►
From YouTube: Kubernetes WG IoT Edge 20220727
Description
July 27 meeting of the CNCF IoT Edge Working Group. edge native white paper draft discussion, Flotta project: covers device onboard and config, OS lifecycle management with update rollback, app distribution. Tolerated low resource and intermittent network connections. 64 bit for now. Use cases: manufacturing, trains, drones. https://project-flotta.io/documentation/latest/intro/overview.html
A
Hi
welcome
to
the
July
27
meeting
of
the
cncf
iot
edge
working
group
on
today's
agenda,
we're
going
to
start
with
a
brief,
10
minute-ish
discussion
of
a
white
paper
on
edge
that
this
group
is
working
on,
and
then
that
will
be
followed
up
with
a
presentation
on
Project
floto,
which
is
has
Edge
device
management
combined
with
kubernetes.
So
with
that
said,
let's
jump
right
into
the
white
paper
discussion.
If
anybody
who
has
who
is
organizing
that
wants
me
to
make
them
a
co-host
because
he
wanted
to
share
anything,
go
for
it.
A
B
You
Stephen,
this
is
Brandon
I've,
just
gone
and
added
a
link
to
the
white
paper
outline
draft
in
the
chat,
quick
reminder:
we've
been
working
on
this
now
for
several
weeks
and
this
group's
had
a
a
couple
discussions
already.
B
Unfortunately,
both
myself
and
my
colleague
Amar
were
unable
to
make
the
meeting
two
weeks
ago
where
the
paper
was
discussed
at
length,
but
we've
looked
through
the
the
comments
and
the
minutes
and
just
wanted
to
respond
to
a
couple
of
things
and
then
also
just
get
this
group's
quick
feedback.
I
realized
we
do
only
have
10
minutes
and
there's
another
topic,
so
I
don't
want
to
take
up
that
time,
but
we
want
to
keep
making
progress
on
this
outline,
so
we
can
look
to
finalize
it
and
then
start
drafting.
B
So,
if
I
to
do
a
summary
of
some
of
the
feedback
that
I
saw,
we
want
to
make
sure
that
we're
not
duplicative
of
existing
efforts
and
we've
looked
through
some
of
the
other
materials.
There
was
a
white
paper
from
our
friends
at
Huawei
and
in
China
that
was
interesting.
It
was
pretty
Broad
in
terms
of
personas
I.
Think
the
cncf
angle
would
be
a
little
bit
more
developer,
specific
and
also
a
little
more
technology
agnostic.
B
C
No
I
think
you
covered
it
and
I
I
also
I
reviewed
the
the
Huawei
paper,
and
it
does
have
some
points
for
developers,
but
it
sort
of
sprinkled
and
it's
it's
not
focused
on
that
so
I
think
I
I,
don't
think
we'll
be
duplicating
that
that
particular
aspect
and
also
it's
I-
think
it's
more
opinionated
in
terms
of
prescribing
Technologies.
It
comes
from
an
Edge
Gallery
point
of
view,
which
has
things
like
sharded
databases
and
all
in
the
platform
itself.
C
Well,
I
think
a
kubernetes
perspective
would
probably
be
more
more
sort
of
neutral
is
what
I'm
thinking
so
I
also
I
also
feel
comfortable
that
it
it's
not
duplicative.
C
B
D
Just
to
comment
along
this
line,
so
one
of
the
goals
of
this
white
paper
also
is
to
get
a
broader
audience
to
feedback
from,
hopefully
kubecon
North
America,
and
so
we
submitted
a
proposal
to
kind
of
discuss
our
status
there
at
kubecon,
North,
America
and
I.
Think
I
can
find
the
I'll
find
the
abstract
and
share
it
here
of
kind
of
what,
if
it's
accepted,
people
would
read
the
session
looking
like,
but
that's
also
a
goal
to
work
towards
of
being
at
a
reasonable
state
by
then
I
think
it's
mid-october.
A
Not
to
clutter
the
waters
but
I
just
got
an
email
from
the
eclipse
foundation
in
the
last
couple
days,
with
a
link
to
a
white
paper.
They
published
I
dropped
that
in
the
chat.
I,
don't
know
if
that
link
is
specific
to
me.
So
let
me
know
if
somebody
tries
to
open
it,
whether
that
actually
works
for
you
or
not.
But
it
looks
like
a
very
general
paper
from
my
perspective
and
you
know:
there's
still
room
for
others.
A
A
But
so
does
the
cncm,
so
I
think
in
some
cases
these
white
papers
coming
out
of
the
orgs
kind
of
Spotlight,
their
own
projects
and
I,
don't
know,
that's
that's
a
goal
for
the
one
we're
working
on
here,
we're
kind
of
so
Broad
in
scope
that
I'm
willing
to
open
it
up
for
anything,
including
covering
projects
coming
out
of
Open
Source
areas
that
aren't
in
the
cncf,
because
I
think
the
whole
point
is
to
serve
potential
users
not
to
serve
the
foundation
itself.
