►
From YouTube: Kubernetes WG IoT Edge 20230322
Description
March 22, 2023 meeting of the CNCF IoT Edge Working Group. No agenda birds of a feather discussion of various edge related open source projects including a number of projects related to AI/ML where a pre trained component runs at edge.
A
Okay,
yes,
but
I
think
there's
a
lot
to
be
said
about
that.
You
know:
there's
you
use
these
laptops
that
might
have
been
recycled
even
if
the
battery
doesn't
work
if
you've
got
them
hooked
up
full
time
and,
like
you
said,
take
take
three
or
four
of
those
tape
them
together.
People
in,
like
you
know,
War
zones,
for
instance,
I
know
that
there's
a
big
push
for
having
mobile
data
centers
things
that
you
can
drive
around
in
vehicles
right.
You
know,
like
you,
said,
take
four
or
five
of
these
things.
A
B
B
And
I
don't
know
that
I'd
recommend,
you
know
for
a
production
scenario
for
a
Fortune
500
company
that
you
base
your
systems.
Uncobbled
together,
used
laptops,
but
hey
all
of
us
I
think
are
in
this
situation,
where
one
of
the
things
that
we
promoted
in
this
whole
home
lab
birds
of
a
feather
is
that
you
know.
When
do
you
get
a
chance
to
learn
things
play
around
to
the
point
where
you
might
break
stuff
on
the
production
systems?
B
You
know
having
kind
of
this
home
lab
to
play
around
with
or
experiment
with,
or
even
as
a
developer
lets.
You
have
a
learning
environment
where
you
can
make
mistakes
without
the
repercussions
of
doing
that
on
production
systems
and
I,
don't
know
it.
It
almost
to
me
brings
back
the
joy
of
being
a
kid
again
where
you
can
go
in
there
and
play
with
your
toys
and
knock
them
down
and
start
over
again
a
little
easier
than
the
production
systems.
A
A
It's
yeah,
I,
couldn't
I,
couldn't
keep
it
organized
in
boxes,
closed
and
be
kind
of
closet.
I
couldn't
find
things
I
couldn't
find
the
cables
I
needed
very
frustrating.
So
now
things
are
a
bit
more
organized,
so
yeah
I
know
what
you
mean
about
keeping
holding
on
to
things.
I've
often
found
myself
tempted
to
throw
something
in
the
bin
and
had
said
you
know,
go
back
in
there
and
dig
it
out
like
I
know.
I
I
think
I
could
use
that
again.
B
So
tomorrow
you
joined
late,
but
just
to
let
you
know
that
we
didn't
have
any
items
on
the
official
agenda,
but
we've
gone
into
birds
of
a
feather
mode
and
people
are
just
having
a
chat
of
things.
They
want
to
talk
about
or
stories
of
things
they've
discovered
recently.
So
if
you've
got
anything
in
those
categories,
you're
welcome
to
put
them
on
the
table,
as
is
anybody
else
here
on
this
call.
D
Might
I
add
them
there's
a
very
nice
I,
don't
know
if
it's
a
proprietary
stuff
something
but
there's
a
company
called
Edge
impulse.
That
does
a
lot
of
nice
things
for
you
when
it
comes
to
audio
and
machine
learning
on
the
edge
and
what
from
what
I
understand,
they
do
have
certain
sdks,
which
can
run
on
Arduino
potentas
or
Arduino
Nanos.
D
So
so
it's
that
well,
optimized
and
I
haven't
had
the
opportunity
to
play
with
it,
but
they
have
some
nice
tutorials
where
you
can
already
test
your
data
with
your
with
some
pretty
determined
data
sets
that
they
have
or
some
training
models
that
they
already
have
so
might
as
well.
It's
my
it's
worth,
checking
it
out
what.
D
Yeah
what
I
do
know
of
is
is
actually
as
far
as
I
recall.
It's
also
one
of
the
main
directors
of
this
company
is
the
person
who
designed
Co-op,
Zach
Shelby,
oh
so
yeah.
So
it's
quite
a
nice
thing.
I
haven't
had
the
opportunity
to
try
it,
but
it's
very
well
documented
and
there's
a
lot
of
activities
that
you
that
people
do
tend
to
do
in
this
particular
thing
and
it's
also
open
source.
