►
From YouTube: Kubernetes WG IoT Edge 20200701
Description
July 1 2020 meeting of the Kubernetes IoT Edge Working Group
A
There's
some
a
lot
of
adopters,
but
they
don't
want
to
release
their
names.
They're
trying
to
sweden
say
they
don't
because
the
majority
contributor
about
40
percent
contributed
by
huawei.
So
but
they
said.
Okay,
we
don't
want
to.
I
mean
get
a
connection
for
now
under
this
environment,
so
it
does
give
us
a
little
bit
trouble
for,
because
the
toc
member
will
mean
survey
all
the
adopters
to
make
sure
it's.
What's
the
feedback
or
something.
B
A
So
so
far
so
good,
so
hopefully
that
will
be
open
to
public
comments
in
you
know
week
also
so
currently,
under
a
secret
review.
B
A
But
some
something
right:
you
can
see
in
the
k3s
the
discussion
even
joe
betta
and
who's,
the
other
guy
from
we
works.
A
They
are.
I
mean
strongly
against
that
k3
that
that's
why
the
k3s
doesn't
make
the
sandbox.
So
it's
your
count
depends
on
who
gives
the
feedback
right
if
the
guy,
who
have
the
influence
on
the
public
that
we
decided
and
yeah
and
they
change
the
standard
for
sandbox.
I
guess
more
and
more
people
can
make
a
sandbox
much
easier.
A
Thank
you
thank
you
and
yeah.
There
will
be
more
and
more
sandbox
product
in
the
cncf
and
I
heard
that
they
are
going
to
lose
the
maintenance
track
for
all
the
sandbox
projects,
because
they
are
going
to,
I
mean,
lose
the
standard
for
sandbox.
They
want
more
projects,
but
they
are.
B
Yeah,
I
think
that
the
I've
heard
definitely
some
debate
that
they
made
the
sandbox
to
lose,
particularly
that
was
the
opinion
in
the
storage
sig.
But
then
it
came
down
to
an
argument
that
once
we
let
the
first
few
in,
we
can't
suddenly
slam
the
gate
shut
on
the
others
who
look
just
as
good
as
the
ones
we
already
let
in
so,
but.
A
They
are
going
tighten
up
on
the
incubation
and
the
graduation
yeah,
I'm
hosting
another
c9
genie
project
we
applied
last
october.
It
never
got
processed.
We
we
stopped
to
the
project
is
almost
idle
for
about
of
three
or
four
months,
but
suddenly
in
making
the
sandboxes.
I
haven't.
A
One,
what
was
it
again
right?
It's
called
cni
genie,
it's
a
it's,
a
more
like
a
cni
improvement
project.
B
A
A
A
So
you
can
see
between
october
to
may
there's
no
activities.
A
It's
just
say
why.
A
To
cook
up
to
multiple
plugins
orange,
as
at
the
very
beginning,
a
very
beginning,
some
production
team
asked
us,
so
we
did
that
and
also
orange
the
technical
company
interest
in
that,
however,
I
haven't
followed
them
up,
so
I
don't
know
what's
going
on
right
now,
so
I
haven't
touched
it
for
almost
a
year.
B
A
B
Multiple
deployment,
especially
if
it
sits
in
the
data
plane
where
it's
subject
to
causing
latency
or
if
it's
control
plane
only
fine,
but
it
seems
like
if
it's
in
the
data
plane
it's
it
can't
improve
performance
right.
A
Yep,
but
you
can
see
the
vmware
in
the
vmware
archive
the
the
awk
there's
a
lot
of
archive
the
open
source
project
whenever
yeah
now
that
activity
put
over
there.
A
C
I
have
a
quick
question
about
cube
edge
itself.
I've
just
been
scrolling
through
the
documents
you
send
around
and
on
the
high
level
architecture.
You
show
connectivity
to
opc
ua,
but
I
can't
really
find
that
in
in
the
documentation.
Is
that
already
implemented?
Or
is
this
like
like
on
the
road
map.
A
To
this
should
be
on
the
roadmap,
I
think
it's
the
documentation,
which
are
you
in
the
really
dog
that
documentation
or.