So
that
that's
my
attitude.
B
Else
volunteered
to
be
an
author
of
the
paper
right
now
we're
starting
with
two
Amar
and
Robbie
from
Verizon.
Has
anyone
expressed
an
interest
on
taking
a
section
or
two.
A
I
may
entertain
it
I
I,
don't
want
to
take
on
the
scope
of
being
sort
of
the
Czar
of
the
whole
document,
but
for
things
that
maybe
I
have
specialized
interest
and
knowledge
in.
If,
if,
if
doing
a
piece
of
it
constitutes
being
an
author
I
guess,
let's
clarify
that
too,
because
you
know
I
wouldn't
expect
that
if
you
write
two
sentences
that
you
necessarily
get
listed
as
an
author,
but
we
could
do
it
that
way.
A
B
Thanks
Steven
yeah
I
think
we
can
work
that
out
when
we
have
a
better
idea
of
how
many
and
how
long
so,
I
see
we're
almost
out
of
time
here.
Any
other
thoughts
from
the
group
about
how
we
proceed.
Should
we
I
put
this
out
on
the
mailing
list,
as
well
as
the
slack
Channel
got
some
feedback,
but
not
a
ton.
Is
there
anything
else
we
need
to
do
to
start
finalizing
this
outline
and
then
start
drafting
well.
A
I
think
this
came
up
for
Chad
in
the
last
meeting
that
you
missed
that
we
wanted
to
try
to
do
this
asynchronously,
because
you
know
pretty
much.
Everybody
is
in
a
situation
where
you
can't
make
every
meeting
and
by
allowing
contributions
in
the
form
of
direct
headache,
contributions
or
comments,
people
can
kind
of
work
offline
and
the
idea
was
that
we
would
allocate
10
minutes
at
every
one
of
these
recurring
meetings
to
go
back
and
visit
it
and
go
over
the
comments.
A
Now
this
we
kind
of
did
broadly
but
I
think
to
get
to
a
final
document.
We
kind
of
probably
need
to
open
that
document,
go
through
comments
and
and
address
them
and
close
them
and
I
think
that
I've
seen
that
work
in
the
past
on
these
white
papers.
So
that's
just
a
proposal
but
yeah
I'm
open
to
others,
but
I
think
that
would
work.
D
Maybe
we
can
aim
to
have
that
line
fully
finalized
by
the
next
meeting
and
I.
Think
a
good
thing
we
could
do
is
go
through
and
assign
people
sections,
so
that
would
be
kind
of
a
good
time
to
decide
whether
any
of
the
sections
would
be
good
for
you
before
the
meeting,
and
then
we
just
go
and
put
someone's
name
by
it
and
obviously
that
can
change
people
can
swap,
but
then
we'll
have
some
working
material
for
the
the
meeting.
After
that.
A
Okay
Kate:
do
you
want
to
be
the
person
in
charge
of
putting
people's
names
on
my
opinion
is
to
get
it
done
fast,
it's
better
for
it
not
to
be
100
democratized
to
where
we
get
consensus
on
micro
decisions,
and
that
sounds
like
a
great
idea
and
if
you
want
to
do
it,
fine,
otherwise,
how
about?
If
we
appoint
somebody-
and
it's
not
to
say
that
you
know
the
rule
is
final.
D
A
Any
objections
to
that.
Okay,
I'll
take
silence
as
no
objection.
So
let's
go
with
that
plan.
So
go
in
that
white
paper
go
assign
yourself
sections.
If
there's
two
on
a
section,
I
have
I.
Think
that's!
Okay,
too.
If
two
people
want
it,
but
we'll
go
with
that
and
then,
if
there's
gaps
later
I
don't
know
Kate
are
you
gonna
take
on
assigning
names
or
should
that
be
somebody
else.
D
Try
to
do
that
as
long
as
we
know
the
name,
it
seems
like
right
now
we
have
a
couple
people
who
are
interested
so
all
put
in
the
slack
kind
of
this.
What
we
just
discussed,
saying
hey
over
the
next
week,
go
through
white
paper,
see
what
you're
interested
in
comment
that
you're
interested
otherwise
comment
in
this
thread
that
you're
interested
in
participating
in
some
way
and
I'll
give
you
a
section
something
along
those
lines.
A
Okay
sounds
good,
so
in
the
interest
of
time
we
already
shaved
about
three
minutes
off
is
allotment,
but
I.
Think
Eloy
is
here
to
speak
on.
Florida
I
believe
I
made
you
a
co-host.