So
it's
not
like
it's
completely
closed
off,
but.
C
B
I'll
help
you
take
a
look
at
it.
I
threw
out
another
one
on
the
chat
that
I
became
aware
of.
Unfortunately,
I
don't
believe
it
is
open
source,
but
it's
since
EML
and
the
context
where
I
became
aware
of
this
is
that
one
of
the
these
embedded
system,
chip
vendors-
has
done
some
workshops
demonstrating
this,
and
this
is
in
that
category
of
predictive
or
preventative
machine
maintenance
where
this
Scentsy
ml
can
hook
up
to
audio
or
ultrasound
sensors
and
anticipate
based
on
machine
learning.
B
It
can
monitor
machinery
for
health
and
preemptive
identification
of
potential
problems
and
looking
at
their
website.
Anyway.
It
looks
like
you
know
it.
It
has
extended
pretty
far
but
like
I
say,
this
is
I.
Think
one
of
the
biggest
opportunities
in
Edge
yeah,
the
whole
world
is
talking
about
things
kind
of
in
user
interface
and
search
replacement
with
chat,
GPT,
but
I
think
when
it
comes
to
machine
learning
and
AI.
B
It's
kind
of
interesting
because
I
suspect
that
there
might
be
two
or
three
more
of
these
and
there's
probably
more
to
come,
and
maybe
this
is
a
unaddressed
niche
for
open
source
Frederick
as
being
with
the
eclipse
Foundation,
where
you
host
a
lot
of
these
Edge
projects.
Are
you
aware
of
any
that
fall
in
this
category
in
kind
of
in
the
eclipse
Stables
that
would
do
ml
for
Edge.
C
We
have
some
of
our
projects
that
will
integrate
third-party
Frameworks
and
things
like
that.
But
I
don't
think
we
have
anything.
You
know
strictly
focused
on
on
that
particular
use
case.
So,
for
example,
Eclipse
Cora,
which
is
a
popular
solution
for
Gateway,
can
run
containers
and
that's
tight
integration.
Now
with
the
Nvidia
AI
stack,
so
they
they
had
to
talk
at
eclipsecon
describing
the
whole
workflow
and
it
was
quite
quite
interesting.
I
can
I
can
fish
out
what
I've
done
speaking
the
the
actual
link
to
the
video.
A
C
Yeah
and
apart
from
that
and
I
know
that
they
have
on
their
roadmap,
that
they
will
add,
support
for
auto,
auto
AI
Stacks
like
Pi
torch
and
and
other
so
that
it's
easier
to
work,
a
certain
that,
of
course.
Of
course,
you
know
our
general
general
purpose.
Edge
platforms
like
Eclipse
io4
can
run
AI
workloads,
but
I,
don't
think
the
team
has
put
together
any
any
specific
guidance
to
run
it
or
you're.
On
your
own
to
package,
your
containers
and
and
and
and
things
like
that,
I'll
do.
C
As
you
know,
the
team
at
edgeworks
has
been
has
been
working
on,
putting
together
AI
powered
cameras,
and
things
like
that.
So
so,
maybe,
if
you
are
nice
and
ask
on
their
slack
they
will.
They
will
give
you
some
pointers
on
how
best
leverage
it
for
for
your
or
your
things
that
that's
about
what
I
can
think
we
have.
We
have
something
called
called
The,
Deep
learning
for
jet
Eclipse
as
well,
but
this
is
really
a
Java
Java
type
of
framework,
so
probably
less
popular
with
this
particular
crowd.
Here.
B
Yeah
I
think
kind
of
it's
just
like
even
at
the
larger
scale,
the
kubernetes
orchestrator
anything
that
can
host
apps
could
host
an
app
that's
in
the
ml
category.
But
you
know
I,
think
there's
an
opportunity
here
for
things
to
provide
kind
of
the
framework
or
library,
of
course,
specifically
made
for
this
use
case
of
both
I
I
think
even
putting
together
the
training
and
the
ml
is
Big.
So
some
of
these
they've
already
pre-trained
a
model
and
the
project
is
delivering
the
model,
but
others
are
more
generic.