C
You
probably
can
because
it's
just
jason
that
comes
out
of
it,
but
that
actually
sounds
sounds
quite
interesting.
Another
question
I
that
I
I
thought
about
was:
can
I
actually
change
device
software
with
cube
edge?
So
can
I
use
containerized
hardware
controllers
with
cube
edge
or
do
I
have
to
use
the
mqtt
broker.
C
So,
for
example,
my
my
opc
ua
server
or
my
my
modbus
server
or
my
my
bluetooth
device.
C
Would
that
be
possible
or
or
do
I
yeah,
because
right
now,
I'm
actually
like,
like
in
the
process
of
actually
thinking
about
how
to
handle
device
communication
with
containers
like
setting
up
drivers
deleting
driver
when
the
driver
is
when
the
container
is
finished
and
so
on,
and
how
to
to
detect
different
hardware
parts
dynamically.
C
C
A
No
yeah,
this
seems
it's
more
like
a
plug-in
plug-and-play
right
yeah.
We,
I
don't
think
we
developed
that
we
had.
I
mean
the
current
in
this
year.
We
are
looking
for
the
ota
update
and
the
firmware
firmware
update,
but
that
one
is
interesting.
I
I
never
thought
about
that.
I
mean.
B
Yeah
there
actually
is
a
downside.
You
know
like
windows,
has
that
plug
and
play,
but
potentially
it
with
an
edge
iot
device.
That
is
unsecure.
You
know
like
there's
nobody,
no
security
guards,
locking
it
down.
It
seems
like
there's
a
real
vector
for
an
attack
there
if
you
were
able
to
plug
in
some
random
usb
device
and
just
get
it
to
do
things.
C
The
idea
I
had
was
that
you
have
some
kind
of
demon
set
running
that
has
like
an
allow
list
in
it
in
its
let's
say,
search
pattern.
So
so
you
kind
of
check
boxes
like
I
want
to
support
this,
and
this
and
this
kind
of
hardware.
C
I
have
no
idea
how
you
can
actually
check
for
that,
like,
of
course,
usb
devices
and
stuff
like
that
is
quite
easy.
But
as
soon
as
you
go
like,
I
don't
know,
gpios
or
spi
interfaces
and
stuff
like
that,
it
gets
really
nasty,
detecting
what
is
actually
plugged
in
there.
C
So
I'm
not
quite
sure
how
to
handle
that
on
the
on
the
detection
level,
but
the
idea
sounds
or
what
was
quite
nice,
that
you
kind
of
just
grab
whatever
you
have
in
your
in
your
let's
say
in
your
closet
and
then
you
just
plug
it
into
your
computer
and
it
gets
the
software
automatically
and
you
don't
have
to
worry
about
installing
it
yeah.
So
that
was
like
like
and
then
I
thought
about
like
how
can
I
handle
that
with
with
initialization
parts,
for
example,
handling
the
installation
on
the
host?
C
How
can
I
delete
the
driver
afterwards,
once
I
kind
of
get
rid
of
it
or
like
when
I
unplug
the
usb
device
and
stuff
like
that,
but
I'm
not
quite
sure
how
to
handle
that
right
now,
and
I
just
thought
about
throwing
it
in
here.
If,
if
you
guys
have
any
ideas
like
yeah.
B
I
got
a
question
for
you
before
I
throw
out
ideas,
but
are
you
using
the
opc
ua
right
now
and
if
so,
it's
been
years
since
I
had
anything
to
do
with
opc,
but
at
one
time
that
used
to
be
windows
only
just
curious,
if
that
is
fully
supportive
of
linux.
These
days.
C
I'm
not
quite
sure
actually,
so
I
have
the
feeling
that
that
the
big
plc
manufacturers
like
siemens,
cements
and
so
on,
are
pretty
much
pushing
the
whole
protocol
right
now.
I
think
it
is.
It
can
be
run
on
linux,
I'm,
I
have
to
admit,
I'm
not
quite
a
big
fan
of
opc
ua,
because
it's
like
the
idea
of
of
microservices
brought
down
to
the
production
or
to
the
manufacturing
level,
but
in
a
very
complicated
matter.
