►
From YouTube: Kubernetes WG IoT Edge 20230308
Description
March 8, 2023 meeting of the CNCF IoT Edge Working Group.KCP and KCP Edge discussion. Discussion of KubeCon EUrope and Edge Native Whitepaper status.
A
We
don't
have
much
on
the
agenda
today,
but,
as
usual,
when
that
happens,
we'll
turn
this
into
just
an
open
discussion,
birds
of
a
feather
and
see
what
comes
about.
Sometimes
those
are
the
most
interesting
meetings.
There
is
one
thing,
I
added,
which
is
just
pointing
out
that
kubecon
Europe
is
coming
up
in
about
a
month.
The
actual
date
is,
let's
see,
let's
get
there
April
18.
A
and
to
the
21st,
the
kubecon,
the
kubernetes,
on
the
edge
Day
event,
that
is
a
co-located
event
on
the
first
day
of
that
conference,
has
actually
already
sold
out
it
sold
out
very
quickly.
In
fact,
remarkably
to
me,
it
sold
out
before
we
even
finished
picking
the
speakers
and
agenda,
so
I
have
to
believe
either.
The
topic
is
really
hot
or
people
base
their
decision
on
buying
a
ticket
on
what
those
events
have
been
like
historically
at
past
kubecons,
but
anyway,
I'm
looking
forward
to
that.
A
But
if
you
haven't
already
managed
to
get
yourself
a
ticket
to
that,
it's
already
sold
out.
Unfortunately,
now
that
said,
there's
a
there
are
some
Edge
and
iot
related
sessions
going
on
in
the
main
part
of
the
kubecon
conference
and
in
addition
to
that,
Kate
who
couldn't
be
here
today
has
made
arrangements
to
try
to
get
some
space
for
a
physical
group
meeting
in
the
pre-covered
days
at
kubecon's.
A
This
group
has
actually
had
physical
meetings
where
I
think
sometimes
we
gathered
in
a
room.
Sometimes
we
went
out
to
dinner.
A
Sometimes
we
did
both
and
it's
kind
of
nice
if
there
is
an
opportunity
in
a
critical
mass
for
people
to
get
together
physically,
so
I
just
wanted
to
bring
up
the
idea
that
we
have
an
opportunity
where
there
is
a
room
available
that
we
could
use
the
morning
of
April
20th
and
we
can
do
whatever
we
want
with
that
space
and
time
the
actual
conference
doesn't
start
until
about
10
and
we
could
get
the
room
as
early
as
eight,
so
it
doesn't
necessarily
impinge
on
your
ability
to
go
to
regular
sessions.
A
If
we
do
something
in
that
time
slot,
so
just
throwing
it
out
there
I'm,
not
even
sure.
A
Maybe
we
can
just
have
a
virtual
Zoom
show
of
hands
of
people
who
think
they
might
be
there
for
the
event,
because
if
nobody's
going
to
be
there,
it's
kind
of
moved,
although
I
intend
to
be
there
and
so
does
Kate
anybody
else,.
A
Looks
like
maybe
no,
but
if
things
change,
why
don't
you
ping
us
on
slack
and
we'll
try
to
see
what
we
can
do
well
we're
there?
Let
me
just
take
a
quick
look
and
see:
okay,
there's
one
other
item
that
got
added
to
the
agenda
by
Brandon
and
that's
Edge
native
white
paper
status.
Let
me
make
you
a
co-host
brand
and
in
case
you
wanted
to
share
something
and
you're
welcome
to
take
over.
B
All
right,
thank
you,
Stephen.
This
will
be
fairly
brief,
but
it's
been
a
I
think
a
couple
months
since
I've
been
able
to
make
this
call
I've
had
a
conflict
that
did
not
exist
today.
So
I
wanted
to
check
in
here
with
this
group,
I
think
Kate
has
been
making
some
updates
on
my
behalf
on
the
last
few
calls,
and
unfortunately,
I
don't
have
a
Ultra,
definitive
answer
on
the
publication
and
the
go
forth
plan,
but
I
can
tell
you
that
we're
we're
getting
very
close.
B
B
They
also
have
agreed
to
promote
this
new
document
in
a
blog
post,
which
they
asked
me
to
write.
Just
a
two
paragraph
post
introducing
the
work
and
the
and
the
piece
that'll
be
going
live
shortly
on
the
cncf,
blog
and
they'll
do
a
feature
as
well.
In
their
newsletter
called
kubecon
weekly,
there
was
an
open
question
whether
we
should
had
a
hold
off
on
publishing
the
paper
until
kubecon
Europe,
but
we
just
decided
that
it
wasn't
worth
waiting
longer
since
we've.
B
This
has
been
in
motion
quite
a
quite
a
ways
and
in
fact
it
it
might
kind
of
get
lost
if
it
were
caught
up
in
the
the
all
the
kubecon
announcements,
but
they
are
including
it
as
a
bullet
item
in
the
Press
package
and
if,
when
speaking
with
press
and
analysts,
if
anyone
wants
to
do
a
briefing,
I
told
them
that
the
authors
of
the
paper
would
be
open
to
do
a
briefing
virtually
seeing
as
it
would
be
hard
to
get
everyone
in
the
same
place.
B
Unfortunately,
myself
or
Amar
will
not
be
at
kubecon
Europe
this
year.
So
I
should
know
more
about
that
here,
hopefully
this
week
or
next
week
by
the
latest,
sorry
for
just
keeping
that
dragging
on
and
on
it's
taking
a
while
to
get
through
their
creative
Cycles.