B
Where
they'd
help
you
build
your
own
for
some
specialized
use
case
and
I.
Think
there's
probably
some
already
out
there
and
more
to
come.
I
dropped
another
one
in
the
chat.
This
tiny,
tiny,
ML
and
you
know
I'm
aware
it
exists,
but
I
haven't
had
any
time
to
go,
look
into
it,
but
yet
another
one
that
is
out
there
and
appears
to
at
least
have
some
people
who
have
applied
it
on
some
hardware
for
Edge
applications.
Yeah.
C
B
I
did
notice
another
one
that
ended
up
on
our
list
of
open
source
for
Edge
too,
that
had
a
special.
It
sounds
like
it
was
originally
developed
to
be
navigation
for
autonomous
drones.
You
know
where
you
might
give
it
GPS
coordinates
that
the
Drone
is
to
fly
to,
and
maybe
even
a
multi-step
path
where
it
flies.
B
Some,
you
know
to
four
waypoints
and
then
returns
home
and
I
believe
some
of
these
are
popular
for
drone,
filming
where
you'd
send
it
on
a
known
path
to
do
reconnaissance
or
even
filming
for
the
purposes
of
making
a
movie.
You
know
that
a
lot
of
these
movies
have
drone
shots
now
and
you
could
have
it
fully
remote
controlled.
But
if
it's
to
go
over
a
path
of
a
mile-
and
you
know
the
the
Waypoint
mechanism-
maybe
is
a
a
better
way
to
do
it
and.
B
B
A
B
Don't
think
the
one
I
saw
was
was
driver
controlled,
though
with
yeah,
what
they
call
first
person
video
where
they
put
in
you
know
a
pair
of
goggles
and
I.
Don't
see
why?
Because
the
tree
track
doesn't
move
around
other
than
avoiding
collisions,
why
you
couldn't
turn
that
fully
autonomous
and
frankly,
even
in
collision
avoidance,
if
the,
if
some
kind
of
ml
got
the
camera
feed,
it
might
actually
be
better
at
avoiding
collisions
than
the.
E
D
D
D
That
one
of
those
projects
from
the
cliff
Foundation,
if
I
remember
correctly,
it's
from
I
think
Angelo
the
Zeno
project
correct
me
from
wrong
Frederick
right.
They
also
have
this
form
of
they
have
this
protocol,
which
apparently
they've
used
it
in
this
Indie
Car
autonomous
racing
systems
or
something
I
haven't
had
the
chance
to
Deep
dive
into
the
protocol,
but
I
think
it's
supported
by
the
eclipse,
Foundation
yeah.
C
Well,
that's
the
thing
both
both
Cyclone,
DDS
and
Xeno
have
been
involved
in
the
the
the
challenge,
but
you
are
correct:
Sean
they
weren't,
encapsulating
DDS
traffic
in
Zeno,
because
DDS
doesn't
doesn't
like
to
hop
from
one
network
type
to
another
church.
So
that's
how
they
they
relate,
Telemetry
and
and
comments
to
the
cars.
C
And
you
know
what
you
just
described:
the
Steve
makes
me
think
about
the
project.
We
that
that's
the
latest
one.
We
welcomed
in
our
working
group,
the
iot
one
that
eclipse
and
it's
called
the
athanis-
it's
not
a
way
to
run
AI
on
drones
on
a
day
like
that.
But
it's
a
way
to
offload
complex
mathematical
calculations
in
real
time,
so
it
could
be.
It
could
be
used,
of
course,
to
to
implement
a
higher
end
AI
framework,
and
then
you
push.
C
You
know
to
a
hedge
server,
the
the
the
the
the
hard
number
crunching
and
get
back
the
results,
and
it
seems
they.
They
are
working
on
a
hardware
accelerator
in
at
the
company
that
submitted
the
project
to
us.
That
will
complement
the
software,
but
you
can
you
can
use
the
software
platform,
it's
open
source
without
the
actual
Hardware.
D
Just
one
question
to
to
the
community
over
here:
if
I
may
I
am
part
of
professionally
I
am
into
Industrial
Automation
stuff,
so
of
course
I
am
into
Edge
Computing
and
per
se.