So
let
me
know
if
that
didn't
appear
to
have
happened
and
how
about
either
Dion
you
can
introduce
him
or
he
can
introduce
himself
and
I'll
turn
over
the
mic
to
you.
E
Yeah,
let
me
set
my
screen.
Let
me
see
if
I
can,
sir.
A
We're
getting
something,
but
it
looks
like
gray
diagonal
hash
lines,
not
something
visible.
A
It
looks
like
it's
zoomed
in
so
that
it's
cut
off
on
the
left
yeah.
Let
me
see.
A
E
No
I
don't
have
a
half
white
word.
Buns
I
have
like
personal
screen,
but.
D
E
E
Yes,
maybe
I
don't
know
if
Modi,
theater
or
Jordi
can
try
to
serve,
but
I
cannot
search
them.
The
demo,
that's
okay,.
E
E
E
C
With
your
laptop
screen,
instead
of
the
the
external
monitor.
C
D
C
Okay,
does
anybody
else,
have
the
slides
at
hand
or
I
need
to
pull
them?
The
other
moti
do.
E
D
C
A
It's
say:
I.
B
E
E
So,
let's
go
to
the
to
floata,
so
I'm
Malay
from
floata
I'm,
with
my
colleaguese,
a
few
other
mods
that
are
not
here.
We
are
working
on
on
credit
flat.
President
bonds
inside
the
team
is
special
plates
inside
a
Rahat.
We
aim
to
work
on
the
in
that
solutions
for
small
footprint
devices
like
Jetson
bars,
Raspberry,
Pi's
interlock.
E
It
has
some
use
cases.
So
let
me
so
what
we
are
focusing
in
Florida.
Let
me
see
because
I
think
that
okay,
so
what
we
are
focusing
go
to
the
slide
before
so
what
is
the
main
focus
for
us
is
three
things
that
is
like
the
device
life
cycle
in
this
case
is
the
the
Bison
boarding
the
base
configuration
and
the
US
life
cycle.
Like
updated
the
robot
from
boarding,
we
have
our
process,
but
we
also
integrate
with
fvdo
the
final
device
on
boarding
process.
E
That
is
something
that
Fedora
iot
is
working
on
and
we
already
integrated
with
that
about
the
bias
configuration
for
Rich
devices.
We
have
like
the
configuration
to
getting
more
loads
and
stuff,
and
we
also
will
integrate
some
kind
of
ancible
workloads
in
there.
So
as
soon
as
a
new
device
joining
the
the
flim
Anonymous
Fleet
Management
solution,
you
can
run
any
any
support
playbook
in
in
there.
That
is
not
in
there
yet,
but
it's
going
to
be
in
the
next
release.
E
E
We
also
care
about
the
application
life
cycle,
like
how
you
define
the
application.
How
do
you
run
the
application?
So
we
use
the
same
aspect
as
the
bot
in
kubernetes
how
you
run
that
is
like
the
application
runtime.
That
is
what
we
use
is
the
podman
play
Cube,
so
we
rely
on
play
Cube
more
than
running
the
applications.
E
We
also
have
like
the
data
collection,
like
you,
can
think,
with
any
F3
like
a
dueling
like
S3
double
BM
menu,
so
we
have
a
way
to
both
data
and
retrieve
data
from
your
stream
inside
the
workloads,
and
we
also
taking
care
about
the
severability
and
the
log
collection
for
for
the
for
the
workloads
about
what
is
the
cases
that
we
aim
for
is
our
the
environment,
that
we
say
that
we
are
good
at
or
at
least
we
want
to
to
be
good.
E
E
We
care
about
the
multi-art,
so
we
build
for
x64
Ram
in
rvm.
We
only
provide
support
for
64.,
even
that
we
can
provide
support
for
rem32,
but
at
the
moment
we
only
focus
on
on
64
bits
as
I
say
the
flacky
networks
that
become
we
can
learn
the
connectivity
in
any
moment
also
because
the
devices
that
we
want
to
use
is
more
close
to
the
iot.
That
the
5D
yet
is
more
like
a
minimum
footprint,
so
we
just
care
about
consuming
low
for
printing
memory
and
CPU.
E
Security
is
something
that
we
care
about,
because
security
is
something
that
security
is
something
that
is
never
ending
job
but
Bridge
Police.
We
are
trying
to
work
in
the
security
stuff,
so
it's
trying
to
getting
better
and
better
in
insecurity
and
Implement
more
features
like
now.
We
have
mtds
connections
for
all
ruthless
podman,
but
we
want
to
implement
USB
guard
marksec
among
this
kind
of
security,
tooling
that
we
we
have-
and
it's
important
on
the
on
that
if
you
can
go
to
the
next
slide.
E
So
the
use
case
that
we
focus
on
are
like
one
is
the
retail
companies
like
thinking
a
company
that
has
multiple
a
store
across
a
different
towns
like
in
USA's?