B
The
other
related
update
I
had
to
the
paper,
was
I've,
been
speaking
with
somebody
at
Cisco,
Frank
brockner's,
who
I've
known
for
many
years
through
his
participation
in
the
LF
networking
project,
world
and
building
on
a
conversation
we
had
in
Seattle
at
one
Summit
last
year.
He
he
shared
the
the
GitHub
version
of
the
paper
with
some
colleagues
at
Cisco
and
they
have
voiced
an
interest
to
help
evolve.
The
document
and
I'm
pulling
this
line
from
him
from
an
email.
B
We're
scheduling
a
call
next
week
to
discuss
further,
but
their
initial
idea
was
now
that
we
have
it
defined
what
the
qualities
of
edge
native
are.
His
idea
would
be
to
perhaps
an
indexed
revision
or
Evolution
or
an
add-on
think
about
what
educative
means
from
a
developer
and
architecture
perspective.
B
So
that's
all
the
information
I
have
with
his
idea
now.
I'll
know
more
when
I
speak
with
him
and
his
team
next
week,
but
I
wanted
to.
Let
you
all
know
that
that
that
idea
has
been
proposed.
He
wanted
to
chat
about
it
with
me
before
bringing
it
to
this
group.
So
if
there's
any
particular
questions,
you
want
me
to
ask
or
to
find
out
going
into
that
conversation,
please
let
me
know
otherwise.
I'll
I'll
hear
what
they
have
to
say.
Take
some
notes
share
that
back
with
the
group
and
I
think.
B
The
next
logical
step
would
then
be
to
ask
them
to
to
join
this
call
in
the
near
future
and
present
their
idea
for
discussion.
A
One
thought
Brandon
on
that
and
I
think
this
came
up
well,
we
were
putting
together
this
whole
Edge
native
white
paper
that
we
wanted
this
to
be
kind
of
succinct,
meaning,
like
paper
usually
implies.
This
is
like
a
quick
read
thing,
like
maybe
three
pages,
and
that
if
we
Branch
off
into
these
things
like
developer
perspectives,
architectures
best
practices
it
it's
additive
and
I'm,
not
saying
that
this
one
act
would
convert
it
to
a
book.
A
A
That
is
the
overview,
and
then
we
have
independent
efforts
to
pursue
Edge
and
iot
related
niches
with
focuses
on
specific
areas
and
I
think
those
are
maybe
easier
to
consume
and
easier
to
keep
updated.
You
know,
there's
always
a
sell
by
date
on
this
and
any
monolithic
paper
just
gets
held
down
by
the
slowest
thing
like.
If
we
cover
three
topics
and
one
of
those
topics,
we
just
can't
get
enough
reviews
and
stuff,
and
it
keeps
the
next
revision
of
the
white
paper
from
going
out.
A
B
I
for
one
like
the
idea
and
I
agree
with
the
logic,
there
definitely
seems
more
streamlined
and
less
dependent
on
other
variables
with
the
idea
of
of
an
add-on
and,
however,
we
want
to
structure
it
in
terms
of
a
hierarchy
or
whatever.
We
want
to
call
it.
A
supplementary,
separate
piece
versus
a
rewrite
makes
sense
to
me,
but
I'd
be
curious
to
hear
thoughts
from
anyone
else.
C
I
had
some
thoughts
on
motivation,
so
I
just
wanted
to
give
them
here
so
Brandon
I'm
also
on
those
emails,
so
in
full
disclosure
just
wanted
to.
Let
you
know
that
I
bumped
into
Frank
at
he
had
given
a
talk
at
I,
believe
it
was
kubecon
last
fall
and
he
wasn't
the
one
who
gave
the
talk.
C
It
was
one
of
his
colleagues
because
he
wasn't
available
and
immediately
I
saw
what
Great
Bear
was
the
product
that
Cisco's
working
on
to
deliver
and
manage
Edge
locations
and
I
thought
it
was
an
incredible
Synergy.
What
we're
doing
in
kcp,
Edge,
so
I
had
some
informal
discussions
with
him
and
said
you
know:
how
do
we
get
to
work
together
and
Cisco's
policy?
Of
course,
is
they
fail
closed?
C
They
don't
they're,
not
really
an
open
source
company,
but
one
area
that
they
did
agree
that
they
would
be
able
to
work
with
us
in
in
the
open
was
in
this
working
group,
and
so
this
that's
the
precipitous
or
the
you
know
the
background.
Motivation
behind
that
set
of
queries
that
came
to
you
Brandon,
is
unfortunately
I'm.
C
I
was
the
one
who
kind
of
initiated
and
prodded
that
so
apologize
for
not
having
knowing
you
earlier
and
being
able
to
fill
you
in,
but
yeah,
then
to
dovetail
on
the
back
of
what
you've
suggested
in
terms
of
supplementary
material.
I.
Think
that's
I
think
it's
a
fair
way
to
be
inclusive,
as
well
as
without
diluting
the
initial
effort,
but
it
would
be
nice
I
I
would
I
would
add
this.
C
One
compromise
if
there
were
any,
is
that,
if
at
all
possible,
to
revisit
the
initial
index
so
to
speak,
to
update
it
with
references
to
other
materials,
so
that
it
would
be
useful
as
an
as
an
index
and
then
therefore,
as
a
way
to
to
connect
those
documents.
Once
they're
made
available.
A
Yeah
I
agree
with
that
that
we
we
need
a
topmost
reference
point
that
just
covers
it
generally
and
leads
you
into
the
points
out.