But
I
wanted
to
ask
the
community
about
one
particular
thing
that
I
think
is
a
bit
like
there's
a
for
me
as
someone
who's
working
on
with
Edge
devices,
because
we
do
VR
timer
should
do
create
a
generate
our
own
hardware
and
we
do
ship
them
with
specific
software.
This.
B
A
B
And
you
know,
there's
a
couple
different
approaches:
I
think
that
are
popular
one
is,
you
know
you
say:
kubernetes
is
too
complex
and
you
know
there's
many
aspects
of
that.
It
is
complex,
but
you
can
put
a
layer
on
it
to
where
you
run
an
application
program
that
does
the
kubernetes
parts
for
you
so
that
you
have
a
framework
where
you're
running
on
top
of
it.
There
are
also
attempts
that
I
would
say
described
as
modified
kubernetes.
B
Even
with
regard
to
compose,
you
could
potentially
still
use
it
but
cover
it
up
and
I
would
say
that
a
tool
like
portaner
could
remote
manage
unattended
nodes
that
actually
are
using
compose,
or
at
least
compose
yaml,
even
if
literally
compose
isn't
there
to
manage
the
workloads
at
Edge
and
cutting
edge.
Things
like
webassembly
are
out.
There
alternate
orchestrators,
some
contend
that
Nomad
is
better
than
kubernetes
and
some
of
these
become
religious
discussions.
B
A
Yeah
our
experience
has
been
in
this
space.
Is
that
and
I
see
this
a
lot
too
I
mean
even
in
engineering?
We
don't
you,
don't
get
oodles
and
oodles
of
people
in
the
population
that
understand
distributed
systems
or
want
to
work
in
that
area.
Right
data
science
as
popular
as
it
is
now
and
low
code.
No
code
begs
the
question.
You
know:
why
should
I
have
to
to
learn
other
Tooling
in
order
to
do
the
job
that
I
need
to
do
right,
I
just
want
to
get
it
done.
A
So
you
don't
get
hurt
and
it
makes
it
a
little
bit
more
attractive
and
easier
to
build
deployments
that
you
can
send
out.
So
I
would
say
that
you
know
it
depends
on
what
area
you're
coming
from.
If
it's,
if
it's
the
area
of
git
Ops,
CI
CD,
you
know
continuous
stuff,
then
I
think
that
that
area
is
pretty
well
maintained.
E
Therein
lies
the
challenge.
We
have
a
tendency
as
Engineers,
to
throw
people
in
the
deep
end
and
tell
them
they
have
to
set
up
their
own
kubernetes
cluster
as
the
first
step,
but
where
a
lot
of
people
have
had
success
is
leveraging
existed,
existing
managed
kubernetes
services,
like
the
cloud
service
providers,
offer
and
I
think
that
might
help
these
people
adopt
these
newer,
Cloud
Technologies
by
allowing
them
to
focus
on
specific
areas
to
work
in
adjacent
domains.
E
B
Make
the
observation
that
I
think
if
we
look
over
into
kind
of
the
data
analytics
realm,
that
I
would
contend
things
like
Jupiter
notebooks
were
a
a
tool
intended
to
help
a
data
analytics
person
who
just
wants
to
make
observations,
conclusions
predictions
from
a
pool
of
data
easy
without
getting
down
into
being
a
hardcore
programmer.
You
know
and
learning
programming
languages
in
detail.
That
was
a
very
successful
project.
B
That's
used
by
a
lot
of
researchers
and
perhaps
in
this
Edge
Community,
whether
it
be
applications
like
control
or
just
data
analytics
again,
there's
a
real
opportunity
for
these
kinds
of
tools
that
allow
people
to
not
get
into
the
details.
Even
in
classic
industrial
engineering,
I
or
industrial
control.
I
would
say
that
things
that
that
were
out
in
the
80s,
like
plc's
or
programmable
controllers,
were
a
rapper
that
gave
people
who
were
more
or
less
closer
to
the
electrician
training
capability
realm.
B
The
ability
to
do
things
without
learning
to
be
an
embedded
system
programmer,
and
you
know,
ultimately,
what
was
going
on
with
those
early
plcs
was
even
getting
down
into
machine
language
to
make
the
the
control
loops
operate.