In
this
case,
this
retail
company
can
have
a
point
of
sale
system.
E
A
video
surveillance
system,
some
sensors
about
how
much
electricity
consumes
or
music
in
the
store
or
something
else,
so
a
lot
of
devices
that
can
be
part
of
that
and
what
we
want
is
to
manage
all
these
devices
inside
this
kubernetes
cluster
using
the
crds
and
you
can
Define
the
like
Edge
device
and
Edge
overload
and
running
this
workloads,
like
our
containers,
inside
of
the
old
retail
stores
that
you
want.
So
the
networking
there
can
be
like
any
store
can
lost
connectivity
in
any
moment
and
it's
not
reliable.
E
E
If
this
robotic
arm
performs
a
suspected
or
recently
down
when
it's
suited,
you
know
like
all
these
woodloads
that
come
running
in
the
photo
line,
so
maybe
this
kubernetes
cluster
can
be
inside
of
the
factory.
There
is
something
that
pushing
this
pushing
change
onto
that
kubernetes
cluster
with
the
edge
administrator.
So
this
is
more
or
less
the
use
case
that
that
we
can
think
of
also
like
you
can
think
about
trains,
drones
or
something
that
is
more
or
less
in
this
case,
but
more
or
less.
E
E
We
rely
on
utcd
share
this,
so
we
have
four
here
this
that
is
such
a
device.
It's
it's
biceps,
that
is
a
group
of
the
device,
a
single
request.
That
is
some
kind
of
approval
process
that
we
provide
flutter
operator,
is
taking
care
about
the
updated
some
do
the
reconciliation
needed
in
the
crdis
and
the
htpi
is
how
the
devices
communicate
with
the
kubernetes
cluster.
E
E
On
the
down,
you
can
see
the
float
that's
device.
In
this
case.
We
need
to
install
the
device
worker.
The
float
advice
worker
that
worker
is
the
only
thing
that
is
doing-
is
pulling
information
from
the
floater.
It's
API
on
her
bit
seconds.
That
is
like
we
have
a
bit
that
this
can
be
configured
by
by
The.
Edge
administrator
can
be
60
seconds
to
five
minutes.
E
This
device
worker
just
interact,
interacts
with
Postman
and
system
D,
because
we
want
to
have
the
minimal
footprint
possible.
We
rely
a
lot
in
the
system
D
like
running
the
workloads
with
using
systemd
that
is
donated
by
podman
to
get
in
the
information
about
the
containers,
but
we
only
pushed
a.
We
are
using
podman
preview
that
is
like
we
can
interact,
we
can
send
the
Pod
spec
directly
to
podman
is
getting
in
there
system.
E
E
The
first
thing
that
you
can
see
in
there
is
like
the
the
first
crd
that
you
can
have
when
you
start,
you
have
a
Raspberry
Pi,
your
style
float
advice,
agent,
RPM
and
as
soon
as
you
install
that
you're
registering
to
your
float
into
your
flutter
cluster,
you
will
have
this
device
finger
request.
That
is
the
approval
process
that
we
have.
E
You
will
get
the
name.
The
ID
can
be
from
the
matching
ID
from
the
installation
or
can
be
the
the
video
ID
uad
and
in
there
the
system
administration
need
to
do
need
to
approve
the
device.
If
you
start
trust
it
or
not,
used
to
run
the
initiable
playbooks
or
to
run
the
or
to
run
the
any
kind
of
the
any
kind
of
the
world
that
you
want
to
run.
E
So,
let's
administrator
need
to
say,
approve
through
it
will
have
like
the
information
that
in
there
like
you,
can
see
in
there
it's
good,
but
it
will
have
like
the
hardware
specification,
the
CPU
information,
the
system
vendor.
E
After
that
you
can
Define
the
target
namespace.
So,
for
example,
if
a
user
want
to
do
to
have
okay,
this
device
is
from
the
New
York
store,
so
come
have
like
a
name
space
for
these
devices
and
can
have
like
custom
config
maps
that
can
be
used
by
workloads
and
stuff
in
that
name
space.
So
the
separation
is,
is
total
in
terms
of
the
how
they
manage
that
and
they
can
also
Supply
labels
in
that
site.
E
So
next
slide
please
as
soon
as
they
the
device's
approval.
We
know
that
and
what
we
do
is
like.
We
created
The
Edge
device
that
edge
device
will
have
like
a
labels
that
are
managed
by
flutter
itself,
so
you
can
see
in
there
like
device,
CPU
device,
CPU
model
host
name
all
the
stuff.
This
is
created
by
by
flutter,
used
to
identification.
E
Because
we
run
the
workloads
based
on
labels,
we
have
like
the
device
selector
on
this
kind
of
tooling.