You
know
what
I
mean
like
there's
a
whole
protocol.
D
C
That
was
the
opposite.
You
see
like
why
can't
I
use
something
similar
like
grpc
or
or
the
simple
json
and
some
kind
of
iding
or
like
like
open
id
with
with
json
and
rest
for
for
authentication.
C
But
of
course
I
know
that
they
want
to
do
like
the
whole
semantic
server
thingy,
like
I
just
plug
in
my
my
opc
ua
server
into
my
network
and
then
it
kind
of
introduces
itself
into
the
network
and
say
hey
guys.
I
have
this
kind
of
function
and
I
have
this
kind
of
data
that
you
can
query
and
so
on.
But
personally
I
believe
it's
like
a
very
complicated
protocol
for
just
doing
that
and
yeah.
So
so
I'm
not
the
biggest
fan,
I'm
not
quite
sure
how
it
will
evolve.
Yeah.
D
B
D
B
And
they
had
so
many
meetings
before
that
effectively
see.
I
did
a
startup
called
wonderware
that
we
literally
invented
that
dde
protocol
and
at
one
time
I
think
I
probably
had
written
this
was
the
very
first
few
years,
but
I
personally
had
written
over
50
percent
of
those
opc
drivers,
and
you
know
it.
It
was
a
scenario
where
it
was
complex
enough
that
companies
in
the
field
probably
devoted
at
least
one
full-time
employee
to
just
keeping
up
with
opc.
C
Yeah,
I
I
think,
like
the
problem
right
now
is
that
that
opc
ua
is
like
this
digitalization
for
manufacturing
protocol,
or
it's
called
like
that.
That
sounds
like
that
and
what
it
basically
does.
It's
it's
enforcing
people
to
not
think
about
networking,
which
I
find
like
really
hard.
If
you
really
want
to
digitalize
your
your
manufacturing
process,
because
you
can't
do
like
digitalization,
because
without
knowing
all
the
network
stuff
underneath
you
know
what
I
mean
right
so
for
me,
it's
kind
of
like
this.
B
Yeah,
one
of
the
other
aspects
in
that
field
is
that
the
end
users,
once
they
find
something
that
works,
are
very
resistant
to
swapping
it
out.
So,
if
they're
convinced
that
this
opc
driver
is
at
least
stable
and
bug-free,
they
don't
want
to
take
the
perceived
risk
of
switching
to
something
different,
because
you've
got
people
who
you
know
a
factory
outage
could
cost
them
a
million
dollars,
and
if
it
was
something
like
a
refinery,
the
whole
plant
could
explode
or
something
so
they're
kind
of
risk-averse.
C
Yeah
yeah,
of
course,
I
think,
that's
the
whole
discussion
like
the
whole.
C
The
whole
discussion
about
agility
in
manufacturing,
for
example,
because
I've
talked
to
people
from
from
siemens,
for
example,
and
they
still
serve
customers
from
20
years
ago
with
plc
repair
parts
that
were
built
like
in
the
80s,
basically,
which
is
crazy.
If
you
think
about
that.
Well,.
B
B
B
C
The
funny
part
is
that
the
whole
industry,
for
example,
like
the
automotive
industry,
now
wants
to
become
more
agile
or
more
more
flexible
in
their
production
system.
And,
of
course,
the
whole
hardware
that
I
have
been
building
for
the
last
20
to
30
years
is
not
prepared
for
that,
because
you
can't
just
plug
and
play
your.
C
What
is
it
called
an
english
be
or
for
your
engine,
for
example,
the
drilling
machines
or
whatever?
Of
course,
the
question
arises
like
how
flexible
do
you
want
to
go
like
when
you,
when
you
think
about
what
you
just
said,
chemical
plants,
where
the
adaptability
doesn't
need
to
be
that
big?
You
can
argue,
okay,
I
can
just
weld
my
software
and
my
hard
way
in
there,
but
of
course,
as
soon
as
you
want
to
change
something
later
on,
that
gets
like
really
hard
to
do
and.