Very
at
very
low
latencies,
very
predictably,
but
they
put
the
wrapper
of
things
like
PLC
ladder,
logic
and
things
up
there
to
bring
it
up
into
an
abstraction
that
people
were
already
comfortable
with
from
day
one
and
those
attempts
have
always
been
pretty
successful.
A
And
this
is
inspiring
because
what
it
allows
you
to
do
is
it
meets
the
developer,
the
engineer
where
they
are
and
and
almost
indicate,
no
matter
which
domain
they're
coming
from.
If
you
could
think
about,
let's
say:
writing
some
go
code,
and
in
that
code
you
dispatch
something
to
take
place,
an
action,
a
function
when
that
gets
dispatched.
A
So
I
think
you're
going
to
see
a
lot
more
of
this
too,
with
the
Advent
of
chat
GPT,
because
people
want
to
run
just
about.
You
know
want
to
execute
Things
based
on
conversation
now
and
so
very
good
possibility
that
you'll
find
I,
think
things
like
chat
GPT
influencing
other
apis
to
to
be
ready
to
take
instruction
and
to
build
out
and
provision
things
on
your
behalf
without
you
knowing
you
know
the
particulars
of
how
it's
being
done.
D
Yeah
everything
that
makes
sense,
however,
like
it's
it's
quite
interesting,
to
observe
I've,
been
observing
this
pattern
quite
a
lot,
at
least
in
the
Industrial
Automation
space
is
there's
a
big
boom,
as
Andy
mentioned
about
low
code
or
no
code,
Logics
of
things
like
tools
like
node
red
right.
That's
that's
a
massive
thing.
It
gets
served
by
almost
all
the
biggest
Industrial
Automation
providers,
because
it
just
makes
it
easier
for
a
control
engineer
to
just
create
some
flows.
D
D
But
a
lot
of
times,
I've
observed
that
it's
mostly
the
case
that
a
lot
of
requirements
in
Industrial
Automation
at
this
point
of
time
is
just
one-to-one
mapping
with
the
physical
device
itself,
which,
at
certain
point
of
time
you
I
feel
like
kubernetes
might
be
an
Overkill,
of
course,
but
I
mean,
of
course,
with
different
filtering
logic
and
some
some
tags
and
everything
you
can
achieve
that
kind
of
things.
But
yeah
it's
quite
it's
it's
an
interesting
observation
that
I've
been
currently
trying
to
look
into
as
well
as
try
to
solve
right.
D
So
everybody
wants
a
Time
series
data
based
on
each
and
every
device
that
they
have
so
that
they
could
have
a
nice
dashboard
to
to
show
the
management
perspective,
to
show
the
management
Personnel,
how
things
are
working.
So
that's,
unfortunately,
the
case
where
they
won't
accept
a
lot
of
high
level
tooling,
like
orchestration
orchestrators
like
Nomad
or
kubernetes,
because
for
for
them
they
also
are
concerned
about
the
fact
that
if
this,
if
things
go
down
really
bad
troubleshooting,
a
kubernetes
cluster
might
be
extremely
difficult
for
them.
On
the
long
run,.
A
Shannon
I
dropped
a
link
in
the
chat.
It
was
a
demo
day
that
we
did
for
kcp
Edge
a
while
back
where
we
showed
an
open,
AI
GPT
model
to
modifying,
manipulating
declare
yaml
declaration
for
right
config
for
seven
different.
You
know
it's
heterogeneous
or
Raspberry
Pi's,
let's
just
say
those:
they
were
each
outfitted
with
either
a
sense
hat,
not
a
sense
hat,
a
camera,
not
a
camera
or
they
were
in
a
region
or
not
in
a
region.
A
So
you
have
to
change
things
like
date
and
time
considerations,
functionality,
capabilities
so
that
the
the
pods
don't
fall
over
right
because
you've
tried
to
address
the
camera
when
the
camera
doesn't
exist.