Like
any
service,
you
can
Define
any
labels
and
put
the
workloads
in
there
that
aspect
of
the
edge
device
is
high,
because
you
can
define
a
lot
of
things
in
there.
In
this
case,
you
can
see
in
there
like
the
hair
bit
like
every
one.
Every
two
minutes
you
will
get
the
latest
configuration
from
from
The
Edge
API.
E
If
something
changed,
it
will
apply
all
these
things
like,
okay
and
it
will
notice,
is
in
the
place
or
something
new
that
is
like
a
new
and
simple
script
need
to
run.
We
also
have
like
the
SSA,
the
RPM
of
three
information.
So
in
this
case
we
have
like
dios
information,
and
we
say:
okay,
we
we
are
going
to
break
if
a
new
commit
is
in
place,
so
it's
the
main
thing
that
we
can
say:
Okay
updated
as
soon
as
the
security
date
happens,
or
something
and
the
request
time.
E
That
is
a
way
to
put
the
configuration
for
all
the
devices.
So
in
there
you
can
see
like
it's
the
same
spec
as
device,
and
you
can
see
in
there
like.
In
this
case
we
use
the
metrics
and
we
say:
okay,
this
sample
set
is
going
to
store
the
metrics
for
200
megabyte
maximum
for
24
hours,
and
we
have
a
lot
of
collection
that
has
this
syslog
with
this
buffer
size
to
apply
this
set
device
on
the
this
Edge
device
that
into
the
device
you
need
to.
You
need
to
add
a
label.
E
And
now
the
device
is
fully
configured
I
mean
you
can
have
like
the
device
ready
to
get
in
workloads
to
getting
the
super
ability
from
the
system
monitor
or
whatever,
and
now
we
focus
more
in
the
overloads.
As
I
said,
we
schedule
the
the
wireless
based
on
labels
like
with
the
device
electros
and
his
service.
E
We
have
the
system,
the
integration
as
I
say
like
to
keeping
the
Bots
running
for
operative
time,
multiple
options
here
to
use
a
private
image
registry.
If
needed,
you
can
use
a
coffee
map
and
secrets
in
the
Pod
spec
too.
If
you
want
to
redefine
the
world
to
support
spec,
so
you
can
use
like
the
config
host
device
and
stuff.
E
We
also
have
a
way
to
say
this
workload
team
match
is
going
to
be
out
updated
in
case
of
the
devices
thinking
like
some
secret
data,
so
you
set
the
device
and
we
checked
that
the
if
the
image
is
updated
or
not
to
do
that
and
something
that
we
are
in
the
last
race
is
like
rude
Las
Palmas,
just
because
if
any
warlord
gets
compromised,
don't
get
access
to
the
whole
s
devices.
E
E
E
E
E
We
have
like
the
edge
device
configuration
like
you
can
say:
okay,
this
device
can
store
200
megabytes
of
megabytes,
200
megabytes
of
metrics
for
a
maximum
of
24
hours,
I'm
thinking
like
the
workload
is
running,
and
we
we
scraped
the
overload
metrics
that
is
defined
in
the
edge
for
load
and
we
Define.
Okay,
this
this
world
has
this
metrics
and
we
escaped
the
information
into
the
from
from
that.
E
We
store
that
information
into
the
device
and
we
send
it
in
batches
to
our
remote
tsdb,
like
Thanos
or
otherwise
time
stream
using
remote
right
from
pen
videos.
The
reason
that
we
did
this
is
because
thinking
that
you
lost
the
connectivity-
and
maybe
this
will
load
start
to
fail,
this
will
not
start
to
fail
like
the
because
there
is
no
connectivity.
E
If
you
don't
store
the
this
in
a
buffer,
you
will
love
that
metrics
and
you
will
just
like
all
this
all
this
stuff.
So
this
is
why
we
put
this
buffers
in
there,
because
we
want
to
store
that
in.
If
something
happens.
At
the
same
time,
we
in
the
next
release,
we
are
going
to
work
in
the
reducing
the
control
plan,
control,
plane,
bandwidth.
E
So
thinking
that
if
you
have
like
Edge
device
connected
by
3D
or
4G
or
whatever
the
Contour
paying
information
is
not
important
at
all.
So
maybe
we
at
a
in-depth
device
like
okay,
the
control
plane
can
only
take
500
kilobits
of
per
second
of
the
bandwidth
is
because
there
are
about
load
that
has
more
Purity
into
the
onto
the
onto
the
internet
connection.
So
this
is
why
we
added
all
this
kind
of
metrics
and
buffers.
E
We
have
the
limits,
because
we
know
that
we
can
kill
the
we
can
if
we
store
all
this
information
in
there,
and
we
know
that
we
can
kill
the
device
for
a
long
period
of
time,
but
we
provide
the
customers
the
way.