B
C
Kind
of
like,
like
a
lot
of
people,
are
trying
to
to
be
more
flexible
in
that
way
by
using
old
hardware,
and
that's
definitely
not
going
to
work
yeah.
I.
B
B
Are
your
key
to
having
long
life
non-disruptive
and
I'd
use
as
a
model,
something
like
if
you
look
at
the
internet.
You've
got
these
ancient
protocols
like
border
gateway
protocol,
for
example,
which
handles
the
routing
and
switching
between
internet
carriers,
and
that
stuff
is
literally
decades
old.
Yet
I
don't
know
of
anybody
even
talking
about
changing
it,
and
the
reason
it
has
lasted
is
simply
that
it's
was
extensible
and
they
they
did
a
decent
architecture
so
that
that
has
been
able
to
evolve
over
time,
even
when
the
time
period
is
decades.
B
You
know
the
lldp
the
link,
the
link
layer,
discovery
protocol
is
another
one
like
that.
That
just
was
really
well
done
and
I
actually
think
the
kubernetes
control
plane,
particularly
with
the
crd
feature
that
you
know
really
got
expanded
out
a
couple
years
ago,
gives
kubernetes
the
opportunity
to
put
an
abstraction
on
things
so
that
you
know
I'm
happy
to
live
with
and
not
rip
and
replace
the
old
stuff
for
many
customers
or
users.
B
C
That's
true
yeah.
I
really
I
really
like
the
the
design
of
of
the
cuban
it
app
api
because
it
just
you
kind
of
feel
that
that
the
people
put
so
much
work
in
at
it
to
make
it
like
plug
in
or
to
use
plug-in
architectures
on
every
level.
C
Like
for
example,
network
plug-ins
and
then
container
plug-ins
or
cni
plug-ins,
so
basically
everything
is
changeable.
In
case
you
want
to.
C
B
B
So
yeah
just
to
let
you
know,
I
think
it's
because
of
the
last
time
I
used
the
master
account
for
the
working
group.
I
you're
showing
my
headshot
on
the
little
icon
for
participants,
so
I
think
of
your
right
mouse
click.
You
can
replace
it
with
your
own
or
set
it
to
blank.
It
just
looks
funny
because
I'm
looking
at
two
of
me
on
there
now,
don't
you
have
a
twin.
B
Well,
I
don't
think
we'll
ever
get
it
built
into
the
official
api,
but
I
think
that
there's
a
potential
for
putting
a
facade
on
there,
they
could
manage
common
devices
and
have
them
be
viable,
a
viable
open
source
project.
If
you
will,
I
don't
think
that
the
kubernetes
steering
committee
at
this
time
would
accept
it
as
part
of
kubernetes
itself.
D
B
B
B
B
Industrial
temperature
controllers
kubernetes
itself
is
never
going
to
do
that
because
it's
too
much
of
a
niche
use
case,
but
nothing
would
stop
you
if
you
could
get
critical
mass
like
if
you
could
get
all
the
manufacturers
of
temperature
loop,
controllers
or
plcs,
which
you
brought
up
to
get
behind
a
standard
and
you
know
unify
on
one
crd
to
rule
them
all.
I
think
you
could
pull
that
off
now.
Some
of
those
device
categories
like
plcs
are
so
different.
B
C
B
C
And
this
stuff
is
really
complicated
like
more
complicated
than
it
should
be.
Yeah
like
kubernetes
is
a
blast
compared
to
setting
up
in
plc.
B
I
don't
know
it
just
it
just
depends.
If
you
take
somebody
like
in
the
us
market,
somebody
like
an
electrician
rather
than
a
software
engineer,
they
kind
of
like
the
way
they
the
things
are
laid
out
for
them,
because
that's
that
was
the
engineering
effort
to
make
it
digestible
from
people
who,
in
the
day,
maybe
even
connected
up
old
physical
relays,
and
you
know
those
people,
I
think,
who
have
never
been
to
a
computer
science
program,
would
have
great
difficulty
absorbing
kubernetes.
C
I've
just
found
out
like,
like
the
the
whole,
what
was
it
home
assistant?