What
else
is
available
and
links
to
them
rather
than
having
it
be,
like
I,
say,
monolithic
right.
C
B
It
does
completely,
and
so
yeah
thanks
for
piping
up
nice
to
meet
you
Andy
and.
B
Think
this
is
the
the
right
group
to
be
having
this
discussion.
I
I
I'd
still
be
happy
to
meet
with
you
and
Frank
next
week,
offline
to
flesh
out
the
the
idea
a
little
bit
more,
but
would
you
and
and
Frank
or
others
be
willing
to
join
this
call
in
in
two
weeks
to
continue
the
discussion?
Well.
A
The
the
other
thing
Brandon
I'd
like
to
throw
out
there
is
I'm,
not
that
familiar
with
this
Cisco
did
you
say
it
was
great
Bear
yeah.
A
I'd
love
to
have
them
present
to
this
guy,
because
I'm
curious,
you
know
we're
always
open
to
those
and
I
think
we
get
better
attendance
when
we
have,
we
can
advertise
and
promote
talks
on
particular
topics.
The
only
thing
is,
if
it
is
entirely
closed,
Source
by
rules
of
the
cncf,
it
maybe
is,
is
not
proper
subject
matter,
however,
even
things
that
are
commercial
if
they
can
be
wrapped
into
here's
a
tool
that
could
be
used
generically
with
kubernetes
or
with
whatever
I
I
I
think
that
we
can
find
room
for
it.
A
B
A
E
Would
it
be
an
appropriate
approach
to
you
know,
invite
them
to
give
their
views
on
the
technology
issues?
You
know,
apart
from
promoting
the
specific
product,
just
saying
this
is
the
way
we
think
you
know
we
the
way
we
think
about
the
problem
space.
You
know
what
we
think
is
a
useful
approach
to
the
problem.
That
kind
of
thing.
A
A
Kubernetes,
there's
I,
don't
know
60
plus
commercial
distributions
they're
out
there.
Everybody
knows
it
saying
they
exist
is
fine,
but
giving
a
canned
pitch
that
tries
to
sell
this
particular
one
to
somebody
that
that
isn't
the
best
use
of
one
of
these
group
meetings
but
acknowledgment
that
yeah
there
is
a
commercial
funding
behind
this
I.
Don't
even
have
trouble
with
that.
C
Yeah
and
you
know
what
I
don't
think
you'll
get
much
pushback
I,
don't
think
they
really
want
to
go
into
the
details
of
of
their
project.
You
know
in
in
the
closed
space
right,
because,
just
because
the
commercial
nature
of
it
right
so
I
don't
think
you'll
get
very
much
of
it
in
the
way
of
a
deep
dive,
but
I
agree
that
shouldn't
be
used
for
marketing
either.
Okay,.
A
So
anyway,
Brandon,
if
you're
talking
to
them,
why
don't
you
throw
that
out
there
because
it
sounds
like
they
might
want
to
come,
but
as
long
as
they're
coming,
it
would
actually
help
as
a
meet
and
greet
for
them
to
introduce
themselves
with
the
you
know,
the
background
on.
A
Is
Andy
by
the
way
you're
welcome
to
expand
on
kcp
too
I
caught
the
last
community
meeting
on
YouTube,
and
you
know
you're
even
welcome
to
throw
out
references
to
those
it
looked
like
it
was
sort
of
at
this
stage.
Still
growing
and
kind
of
my
wrap-up
would
be
inside
poker
where
it's
kind
of
the
people
on
it
rather
than
building
Community,
but
you
got
to
start
somewhere
and
I
found
I
found
your
meeting
interesting
as
a
lurker.
A
Not
joining
live,
but
somehow
you
know
YouTube
gives
recommendations
that
have
popped
it
into
something
on
the
list
of
something
I
might
like,
and
I
watched
it
and
I
did
and
watching
it.
I
realized
that,
with
regard
to
kcp
I
have
a
lot
to
learn.
There
were
there
were
buzzwords
and
things
that
are
definitely
things
I'm
interested
in,
but
I
think
you
have
given
a
little
background
and
I
think
you
may
be
slated
to
give
some
more
at
kubernetes
on
the
edge
day.
A
But
not
everybody
you
know
will
be
at
a
physical
event
like
that.
So
you're
welcome
to
sign
yourself
up
for
a
future
date
and
even
cross-pollinate
it
since
I
know
you're
operating
a
kcp
group,
so
you
you
know
part
of
that.
You
could
build
awareness
that
that
even
exists
for
people
who
want
to
first
take
an
approach
of
seeing
what
kcp
is
and
whether
it's
useful
to
them,
and
maybe
the
Second
Step
would
be.
You
know
the
deeper
dive
into
going
to
a
you
know,
one
of
one
of
its
Community
meetings,
yeah.
C
A
C
E
Okay,
briefly,
kcp
is
a
project
that
is
doing
I.
Several
things,
I
I,
think
of
it
in
layers.
They
generally
think
of
it
in
layers.
E
E
The
lowest
most
Central
layer
is
about
virtualizing,
adding
an
extra
layer
of
virtualization
in
the
Kube
API
server,
and
this
has
also
been
called
things
like
higher
level
namespaces
or
super
namespaces
So
within
one
API
server
you
can
offer
it
can
offer
the
the
you
know,
basically,
a
bunch
of
independent
virtual
API
servers,
so
each
one
can
have
its
own
set
of
crds,
for
example,
and
I
specifically
call
this
virtualization
rather
than
isolation,
because
it's
not
complete
isolation.
