards
policies,
but
we're
seeing
more
of
that
in
in
the
cloud-native
world
and
kubernetes
as
well.
C
So
policies
are
really
key
here
right
you,
they
are
hints
for
the
orchestration
to
tell
you
what
you
hope
to
get
right
and
then
those
different
policies
are
compiled
together
and
distilled
into
a
final
decision.
That's
that's
made.
Sometimes
that
decision
is
made
by
machine
learning
right,
because
the
policies
give
you
the
guidelines,
the
limitations,
but,
for
example,
I
in
the
policy
here
I
said
that
I
wanted
that
one
of
the
edge
cluster
was
very
small,
so
that
means
that
it
needs
to
fit
a
small
pbx.
C
Now
my
inventory
might
have
several
of
them
and
the
final
decision
might
have
to
do
with
runtime
data.
Being
you
know
what
what
is
the?
What
is
the
throughput
needed
at
that
edge
cluster
right
now
right?
What
if
there's
this
edge?
Cluster
is
near
football
stadium
and
people
are
using
their
5g
phones
to
connect,
and
I
want
to
deploy
something
very
quickly.
The
ai
can
choose,
according
to
runtime
characteristics
which
which
package
and
how
to
configure
it
right
so
that
substitution
can
happen
once
it's
policy-based.
G
Can
I
have
a
question
about
the
the
profiles
and
the
and
the
like
being
object
oriented,
so
you
mentioned
that
that
tosca
doesn't
have
this
simple
profile
more
in
in
2.
zero
version,
but
I
think
one
of
the
big
problems.
What
we
will
run
into
with
using
operators
and
and
helm,
is
that
none
of
these
are
standardized
in
any
sense.
So,
like
everybody,
is
now
having
their
own
great
ideas,
how
their
crds
should
be
structured
and
how
how
to
implement
these
these
crds,
but
all
of
these
are
are
different
ideas.
G
So
is
there
any
way
in
using
tusca
and
all
to
another
to
to
solve
these
problems?
Somehow,
I'm
not
saying
that
we
should
have
profiles
again,
but.
C
C
Yeah,
I
think
you
you're
really
pointing
to
a
growing
problem
that
we're
seeing
right
the
explosion
of
crds
and
operators,
and
let
me
tell
you
too,
a
secondary
problem.
It's
very
hard
to
create
operators.
I
think.
D
A
Is
red
hat
kindly
contributed
last
month,
but,
and
I
think
we
can
agree
that
it
definitely
makes
it
easier
to
create
an
operator
than
starting
from
scratch,
but
definitely
does
not
fall
into
the
easy
category,
and
but
I
I
would
really
sort
of
jump
up
and
down
and
take
the
industry
by
the
lapels
and
say
I'm
not
certain.
That
operator
framework
is
going
to
be
the
kind
of
winning
or
or
most
most
ubiquitous
way
of
deploying
operators.
A
But
I
think
the
operator
pattern
on
kubernetes
absolutely
is
the
future
of
telecoms
that
the
idea
of
trying
to
get
away
from
having
this,
you
know
pdf
playbook,
that
you're
sharing
with
the
operations
team
and
saying
no,
no,
we
can
actually
code
most
of
that
into
software
and
have
our
applications
just
be
smarter
and
update
themselves
and
optimize
themselves.
All
those
things
over
time
using
the
the
kubernetes
crd
patterns,
I
think,
is,
is
incredibly
powerful.
Insight,
yeah.
G
A
A
Yeah,
so
it's
that
composition
issue
last
words
tall,
because
we're
out
of
time
absolutely
so
too,
where,
where
can
we
reach
you
for
more
or
to
chat
more
about
trinidad.
C
Oh
so
I'll
I'll
put
my
email
somewhere
I'll
put
it
on
the
agenda.
If
that's
fine-
and
I
think
it's
already
there
but
yeah.
C
The
link
is
also
on
the
agenda,
so
it's
just
on
github
you're
definitely
welcome
to
play
with
it
there.
I
did
want
to
say
you
know.
Concerning
operators.
You
know
the
big
thing
that
I
keep
trying
to
say
here
that
in
the
etsy
mana
language
you
know,
there's
the
vim
and
kubernetes
is
not
a
vim
kubernetes
plus
operators
equals
an
orchestration
platform.
That's
that's
the
big
point
that
keeps
trying
to
be
made
here
task
is:
it
helps
us
to
model
it.
You
know
in
terms
of
standards.
C
C
C
It
was
very
easy
for
them
to
take
those
models
and
connect
it
to
their
orchestrators.
They
are
using
cloudify
to
make
that
connection
for
them
right,
so
they
did
have
to
write
code.
C
They
had
to
create
the
models
in
tosca,
so
it's
just
knowing
the
yaml
task,
the
language
and
they
did
have
to
know
some
python
to
create
the
connect
connectivity,
the
glue
that
connects
between
those
models
and
the
implementations
and
yeah.
There
is
some
work
to
be
done.
Integration
will
never
disappear.
The
idea
is
just
to
make
it
easier,
and
I
do
think
task
is
a
step
forward.
As
I
said,
if
it
didn't
exist,
we
would
have
to
invent
it.
A
Okay,
well,
let's
stop
there
thanks
everybody
for
the
call
we
are
going
to
be
on
the
slack.
If
you
want
to
chat
more
about
this
and
talking
can
come
on
as
well
and
then
our
next
call
is
a
month
from
now
at
a
more
chinese
friendly
time.
Is
your
friend
time.
A
Can
do
that
if
you're
available
yeah
we'll
put
you
on
the
top
of
the
agenda
for
next
time?
I'm
sorry.
We
ran
over.
D
Yeah,
that's
okay!
Thank
you
then
yeah.
I
can
talk
over
next
time.
Basically
I
mean
it's
more.
Like
a
query.
I
was
trying
to
ask
like
you
know:
if
we
have
these
best
practices
to
our
deployment
of
multi-tenancy
in
you
know,
if
it
already
exists
and
as
we
are
in
the
process
of
deploying,
maybe
we
can,
you
know,
give
some
suggestions
and
add
inputs
to
that.
So
that
was
basically
the
question.