At
the
same
time,
this
is
pushing
information
from
device
to
the
parameters,
so
it
doesn't
need.
Parameters
doesn't
does
not
need
to
have
the
option
to
to
have
direct
access
to
the
to
the
workloads
itself,
and
this
case
is
spotman.
There
is
no
cni
at
all.
E
We
open
the
port,
so
it's
hard
to
get
that,
like
the
all
the
super
launch
Matrix
without
the
position
method
and
next
slide,
please
so
the
same
for
deluxe
collection.
So
we
assume
that
the
network
is
not
reliable
at
the
same
way
as
we
do
with
at
the
same
way
that
we
did
with
with
metrics.
We
have
like
the
buffer
size,
you
say
like
okay,
we
have
the
board
load,
the
container.
E
We
read
the
logs,
we
save
in
the
buffer,
and
we
send
it
to
to
the
server
that
we
need
to
send
to
the
server
that
we
need
to
send
to
the
using
the
DCP
organization
buttons
and
all
this
stuff.
The
main
reason
is
that
it's
like
we
don't
want
to
lose
the
information,
even
that
we
know
that
the
buffer
is
full.
If
the
buffer
is
full,
we
cannot
do
anything
at
all.
E
You
need
to
drop
the
the
Old
Logging,
but
we
know
that
we
are
not
going
to
kill
the
the
device,
but
we
know
that
we
certainly
love
the
connectivity.
We
make
sure
that
decompose
at
the
moment.
We
all
implement
this
law,
but
the
idea
is
to
implement
more
protocols
here,
like
Kafka
message
or
lucky
or
S3
directly.
So
it's
a
matter
of
a
user
trying
to
implement
it.
We
we
already
work
on
the
way
to
to
have
that
next
slide.
Please.
E
About
the
life
cycle
of
the
devices,
so
what
we?
What
we
do
is
like
we
do,
the
RPM
of
Street
technology,
you
don't
know,
is
you
have
a
PM,
you
install
it
and
you
rebooted
them
with
the
new
RPMs
in
there.
So
you
cannot
write
it
into
the
disk
itself.
E
E
Tooling,
like
okay
I
want
to
update
these
devices
to
this
commit
ID
or
you
can
say,
I
want
these
devices
to
be
out
updated
because
I
don't
really
care
about
the
the
scenes
and
we
also
implement
the
green
boot
scripts
like
if
something
happens,
you
get
updated.
You
get
update,
you
get
update
and
when
rebooted
the
the
server
something
is
failing-
and
there
is
like
this
kind
of
input
scripts
that
is
like
okay,
flutter
is
running
system
is
running.
What
menu
is
running.
E
Everything
is
working
as
we
would
like
we're
starting
the
workloads,
something
that
we
want
to
implement
in
the
next
workload
is
like:
okay,
what
happens
if
I
workload?
Thinking
like
just
some
words,
you
get
a
kernel
update
for
some
reason,
you
lost
the
GPU
driver
or
something
like
that
we
are
going
to
update
and
for
us
everything
is
working,
but
not
for
the
workload.
So
maybe
we
we
are
thinking
a
way
to
have
like
the
workload
making
a
green
boot
script.
E
E
E
Everything
is
in
the
in
the
flutter
project.
Now
is
the
moment
for
the
demo,
but
I
cannot
share
my
screen,
so
I
cannot
demo
that.
But
if
you
are
okay,
maybe
we
can
go
to
the
web
and
there
is
a
demo
like
RC
schema
demo
I'm
trying
to
see
that
demo
and
you
can
ask
questions
sooner.
If
you
are
okay.
E
D
E
So
now
we
see
like
there
is
no
a
single
request
from
from
there,
so
you
can
see
like
we
started
a
few
devices
and
we
created
the
the
namespace
for
New
York
now,
which
are
the
approve
it
to
all
the
devices.
D
E
Okay,
so
now
in
there
you
can
see
like
there
are
three
it's
devices
can
be
in
this
case
are
fake
I
mean
there
is
like
a
BMS
when
I'm
running
this,
so
many
research
devices
is
like
how
you
run
a
customer
will
see
like
the
devices
itself.
E
The
name
here
is
it's
fake
two,
but
can
be
the
YouTube
ID
or
any
kind
of
information
that
can
be
part
of
the
metadata
of
the
video
that
The
Fighter
device
supporting.
So
this
is
something
that
you
can,
the
customer
can
say,
or
the
user
can
say.
Okay,
this
is
the
the
right
way.
Now
we
label
some.
E
Some
devices
go
in
this
case
we
say:
IPP
camera
the
kisk,
and
now
it's
going
to
deploy
the
workloads.