E
It
is
just
virtualizing
some
things
and
giving
an
extra
layer
of
organization
or
separation
in
in
some
matters.
Next
layer
up
is
a
particular
user
interface
to
that
called
works.
So
they
call
those
virtual
virtual
that
the
the
word
they
use
there
is
logical
clusters,
they
say
one
server
supports
multiple
logical
clusters,
but
the
using
cluster
there
is
actually
very
misleading.
It's
not
a
cluster.
It's
just
part
of
the
namespace,
the
next
layer
up.
They
call
workspaces,
it's
a
user
interface
to
logical
clusters.
E
It
is
a
particular
way
of
interfacing
to
them.
Next
layer
up
is
API,
export
and
binding.
You
may
know:
Stefan
shamansky
also
presented
a
version
of
this
that
was
separated
from
kcp
at
a
rest
kubecon.
He
called
it
Cube
bind
and
in
kcp
they
have
a
version
of
it.
E
That's
integrated
with
workspaces
and
logical
clusters,
and
then
next
layer
up
and
most
easily
separated
is
something
they
call
transparent,
multi-cluster,
which
is
a
use
of
all
those
other
Technologies
and
and
added
detect
and,
of
course,
some
more
technology
to
enable
basically
kubernetes
as
a
service
or
a
multi-user
kubernetes.
So
the
idea
is
each
user.
You
know,
has
the
illusion
that
they
have
their
own
coup
cluster,
but
in
fact
the
way
it
actually
works
is
there's
a
kcp
server
with
workspaces.
E
You
know
each
Works,
each
user
gets
its
own
workspace
in
which
he
puts
his
these
objects,
API
objects,
the
containers
actually
run
elsewhere.
They
have
they
use
so-called
back-end
clusters
or
p-clust
physical
clusters
or
compute
clusters
and
the
transparent
multi-cluster
mechanism.
E
Propagates
the
necessary
objects
into
the
compute
clusters
to
actually
run
the
containers,
while,
while
connecting
those
containers
back
to
the
original
workspace
API
service,
so
they
have
the
view
of
their
Source
API
server.
E
So
very
briefly,
another
interesting
thing
about
it
is
they've
produced
a
unified
binary
that
includes
the
API
server
and
the
essential
controllers
from
controller
manager
and
an
NCD
server,
and
it
runs
with
a
single
server
at
CD
cluster.
So
it's
a
single
server
CD
cluster.
So
that's
a
very
brief
summary.
A
One
one
question:
expansion:
if
you
will,
when
you
say
multi-cluster,
does
this
support
dispersed
clusters,
so
these
could
be
at
different
physical
locations
or
is
the
the
angle
here
that
they'd
all.
E
Be
well
first,
I
should
admit
that
one
of
the
things
that
I
left
out
is
they
do
also
talk
about
sharding
for
Geographic
locality.
So,
instead
of
one
kcp
server
total,
you
can
have
like
one
in
each
region.
So
that's
one
level
of
dispersion
and
that's
part
of
an
answer
to
your
question.
The
other
part
of
the
answer
is
these
compute
clusters.
E
You
know
they're
distinct
from
the
kcp
servers,
so
they
can
be
distant.
What
they
cannot
be
is
a
will
very
successfully
or
happily
is
disconnected
for
a
long
time.
But
a
really
long
time
is
a
configured
batter
since
it
they
use.
You
know
kubernetes
controllers,
to
connect
them,
so
it's
a
state-based
paradigm.
So
essentially
it
can
tolerate
disconnection,
but
there's
a
health
checking
that
is
configured
with
a
timeout.
E
So
if
something
is
disconnected
for
longer
than
that
Health
period
it's
considered
broken,
and
the
idea
here
is
that
there
is
a
compute
service
provider
who
provides
a
collection
of
compute
clusters.
The
the
users
of
this
virtual
kubernetes
clusters.
Don't
know
about
that.
So
if
a
user
says
you
know,
I
want
to
say
a
deployment
object
with
three
replicas.
E
E
A
E
Oh,
you
should
remember
one
other
thing
that
I
think
I've
meant
to
mention
about
disconnected
operation
right,
because
TMC
connects
the
containers
back
to
the
original
API
server.
While
there's
no
connectivity,
those
containers
are
going
to
have
trouble
accessing
the
API
server
so
again,
depending
on
how
they're
doing
what
they're
doing
with
it.
You
know
they
may
or
may
not
have
grief
with
that.
So
anyway,
yes,
let's
get
back
to
the
larger
issues.
A
E
C
Yeah
so
kcp
Edge
takes
it
a
bit
further
we're
we're
concerned
with
placement,
so
location
where
we're
going
to
place
workloads,
and
we
have
the
notion
of
you-
know-
put
placing
workloads
at
many
locations,
not
just
any
location,
so
instead
of
one
to
any
you're
going
one
to
many
and
that's
an
extension
of
what
TMC
does
today
in
kcp
and
so
we're
for
the
edge
we're
taking
it
one
step
further
and
going
one
to
many
and
then
from
there
we
can
scope
those
back
with
predicates
and
filters
and
and
and
so
forth,
within
the
the
definition
of
the
workload.
C
So
that's
where
we're
taking
it
we're
working
on
also
scale
a
scalability
issues.
We
want
this
to
be
a
consistent
interface.