I
think
this
open
source
project
this
can
actually
be
run
on
kubernetes.
Yes,.
B
I
chose
z-wave,
but
regardless
of
what
you
chose
to
interface
to
your,
I
o,
like
your
light
switches
and
sensors,
you
could
use
zigbee
or
z,
wave
or
there's
probably,
at
least
in
the
u.s
market,
a
couple
other
plausible
things,
but
I
chose
to
have
those
route
to
a
raspberry
pi,
which
then
publishes
it
mqtt
using
a
mosquito
broker,
and
then
the
home
assistant
talks
to
the
mosquito
broker
to
do
its
io,
and
I
like
that
model
just
because
home
assistant
takes
patches
like
at
least
once
a
week
and
by
the
time
you
bundle
in
the
z-wave
thing
and
everything
else,
you're,
probably
getting
hit
with
a
patch
once
a
day,
and
it
was
just
too
disruptive
and
being
able
to
separate
it
out.
B
B
C
A
similar
situation,
I
think,
three
weeks
ago,
with
my
dad
who
tried
to
install
net
next
cloud
on
on
the
server,
and
I
I
was
just
like-
really
frustrated
with
the
whole
deployment
process,
because
you
have
to
set
up
engine
x
and
you
have
to
set
up
some
kind
of
certificate
in
the
engine
x
and
then
you
have
to
port
forward
the
whole
thing,
and
I
was
just
like
wow.
If
you
we
just
would
like
pull
up
a
cluster
right
now
and
then
put
let's
encrypt
on
there.
C
B
Well,
even
kubernetes
has
that
stuff
going
on
if
you're
talking
about
running
an
on-prem
like
in
your
house,
you're
gonna
need
ingress
and
that
might
quite
likely
end
up
being
nginx.
So
there's
a
lot
of
moving
parts.
Some
distros
will
configure
all
that
stuff
for
you
pretty
easily.
But
if
you
do
it
the
hard
way
from
the
pure
open
source,
there's
a
fair
amount
of
work.
Getting
that
up.
C
Yeah,
but
I
have
a
like
running
at
my
home
with
like
metal
and
the
metal
b
and
x
ingress,
and
the
nice
part
about
that
is
that
everything
is
like
very
easy
deployed.
It's
not
like,
like,
for
example,
nginx
where
you
have
like
just
the
bare
bones.
Once
you
install
it,
yeah.
C
B
But
yeah,
if
you
look
at
home
assistant,
even
when
you
run
it
on
a
pie,
they
don't
really
document
it.
Well,
you
have
to
go
over
to
the
developer
docs
to
see
how
this
works,
and
then
I
sshed
into
my
pie
and
started
poking
around
and
they're
running,
all
the
components
on
in
docker
containers,
even
on
a
pie
and
there's
no
orchestrator
there.
B
You
can
install
it
on
a
pie
and
have
cube
cuddle
talking
to
it
in
an
hour
and
it's
pretty
straight
forward,
but
I
found
it
wasn't
that
stable
and
the
other
trouble
is
that
in
a
pi's
memory
footprint
by
the
time
you
lay
down
kubernetes,
you
don't
have
that
much
left
over
for
apps.
B
B
Actually
worth
it
in
terms
of
complexity
and
cpu
and
memory
consumption,
you
know
if
you
are
only
single
node,
maybe
a
better
approach.
Even
if
you
wanted
to
use
a
kubernetes
control.
Plane
is
to
go
down
to
a
container
runtime
and
have
your
kubernetes
control,
plane,
external
and
use
crds
to
control
the
containerized
workloads,
of
which
there
may
only
be
one
or
two
yeah.
C
B
B
If
you
had
multiple
clusters
more
than
likely,
I
o
devices
are
locked
to
one
cluster
node.
You
know
there's
a
lot
of
this
device,
iot
stuff
that
isn't
really
on
a
bus
that
could
talk
to
three
different
kubernetes
nodes
so
that
you
were
free
to
schedule
the
gateway
talking
to
that
to
any
one
of
three
nodes.