E
E
And
this
is
the
same
for
cameras
and
you
see
like
a
everything's
in
there
and
now,
if
you
see
the
device
in
there,
it
is
like
the
status
of
the
devices.
In
this
case,
you
can
see
like
the
devices
are
deployed,
like
the
the
devices
are
the
plugin
like
the
the
tooling,
and
you
can
see
like.
Okay,
there
is
label
some
is
trying
to
to
deploy
the
contentions,
and
this
is
the
workloads
that
are
running
into
the
device
for
this
War
loads.
We
have
metrics
that
you
can
use
like.
E
There
are
metrics
that
you
can
use
in
grafana,
so
we
try
to
deploy
all
the
stuff
in
the
metrics
and
we
provide
some
kind
of
grafana.
That's
worse,
that
you
can
see
like
okay,
this
this
device
has
a
forward
loads.
E
Three
of
them
are
not
running
or
are
created,
but
are
aborted
or
what
happens
with
that,
and
also
all
the
tooling
that
I
mentioned
that
that
upload
possible
ability
and
other
stuff
and
more
or
less
this
is
flutter
My
Demo
today
was
a
bit
cooler,
I
mean
I'm.
Sorry,.
A
Great
I
know
it's
tough,
while
you're
talking
to
be
keeping
up
with
the
chat,
but
I
think
a
number
of
questions
came
in
to
the
chat
most
of
them,
but
not
all
got
answered
by
your
fellow
teammates
on
the
project.
A
Let
me
just
call
out
the
ones
that
didn't
get
answered
in
the
chat.
I'm
just
curious
as
to
how
the
device
to
control
plane
connection
works.
That
does
the
device
call
in
like
an
agent
where
it
initiates
the
an
outbound,
IP
or
other
connection,
or
does
it
go
the
other
direction.
E
E
We
use
that.
Oh
you.
D
C
E
A
E
Part
of
this
device,
the
I
think
that
the
default
is
60
or
120.,
but
can
be
like
for
hours
or
it
could
be
for
for
four
minutes.
A
E
And
after
that
is
what
podman
use
on
the
and
the
lens
itself,
the
last
time
that
I
saw
I
think
that
we
were
around
a
half
a
gigabyte
like
a
500
MX.
We
thought
with
all
the
stuff
in
there.
Okay.
E
A
E
Of
memory
I'm,
you
know
it's
like
it's
no
longer
two
gigabytes
but
I
mean
we
are.
We
would
like
to
work
on
the
Raspberry
Pi
zero,
two
that
is
like
for
kiosk
and
stuff,
but
we
have
problems
with
the
Wi-Fi.
This
is
still
things
to
work
in.
A
E
A
Does
this
anticipate
being
able
to
support
100
air
gapped
environments,
meaning
your
air
gap
100
of
the
time
you
know
and
I
realize
you
might
have
to
put
up
your
own
container
image
registry
and
things
like
that
be
behind
the
air
gap,
but
is
that
something
that
you're
prototyping
and
expecting
it
to
be
useful?
For
yes,.
E
C
E
S3
is
like
a
weight
that
to
share
data
between
devices
and
mystery
buckets.
Oh.
E
A
E
A
So
shared
storage,
basically
across
kind
of
a
wide
area,
out
an
edge.
That's
that
sounds
really
cool.
A
Okay,
I'll
have
to
I'll
have
to
think
about
that,
but
that
that
is
kind
of
a
unique
aspect
that
I
I
haven't
heard,
come
up
much,
but
should
have
a
lot
of
usefulness
and
I
assume
that
that's
sort
of
designed
to
be
asynchronous
too,
so
that
if
you
did
something
like
event,
driven
programming
and
one
device
got
severed,
that
it
could
kind
of
catch
up
at
at
a
later
date
as
to
what
was
going
on
on
other
Edge
devices.
E
C
Yeah,
basically,
we
do
a
sync
between
the
contents
of
the
bucket
and
what's
already
in
the
in
the
volume,
so
we
leave
it
up
to
the
to
the
Sim
process
to
catch
up
whenever
there's
a
connectivity,
but
we
we
aim
to
run
Florida
on
a
disconnected
environments.
So
we
assume
that
the
device
can
be
disconnected
at
any
point
in
time
during
some
time
is
not
specific.
C
But
that's
why
we
added
the
tsdb
inside
of
the
of
the
agent
and
we
added
a
log
capturing
and
all
these
things,
because
we
believe
this
is
important
to
keep
the
observability
alive.
Even
though
the
device
is
disconnected
and
then
you
can
once
connected,
you
can
push
it
to
the
contact
line
or
whatever
needs
to
be
pushed.
A
Okay,
thanks
I'll
at
this
point,
I
better,
stop
monopolizing
all
the
questions,
I'll
point
out
for
people
who
Miss
some
that
if
you
open
the
chat
thread,
there's
a
lot
of
questions.