You
can
use
it
across
heterogeneous
Edge
locations,
so
different
flavors
of
those,
and
we
want
it
to
be
done
at
scale.
So
that's
where
kcp
Edge
is
really
honing.
In
the
demo
day
that
you
saw
potentially
Steve
on
our
YouTube
channel
or
some
of
the
community
meetings,
we
had
a
demo
day
where
we
did
some
work.
C
We
with
Argo
CD
scalability
so
understanding
where
those
bottlenecks
are
so
we're,
starting
to
work
with
the
Argo
CD
scale,
Sig
that
they're
spinning
up
and
some
of
the
folks
there.
So
they've
they've
caught
some
interest
and
we
also
did
something
on
using
openai
GPT
models
to
manipulate
the
Declaration
of
kubernetes
Declaration
for
our
workload
or
deployment
and
then
changing
it
to
adapt
to
different
heterogeneous
locations.
C
So
that's
pretty
cool,
and
so
we
realized
that
when
we
get
to
a
certain
level
of
scale
where
we're
going
with
kcp
Edge
is
that
people
don't
want
to
do
the
operations
they
need
to
to
change
where
workloads
will
be
deposited
by
hand.
So
we're
looking
at
other
ways
to
do
that
through
automation,
things
like
open,
AI
came
to
mind.
C
We
have
an
internal
product
at
IBM
that
is
called
turbonomic,
that
does
placement
decisions
based
on
metrics
and
and
so
forth,
and
we're
also
working
with
the
observability
folks
for
the
ability
to
detect
when
there's
duress
or
maybe
there's
carbon
footprint
or
cves.
That
are
aware
that,
where
we
don't
want
to
put
workloads,
so
all
those
things
come
in
into
play.
So
it's
kind
of
taking
kcp
one
step
further
than
where
it
is
today
and
bringing
it
into
Edge
location,
specific
concerns,
foreign.
A
C
You're
right
we're
shouting,
Into
The
Ether
right
now,
Steve.
You
know
where,
as
you
know,
that
many
of
these
projects
go
through
that
phase
and
we're
starting
to
we're
trying
to
develop
a
certain
level
of
maturity.
If
we
can
I
dropped
a
link
in
there
to
to
our
site
where
our
docs
are
held
and
our
GitHub
is,
is
also
linked
there,
as
well
as
our
medium
site.
So
we've
done
a
lot
of
blogging
on
this.
Also.
A
A
A
Essentially,
I'll
call
it
a
control
plane
if
you
will,
or
a
control
plane
with
virtualization
features
on
top
of
underlying
actual
implementations
of
kubernetes
clusters,
and
is
is
this
versatile
enough
that
you
could
do
something
as
Maybe
strange
or
maybe
this
isn't
strange
as
Landing
this
on
kubernetes
input
clusters
implemented
by
different
Cloud
providers,
like
suppose,
as
an
end
user,
I
wanted
to
consume
for
some
strange
reason
from
both
Amazon's
eks
kubernetes,
azures
kubernetes,
and
let's
throw
Google
in
there
too,
to
recognize
all
the
big
ones
and
not
to
leave
anybody
on
digital
ocean
and
whoever
else
anyway,
pick
three
lay
this
on
top
and
then
you
what
you
operate
with
looks
like
a
virtual
kubernetes
to
you.
A
E
E
E
In
edgemc
we
have
the
idea,
there's
a
center
and
then
multiple
clusters
out
at
the
edge
and
that's
one
use
of
multi-cluster
and
those
clusters
out
at
the
edge
they
could
be
in
different
clouds.
We
do
talk
about
multi-cloud,
and
so
one
of
the
problems
is:
if
you've
got,
you
know
some
some
workload
or
collection
of
workloads.
You
want
to
spread
across
clouds
and,
if
obviously
got
a
problem
with
managing
that.
So
you
know
the
multi-cluster
problem
or
multi-cluster
management
problem
is
part
of
the
multi-cloud
problem,
also
in
the
center.
E
We
have
this
problem
of
you
know
exactly
because
we
have
this
vision
of
in
the
center.
You
describe
your
workload
or
prescribe
your
workload.
It
gets
copied
and
instantiated
at
all
the
places
where
you
want
it
to
go
and
it
actually
runs
out
on
the
edge
clusters
and
then
status.
Information
from
the
edge
is
is
propagated
back
and
summarized
back
into
the
center.
So
in
the
center
you've
got
this
description,
which
is
basically
a
desired
State
and
a
summary
of
reported
state.
E
These
have
to
you
know.
These
are
basically
multiple
workload
descriptions
they
have
to
be
held
in
something,
and
there
are
basically
two
approaches
I've
seen
out
in
the
world,
one
is
to
define
a
container
object
type
and
the
other
is
to
use
one
of
these
logical
clusters.
You
know
basically
a
denatured,
API
server,
one
sort
or
another
as
the
container
for
a
workload
description
that
is
also
multi-cluster,
because
you
in
general
may
have
a
multiplicity
of
workloads.
E
You
may
want
them
to
not
all
go
in
the
same
denatured,
API
server,
so
there's
two
sorts
of
multi-cluster
they're,
very
different
they're,
both
involved
in
our
story,
as
I
said
the
first
one
obviously
is
involved
in
multi-cloud
the
second
one
you
may
have
been
trying
to
touch
on
this
and
even
if
you
didn't
I'll
bring
it
up
anyway.
E
The
question
is,
you
know:
can
this
concept
of
a
a
fleet
of
denatured,
API
servers
or
Services?
Really?