So
this
you
you're
talking
about
it
as
a
one-to-one
relationship
I'm
talking
about
the
cardinality,
as
in
million
to
one,
how
do
I
get
a
million
different
objects
or
different
heterogeneous
configurations
to
deploy
without
having
to
Hand
by
you
know,
go
in
there
by
hand
and
change
the
configuration,
and
so
that's
definitely
something
that's
something
we're
researching.
A
D
Yeah
yo,
so
just
back
to
the
point,
I
think
Andy
you're,
absolutely
right,
I
completely
agree
with
it.
I
I
was
just
trying
to
play
around
with
this
concept
of.
If,
if
I
were
to
add
some
templating
logic
to
my
Docker
compose
applications
right,
how
would
I
go
around
doing
it?
So
I
I
just
recently
started
working
on
this
repo
of
mine,
which
is
completely
open,
source
and
I
just
recently
found
out
that
Frederick
might
be.
They
still
open
places
for
for
the
virtual
Eclipse
virtual
iot
day.
D
So
I
probably
might
give
a
talk
about
it,
but
it's
just
a
it's.
The
same
logic
as
Andy
was
mentioning
right:
a
wrap,
an
abstraction
layer
over
your
Docker
compose
file.
So
all
you
have
to
do
is
take
care
of
two
files
and
it
will
do
the
rest
for
you
and
I
just
recently
found
out
that
you
can
use
a
Docker
compose
command
on
your
let's
say,
host
machine
and
tell
the
command
to
execute
the
whole
stack
on
a
remote
machine
too.
So,
which
is
quite
interesting.
It
tends
to
serve
a
lot
of
problem.
B
C
E
B
Watched
that-
and
it
was
really
cool
but
there's
a
bigger
picture.
You
know
a
lot
of
people
make
the
mistake
of
doing
a
proof
of
a
concept
and
they
get
one
of
them
working
by
tuning
it
by
hand
right,
but
what
they
don't
think
of.
They
remember
the
effort
of
how
hard
it
was
to
do
their
compose
file
and
get
it
to
work.
B
But
the
bigger
picture
is
doing
this
at
scale
with
things
like
policy
enforcement,
so
people
don't
make
security
mistakes
keeping
audit
Trails,
because,
unfortunately,
hackers
are
out
there
in
these
Edge
locations
being
unattended
with
no
physical
security
are
going
to
be
targets
and
they
historically
have
been.
So
you
know
it's
not
if
it's
going
to
happen,
it's
just
how
often
you
need
the
ability
to
do
rollbacks
because
mistakes
happen,
human
error
or
whatever
absolutely
so.
B
You
know
I
think
that
people
have
had
a
lot
of
success
using
this
get
Ops
Paradigm,
where
you
basically
just
like
get,
has
a
reversible
audit
trail
of
what
changes
were
made.
Made
To
Source,
if
you
put
kind
of
your
configuration
and
infrastructure
settings
in
a
git
repository,
and
do
that
in
a
way
that
you
can
operate
it
at
scale
that,
like
I,
say
a
number
of
people
have
been
successful
at
that.
B
Maybe
people
will
come
up
with
even
better
ideas,
but
what
goes
on
in
these
things
is
bigger
than
just
the
yaml
for
one
app
at
a
time,
I.
Think
kind
of
the
looking
ahead.
Two
or
three
moves
on
the
chessboard.
You
got
to
look
at
observability
and
tracing
and
I,
don't
think
you'll
ever
if
you're
operating
a
hundred
thousand
nodes,
even
10
000.
Certainly
a
million
you'll
never
get
away
with
the
need
for
experts
who
understand
all
these
abstraction
layers.
B
The
abstraction
layers
will
be
a
force
multiplier
for
people
who
don't
want
to
go
drill
down
into
the
deep
dive
details.
But
sooner
or
later
you
run
into
a
tough
to
diagnose,
root,
cause
analysis
problem.
Where
you
know
people
are
reporting.
This
thing
just
is
running
slow
and
if
you've
got
in
abstraction
layers,
I've
never
I've
been
around
long
enough
with
enough
scar
tissue.
B
That
I
think
you
always
need
that
backstop
of
either
one
person
or
a
team
of
people
who
can
go
in
there
and
get
together
in
a
room
whether
it's
physical
or
zoom,
and
be
able
to
drill
down
to
analyze,
what's
really
going
on
and
to
enable
that
you
need
things
like
logs
tracing
data
classic
observability
and
you
kind
of
need
to
put
that
in
place
from
the
get-go.