B
Non-Deterministically,
and
if
your
I
o,
like
you,
have
one
usb
z,
wave
interface,
the
gateway
is
gonna
have
to
run
on
the
node
that
that
thing
is
plugged
into
and
if
you're
you're
going
to
have
to
get
the
kubernetes
scheduler
to
always
deploy
it
on
that
one
node.
It's
sort
of
like
you're
fighting
the
scheduler
with
a
bunch
of
labeling
and
things
to
get
that
to
happen.
Rather
than
have
anything
to
be
your
friend.
B
B
B
A
B
So
I
think
one
of
the
areas
that
isn't
covered
is
actually
some
kind
of
you
know.
Kubernetes
manages
like
a
control,
plane
and
deploying
containerized
images.
It
is
less
than
standardized
for
doing
things
like
firmware
images
out
to
devices
like
you
wanted
to
replace
the
firmware
on
whatever
device.
B
Now
well,
to
some
extent,
if
you're
in
a
public
cloud
provider,
there's
something
called
cluster
api
where
you
can
deploy
or
replace
hosts,
but
if
you're
out
on-prem
on
bare
metal,
let's
just
say,
you're
running
on
a
server-
I
don't
know
a
dell
or
an
hp
server.
Those
things
have
a
lot
of
firmware
and
if
you,
and
in
the
course
of
a
year
it's
more
than
likely
the
bios
and
maybe
firmware
on
host
bus
adapters
need
patches
some
of
them
security
related.
C
Yeah,
because
this
is
quite
similar
to
the
question
I
had
earlier
like
how
do
I
handle
device
drivers
for,
for
example,
usb
cameras
like
that?
Why,
like
how
do
you
handle
firmware
on
device
level.
B
A
Yeah
but
the
for
the
ota.
We
are
looking
for
some
other
solutions,
as
I
mentioned.
I
remember
probably
two
weeks
ago
or
a
month
ago,
in
this
meeting,
kyotan
also
mentioned.
Probably
the
hobbit.
I
mean
young
yeah
yeah,
that's
a
possible
solution.
I
don't
think
kubernetes
itself
will
do
that,
but
it
can
integrate
with
other
open
source
project
to
solve
the
problem.
D
That
I
am
aware
of
and
and
what
I
think,
kilton
and
and
and
the
guys
from
bosch
are
doing,
is
they're
trying
to
model
the
gateways
and
and
the
physical
resources
not
with
with
the
crds
but
with
borto
eclipse
portal,
which
is
a
kind
of
a
modeling
language
used
for
the
for
the
devices.
D
D
D
Unfortunately,
okay
yeah,
so
it's
a
modeling
modeling
tool
for
different
kind
of
payloads,
like
how
do
you
define
the
payload
payload
of
your
sensors?
So
what's
the
temperature?
What's
the
unit?
What's
the
like?
How
do
you
parse
the
data
coming
from
the
sensors?
So
they
got
to
the
idea
to
to
use
the
similar
tool
to
to
basically
model
the
the
edge
nodes
so
to
say
so
that
that
and
and
the
whole
bit
uses
or
we
can
use
water.
B
B
What's
the
name
fledged
glitch,
okay,
here
I'll
put
it
in
chat.
B
C
Preston
holmes
is
also
on
the
eclipse
water
page.
I
think
he's
quoted
down
there
so.
B
D
B
And
by
the
way,
speaking
of
meetings
we're,
I
guess
a
little
over
time,
but
we
started
late,
but
I
think
we
will
be
able
to
do
a
maintainer
track
set
session
for
the
boston
online
conference.
So
if
anybody's
got
ideas
for
what
we
might
want
to
cover,
there.
B
No,
they
well
it
it
didn't
even
open
for
submits
yet
so
yeah.
So
usually
they
do
the
deadline.
After
and
some
of
these
they've
stretched,
I
think
some
of
these
conferences
they
seem
to
have
you
know,
based
on
the
extensions
they
seem
to
be
getting.
Maybe
issues
with
people
not
submitting
talks.
I
think
open
source
summit.
Japan
has
been
extended
like
three
times
now,.