The
point
is
from
a
client's
point
of
view.
It's
a
service
doesn't
matter
whether
it's
coming
from
one
server
or
a
collection
to
servers,
and
that's
really
the
important
Point
here.
Can
we
come
up
with
a
client-side
abstraction,
for
you
know
dealing
with
one
or
a
few
or
all
of
a
fleet
of
API
services
and
maybe
even
and
allow
for
different
implementations.
E
So
in
last
week's
kcp
community
meeting,
you
saw
a
guy
actually
working
on
controller
runtime,
trying
to
introduce
this
in
the
context
of
controller
runtime,
an
abstraction
for
a
fleet
of
logical
clusters.
His
first
implementation
was
actually
based
on
kind
clusters
rather
than
kcp
workspaces,
but
I
think
that's.
You
know.
A
very
interesting
question
that
this
begs
I
think
it
should
be
answered
for
kubernetes
in
general,
not
just
controller
runtime,
so
I'll
stop
there.
A
Okay,
I
think
you
did
answer
it
and
it
sounds
like
a
really
cool
idea,
although
sometimes
really
cool
ideas
bring
along
baggage
too.
So
my
The
Geek
in
me
says
that
I'd
love
to
get
my
hands
on
and
start
playing
with
this
kind
of
thing
and
push
the
envelope.
But,
having
seen
you
know,
these
layered
approaches,
like
layers
of,
say,
running
kubernetes
on
top
of
a
hypervisor
on
top
of
physical
infrastructure,
the
it
looks
cool
when
everything
is
working.
A
A
hundred
percent,
but
the
downside
that
has
been
communicated
to
me
by
users
are
what
happens
when
things
go
wrong,
where
something
is
quote,
slow
or
doesn't
work
at
all.
Do
you
need
somebody
who
understands
every
one
of
these
layers
in
order
to
troubleshoot
and
figure
out?
You
know
where
the
bottleneck
is
or
something
so,
I
don't
have
the
answer,
but.
A
E
In
short,
I'll
just
answer
I
think
yeah
in
terms
of
functionality.
You
know
we're
dealing
with
stuff,
that's
state-based
and
resilient,
so
that
actually
has
kind
of
a
downside
in
that
it
tends
to
hide
bugs
you
can
test
stuff
and
not
even
notice
that
a
lot
of
stuff's
going
wrong,
but
when
you
start
to
look
at
performance
yeah
in
my
experience
you
kind
of
have
to
be
able
to
look
at
all
the
layers,
because
you
can
have
performance
problems
at
any
layer
and
in
interactions
between
layers.
E
So
anyway,
that's
that's
a
whole
topic.
You
know
this
as
far
as
I'm
concerned.
There's
no
free
lunch.
We
are
building
big
piles
of
stuff
of
one
sort
or
another.
There's
always
questions
of
what
ghost
happens
when
some
you
know
what
goes
bad
when
something
breaks
or
slows
down
all
the
answers.
You
know
all
the
architectures.
You
know
that
there's
there's
issues
there.
Some
are
better
for
dealing
with
it
than
others.
I
think
you
know
the
more
real.
The
separation
is,
the
the
more
resilience
you
get.
But
that's
that's.
A
Yeah,
thankfully,
we
have
tools
in
the
cloud
native
Community
for
observability
and
tracing
that
might
be
out
applicant.
We
are
to
give
you
a
chance
to
put
something
together
to
to
keep
something
like
that.
That's
multi-layer,
multiple
abstraction
layers
in
a
situation
where
you
can
at
least
find
your
way
to
the
next
layer.
D
Is
a
small
small
ask,
so
I've
been
asked
for
by
the
cube
by
example:
Community
there
they're
doing
like
a
live
stream
shows
for
for
the
kubernetes
community,
I'll
be
there
for
the
next
show
talking
generally
about
Edge
Computing
and
this
working
group,
but
they
would
like
to
hear
somebody
from
the
cube
Edge
to
to
to
also
you
know,
represent
their
project,
and
previously
Cindy
was
here
right.
She
left
I
tried
to
contact
Yin.
He
was
also
the
manager,
but
I
think
he
also
Departed
the
project.
So
just
wanted
to
see.
A
I
have
a
way
you
could
find
somebody,
and
that
is
I
noticed
I,
believe
there's
a
cube,
Edge
talk
going
on
at
kubecon
Europe,
so
you
could
look
at
whoever
is
giving
that
talk
and
that
would
seem
like
that
would
be
a
good
person
right
and
then
qbedge
now
I
believe
is
a
cncf
project,
and
sometimes
these
groups,
even
our
own,
don't
necessarily
keep
their
GitHub
page
fully
updated.
A
But
there
is
supposed
to
be
a
list
of
who
the
organizers
are
of
that
group,
and
that
typically
includes
at
least
GitHub
handles
and
often
emails,
yeah.
D
A
Yeah
I
I
haven't
kept
tabs
on
it.
If
you
want
I'll
step
away,
I'll
I'll,
look
I'll,
get
you
a
name
in
the
next
two
minutes
by
looking
at
the
kubecon
schedule,
but
perfect
yeah.
A
In
the
meantime,
while
I
go
off
on,
do
the
and
do
that
this
is
still
birds
of
a
feather,
so
anybody
else
want
to
throw
anything
out
there.
Even
topics
as
broad
as
hey.
Just
like
you
know,
I
go
there
and
look
at
YouTube
videos
on
and
call
it
the
kcp
while
I
was
eating
lunch.