Not
when
you
have
this
horrible
problem,
and
you
retroactively
wish
you
had
this
instrumentation
that
you
didn't
think
of
at
the
beginning.
Yeah.
A
Yeah
this
notion
of
day
two
existing
while
day
one
is
being
you
know,
worked
on
is
I
think
is
a
is
a
great
idea.
Not
many
people
follow
it,
but
we're
trying
to
bring
those
ideas
together,
but
as
I
was
saying
to
Shannon
earlier,
is
that
sometimes
you
might
even
be
able
to
fix
the
problem
on
your
own
right,
but
you
can't
explain
it
necessarily
in
a
way
that
you
know
you're
talking
about
a
kubernetes
deployment
or
a
pod
or
a
container
has
failed.
A
B
B
D
D
Yeah
I
think
I
agree,
but
there's
one
one
caveat
to
I.
Think
the
AIML
talks
is
the
case
that
predictive
maintenance
for
one
person
might
be
absolutely
different
in
terms
of
context
for
someone
else
right,
like
I
mean
we
get
it.
It's
a
predict.
You
try
to
predict
all
this
this
and
that,
but
like,
for
example,
the
data
sets
the
the
ontology
behind
it.
The
the
naming
structures
or
the
schemas
or
something
might
apps,
might
vary
drastically
between
just
two
entities
of
the
same
huge
Enterprise
or
Corporation.
So
I
think
it
tools.
D
B
Before
we
run
out
of
time,
one
last
thing
I
wanted
to
throw
out
and
maybe
there's
more
if
other
people
have
other
items,
but
the
kubernetes
on
the
edge
Day
is
coming
up
in
Europe
in
Amsterdam
in
about
a
month
they
just
posted
the
schedule.
I
put
a
link
in
the
chat
it
rolled
off,
so
I
put
it
in
the
chat
about
15
minutes
ago.
B
Unfortunately,
the
event
is
already
sold
out,
and
maybe
this
is
a
testament
to
how
hot
the
edge
topic
is,
but
the
event
actually
sold
out
before
the
schedule
of
what
the
talks
were
was
posted.
So
you
know
they
need
a
bigger
bigger
facility,
a
room,
I
guess,
but
it
will
be
there
and
if
anybody
was
headed
to
kubecon,
I
will
be
there
so
track
me
down
and
let
Let's,
let's
maybe
have
some
kind
of
face-to-face
physical
meeting
of
the
people
who
are
active
in
kubernetes
on
the
edge
space.
A
A
A
B
If
you
were
here,
I'd
say
you
know,
maybe
we
can
have
a
meeting
at
a
physical
Meetup
we're
trying
to
bring
back
the
physical
kubernetes
meetup
groups
that
we're
operating
monthly
in
Los
Angeles
and
got
killed
by
covet.
But
we've
had
recent
discussions
to
start
that
series
up
again
here
in
La
they
were
drawing
a
couple
hundred
people
a
month,
wow
back
at
the
peak
and
I'm
sure.
Boston
is
big
enough
to
possibly
have.
B
A
Yeah
we've
got
a
we've,
got
an
entry
and
we're
accepted
the
get
Ops
con.
We're
going
to
do
a
talk
on
so
I
guess:
I'll
I'll
plug
that
here
we're
going
to
do
a
upload,
a
Dem
show
in
a
lightning
talk
how
we've
experimented
with
Argo
CD
at
scale
and
where
the
bottlenecks
are
and
we're
also
working
with
the
Argo
CD
folks
to
to
fix
those
and
work
on
benchmarking.
C
B
Okay,
so
we've
reached
the
top
of
the
hour
and
some
people
already
in
the
chat
indicated
they
have
to
run
so,
let's
close
this,
so
thanks,
everybody
for
coming
I
think
there
is
an
agenda
item
for
the
next
meeting.
That's
in
a
couple
weeks,
so
maybe
it'll
be
a
little
more
structured,
but
thanks
for
coming
and
we'll
see,
you
then
bye.