Plus
answers
there
and
I've
made
an
attempt
to
go
and
cut
and
paste
things
from
the
chat
and
also
the
audio
into
the
agenda.
Node
stock.
So
there's
q
a
questions
and
answers
there.
But
if
anybody
else
has
anything
speak
up
now,
we've
still
got
a
few
more
minutes
left.
E
Regarding
the
regarding
the
deployment
and
the
bot
I
hope
we
move
that
way
just
card
I
mean
there
is
no
good
way
to
updated.
We
it's
something
that
we
maybe
we
work
on
like
how
to
do
the
upload,
like
a
blue
green
development,
something
that
we
need
to
figure
out
how
we
do,
and
mainly
because
this
disconnected
environments-
and
you
know
it's
like
it's
hard
to
to
get
it
right.
E
D
D
E
For
devices
it's
like
more
like
we
are
fighting
now
with
that
there
is
like
the
salinity
stuff,
and
you
can
say
you
can
mount
the
devices
I'm
sharing
into
the
process
pack
itself.
If
you
need
to
privilege
on
that,
you
can
use
the
sec.com
profiles
like
or
the
privileged
mode
there.
E
So
you
can,
you
can
say:
okay,
this
spot
has
access
to
the
carpet
Ming
or
you
know
all
this
method,
meaning
or
is
the
way
it's
going
to
be
laundry
this
one,
because
we
know
that
it's
difficult
I'm,
you
know
when
implements
like
devices,
is
always
a
tricky
to
do
like
they
will
wait.
We
are
working
on
that.
C
Yeah
also,
another
thing
to
consider
is
that
the
way
I
see
this
next
device
is
that
something
that
is
going
to
be
running
a
specifics
or
workloads
for
as
long
as
the
device
is
alive.
So
when
you
shoot
the
device
and
you
onboard
the
device
locally,
first
I
guess
that
you
will
first
set
it
up
properly
with
all
the
permissions,
so
you
can
run
those
workloads
once
they
are
deployed.
E
It's
it's
difficult
to
manage.
The
way
has
the
many
implications
that
you
know
even
that
we
love
to
do.
That
is
it's
difficult,
but
it
should
be
something
like
sit
down.
One
board
and
Trigger
another
one
see
that
everything
works.
If
not
coming
back-
and
you
know
it's
like
so
it's
super
tricky.
Sometimes
we
need
maybe
this
this
when
users
need
to
say.
Oh,
this
is
the
right
way
to
use
that
I
mean
when
we
see
like
iot
Solutions,
it's
more
like
a.
E
It
could
work
it's
something
that
we
aim
for,
like
you
have
like
your
application
in
your
Git
You.
E
Have
your
containers
something
get
updated
and
it's
like
yeah
use,
update
all
the
devices
that
you
have
in
there
like
thinking
about
any
company
that
has
a
lot
of
stores
across
the
USA
I
want
to
update
my
point
of
sale
system
in
the
night
just
putting
in
there
like
the
githubs
and
do
the
updates
over
the
post
system,
one
by
one
or
ten
by
ten-
and
you
know
it's
like
you-
have
multiple
options
to
to
work.
A
Okay,
we're
three
minutes
after
the
hour,
so
we're
pretty
much
at
the
limit,
but
last
call
if
anybody's
got
any
questions
by
the
way.
That
was
a
a
great
presentation.
I
really
appreciate
the
effort
you
put
into
both
the
project
itself,
as
well
as
the
presentation
I
think.
Even
though
we
got
a
little
rough
start,
a
lot
of
information
got
delivered
and
I
appreciate
it
anyway.
Last
call
for
questions.
A
Okay,
I'll
take
silence
as
none.
Thank
you
very
much
for
speaking
on
this
topic.
I
think
that
we
might
it's,
interestingly
enough
and
aligned
enough
with
this
group
that
maybe
we'll
ask
for
a
refresh
in
six
months
or
so,
because.
E
E
I
mean
that's
it
for
me.
Thank
you
for
having
us
I'm
I'm,
the
kubernetes
tunnel.
So
if
something
be
mean
there,
I'm
happy
to
reply
to
questions
and
I
mean
I
will
keep
you
updated
if
we
have
a
new
release
or
something
new
in
there.
I'm
happy
to
point
do
and
do
another
theme
with.
A
You
and
feel
free
to
drop
updates
in
the
channel
for
this
group
as
well.
We're
open
to
you
know
anything
that
is
under
the
open
source
umbrella
coming
in
as
the
news
or
promos.
So
thanks
thanks
again,
thanks
everybody
for
joining,
and
this
me
this
group
meets
every
two
weeks.
So
we'll
see
you
at
the
next
one
bye
bye-bye.