If
anybody
else
has
just
found
an
interesting
blog
or
a
video
lately,
throw
it
out
there.
D
So,
regarding
cubich
yeah,
so
I
found
I
also
found
Kevin
as
well,
which
is
on
the
on
the
on
the
qubit
stock
and
Ian
is
representing
so
yeah,
so
I'm
all
good
on
that
front.
Just
just
haven't
been
able
to
contact
anybody
in
person
thanks.
Okay,.
D
A
A
F
F
D
F
D
A
You
know
here:
I,
I,
think
you'll
hope,
you'll
Forgive,
Me
Andy,
but
Andy.
You
dm'd
me
in
the
chat
asking
if
I
had
any
tips
on
building
Community.
Do
you
mind
if
we
just
do
this
out
in
the
open
in
the
meeting?
I,
don't
mind.
A
To
declare
myself
to
be
an
authority
on
building
Community,
although
I
have
been
a
lead
or
a
co-chair
of
this
group
and
I
do
another
kubernetes
user
group
for
kubernetes
users
running
on
the
VMware
hypervisors
and
I've
been
doing
that
for
years.
So
I've
got
experience,
maybe
not
perfect
success,
but
some
of
the
tips
I'll
have
and
Dion
can
back
me
on
this.
Is
you
know
they
start
slow?
They
don't
change
your
meeting
time
we'll.
We
learn
that
the
hard
time,
because.
F
A
Write
down
what
time
and
day
it
is
and
kind
of
pick
one
and
stick
with
it,
although,
when
covet
hit,
I
think
a
lot
of
people
started
joining
remotely
and
they
watched
the
recordings
and
in
a
way
it's
unfortunate
because
there's
some
real
value
to
the
interactivity
of
people
getting
there
and
actually
chatting
and
people
got
used
to
this
idea
that
there
was
no
sense
of
urgency
to
participate
live,
but
it
meant
that
I
would
contend
that
all
of
these
regular
Community
meetings
with
zoom
had
big
attendance
drops
when
covet
hit
and
they
they
haven't
come
back,
and
there
was
a
time
when
I
was
in
some
of
these
meetings
that
you
would
have
close
to
100
people.
A
It's
almost
too
many
when
it
gets
to
that
level.
But
those
days
seem
to
be
over
whether
they'll
ever
come
back
remains
to
be
seen.
I
think
you
get
better
attendance.
If
you
can
communicate
an
agenda
up
front
where
people
can
shop
and
choose
you
know
what's
there
and
maybe
you
can
promote
it
a
little
and
cross-pollinate
with
different
groups.
A
You
know,
because
the
only
way
I
became
aware
that
your
group
even
existed
was
that
whatever
the
YouTube,
probably
some
AI
bot
or
something
tries
to
discern
what
you
might
be
interested
in
it's
wrong
more
often
than
it's
right
but
I.
Just
the
the
caption
sounded
interesting.
A
So
I
tried
it
and
just
trying
to
build
up
awareness
first
rather
than
you
know,
grade
yourself
on
how
many
bodies
actually
show
up
and
I'm,
not
sure
you
can
even
tell
how
many
really
showed
up,
because
when,
when
you're
recording
your
meetings
on
YouTube,
people
can
voluntarily
hit
that
thumbs
up,
saying
they
liked
it.
But
I'm
not
sure
how
reliable
some
of
these
other
metrics
are.
So
that
you
even
have
any
earthly
idea
of.
You
know
whether
you're
some
guy
giving
a
speech
out
in
the
woods
and
nobody's.
A
Go
on
the
conference
circuit
and
just
get
word
out.
There
I
think
the
yeah
I
I
think
that
would
be
the
summary
of
the
advice
get
get
out
there
on
a
consistent
schedule,
publish
your
agenda
in
advance.
If
you
can
get
speakers
and
things
because
people
have
Niche
interests
in
it,
all
you
have
to
do
is
catch
one
and
then,
if
they
like
the
thing,
they'll
come
back.
Perhaps
yeah.
D
C
F
A
Any
ideas
on
helping
build
an
open
source
project
related,
Community,
good
or
bad
you're,
even
free
to
take
shots
at
how
this
group
is
operated.
I
I
got
a
thick
skin
and
everybody
has
room
for
improvement.
D
C
C
D
C
Take
so
that
the
word
you
know
that
the
word
Bots
pick
up
and
start
to
proliferate,
like
maybe
that's,
maybe
how
it
got
in
front
of
you.
I.
A
A
And
but
that's
I'm
just
pointing
out
that's
how
I
discovered
it
even
existed,
otherwise
I
wouldn't
have
had
any
clue.
I,
don't
think
it
it
awareness
of
that
didn't
hit
through
any
other
channel,
but
that,
for
me,
that's
interesting.
C
So
yeah
I'd
pay
a
lot
of
attention
to
things
like
word:
clouds
right
tags,
those
types
of
things
because
I
know
you
know
they're
all
that's
what
they're
using
as
metadata.
You
know
and
scraping
that
kind
of
thing,
and
you
know,
we've
paid
a
lot
of
attention,
I
think
also
to
the
graph
and
what
I
mean
by
that
is.
C
If
you
look
at
any
of
our
medium
posts,
they're
all
interconnected
with
other
colleagues
or
community
members
that
have
also
written
something
so
we're
making
sure
that
everything's,
interconnected
and
and
is
grounded
as
far
as
an
agenda
is
concerned.
What
is
your
take
on?
You
know
in
our
community
we
use
the
agenda
that
I
post
for
the
community
meeting
in
an
issue
in
the
GitHub
repo,
but
I
see
other
people
using
like
just
a
Google,
doc
and
I.
Think
you
do
use
that
too
for
an
agenda
yeah.
C
What
do
you
think
is
the
most
useful
for
a
community
to
Rally
around,
which
is,
which
is
the
one
you've
had
better
success
with
well
I.
A
Think
unless
your
community
is
devs,
GitHub
is
maybe
problematic,
because
people
aren't
used
to
using
it
as
opposed
I.
I
think
the
Google
Doc
is
more
democratic,
especially.
The
fact
is
that
some
of
the
better
presenters
are
not
the
software
Engineers
who
aren't
used
to
public
speaking.
Sometimes
it
crosses
over
into
people
who
maybe
weren't,
wear
more
of
a
marketing
hat
right
and
they're
more
comfortable
with
that
Google
doc.
A
In
my
experience,
got
it
well
great
yeah
the
Google
Doc,
however
I'll
give
you
the
downside
of
it
that
I've
been
in
some
open
source
projects
where
there
are
unfortunately
trolls
out
there
on
the
internet,
who
will
deface
those
so
putting
it
out
edit
anywhere
may
or
may
not
work
out.
Well,
for
you,
the
nice
thing
is
a
Google.
A
Doc
can
be
rolled
back,
but
in
this
group
we
do
have
theoretically
some
control
in
that
you're
supposed
to
join
the
Google
group,
and
then
you
get
edit
access
automatically
if
you're
logged
into
the
same
Google
account
you
use
to
join
the
group,
so
maybe
putting
it
out
there
literally
open
season
for
any
everything
is
asking
for
trouble
to
materialize.
Downstream.
C
That's
good
advice:
yeah,
we've!
We've
got
a
a
way
to
do
that
and
we've
used
it
for
our
own
documents
internally.
So
I
think
that
that's
safe
to
assume
that
we
could
probably
do
something
similar,
I'll,
I'm
gonna
go
ahead
and
that's
actually
one
action.
C
I'm
going
to
take
after
this
meeting
is
to
create
an
agenda
in
a
Google
doc,
because
I
I
can
I'm
starting
to
see
that
more
and
more
to
be
the
status
quo
and,
as
you
point
out
rightly
so,
I
think
it's
it's
a
lower
bar
of
entry
for
folks
that
are
not
specifically
Dev.
You
know
and
I
think
that's
great
great.
All.
C
A
To
maintain
it,
it
doesn't
do
it
itself.
The
Google
Doc
is
sort
of
easy.
You
know
you
just
you'll
notice
in
ours.
It's
our
agenda
notes,
we
just
let
it
run
for
theoretically
a
year
and
then
start
a
new
one,
although
sometimes
we
haven't
even
done
that
and
we're
expected
under
cncf
rules
to
give
up
to
steering
a
recounting
of
you
know
how.
Why
should
your
group
exist?
How
many
people
are
active
here
that
kind
of
stuff?
You
have
one
place
to
go
where
you
can
get
that
summary
people
can
go.
A
Look
if
they're
curious
about
your
group.
It
gives
them
one
place
where
they
can
see
the
whole
track
record.
Do
I
even
want
to
join
this
well,
you
can
look
at
what's
going
on
for
the
past
year
in
one
place,
and
you
know
answer
that
kind
of
question
or
I.
Often
a
lot
of
kubernetes
groups
operate
with
this,
where,
with
a
running
agenda,
notes,
I
think
I
first
saw
it
in
Sig
storage,
but
the
other
thing
that
I
found
it
solves
really
nicely
is
gee.
A
A
Find
my
link
to
what
I
was
looking
for,
so
in
a
way
this
ends
up
being
like
my
own
notepad,
where
yeah
and
what
you
do
have
to
do
on
these
by
the
way
is,
if
you're
running
Zoom
meetings,
I
always
ask
people
to
help
edit
the
agenda
notes.
A
E
A
Hours
later
and
polish
up
your
agenda
notes
and
fill
it
in
and
if
you
wanted
to
cut
and
paste
that
into
your
YouTube
description,
you
know
that
thing
where
you
hit
more,
that
same
technique
of
if
it
actually
was
a
zoom
recording
going
at
that
text
to
speech,
can
kind
of
give
you
those
things
and
in
some
cases,
if
you
want
to
deal
the
extra
mile
and
get
time
stamp
readings
of
where
the.
C
A
C
A
Think
it's
at
least
a
third
of
a
time
job
for
a
really
decent
project
that
has
contributions
from
multiple
orgs.
It
reaches
that
point
where
the
Community
Management
it's
up
to
you,
how
you're
going
to
get
that
funded
by
somebody,
because
you
know
people
have
to
eat
so
expecting
a
volunteer
to
show
up
and
just
do
it
for
cold
free
is
what.
C
C
C
So
yeah,
no,
these
are
all
really
really
good
points.
If
you
don't
mind,
I'm
gonna
actually
look
at
your
agenda
and
mirror
it
if
it's
possible
for
use
in
our
community,
not
the
not
everything
but
the.
C
A
C
Hey
well,
thank
you
for
welcoming
us
with
open
arms
and
looking
forward
to
continuing
working
together
with
you
and.
A
Like
I
say
the
agendas,
the
agenda
isn't
just
for
notes.
If
you
want
to
speak
at
a
future
session,
just
go
put
your
name
in
and
you're
free
to
have
a
slot
great.