►
From YouTube: KBE Insider Detroit - Josh Berkus, Red Hat
Description
We catch up with Josh Berkus, Kubernetes Community Manager at Red Hat on his various involvement with CNCF and he shares the latest and greatest from the Kubernetes Contributors Summit North America. Josh talks about stateful applications in Kubernetes, how he enjoys coaching speakers for KubeCon, and we discuss the importance of inclusion and the value of different perspectives and its impact on improving open source technologies. Watch this KBE Insider episode and learn more on how you can get involved and contribute to the community.
A
B
Yeah
yeah,
so
one
of
the
the
events
that
always
happens
at
the
beginning
of
kubecon
is
the
kubernetes
contributor
Summit
right.
So
this
is
run
by
my
Sig.
My
committing
kubernetes
called
Sig
contributor
experience
in
order
to
have
a
place
for
kubernetes
contributors
in
this
case
about
was
150
200
kubernetes
contributors
to
get
together
and
discuss
what
they're
working
on
in
all
sorts
of
different
ways.
So
we
had
discussions
about
the
process
of
proposing
major
changes
and
discussions
about
I.
B
B
B
B
The
so
yeah
so
everybody
gets
together
and-
and
we
have
all
these
different
discussions
throughout
the
day
about
you
know
and
and
honestly
for
a
lot
of
people.
It's
just
a
chance
to
be
face
to
face,
because
you
know
you're
on
Zoom
calls
right.
You
know
chat
with
somebody
for
like
a
year
straight
right
and
it's
really
super
helpful
for
for
seeing
people
face
to
face
somebody.
A
No
way
really
yeah,
oh
that's
so
interesting.
I
still
remember
like-
and
this
is
you
know,
but
I
still
remember
our
guys.
You
know
who
worked
on
a
remote
team
team
was
all
fully
remote,
and
this
was
like
20
years
ago
so
like
it
was
really
unusual
to
have
a
remote
team,
but
it
was
at
Oracle
and
on
a
quarterly
basis,
part
of
their
commitment
to
like,
because
they
wanted
this
team
to
exist
and
be
remote.
B
A
A
Yeah
no
I
totally
get
that
it
is
and
I
think
that's
one
of
the
things
that
we're
we're
really
experiencing.
You
know
with
the
pandemic
and
everything
else
we're
starting
at
the
hang
of
well,
that's
cool,
so
tell
me
some
more
about
stateful
applications,
so
I
mean
aside
from
they
are
a
pain
in
the
butt.
B
Yeah,
what
is
the
opinion
about
the?
So
we
were
actually
discussing
this
in
in
the
session
that
sort
of
thing
talking
about
things
that
we
still
wanted
to
do
in
kubernetes
for
stateful
applications,
because
we
had
sort
of
a
couple
first,
cuts
which
is
stateful
sets
and
crds
and
operators,
but
nobody's
really
satisfied
with
that,
because
there's
a
whole
bunch
of
problems
with
real
life,
stateful
applications
that
doesn't
leave
solved
right
because
I
mean
first
of
all,
they're
all
different
kinds
of
stateful
applications.
B
B
So
you
need
this
whole.
This
huge
amount
of
customization,
which
led
us
to
things
like
crds
and
operators.
But
the
problem
with
an
operator
is
that
you
are
basically
writing
your
own
infrastructure
software
right
or
using
something
written
for
somebody
else
to
attach
that
to
kubernetes
and
kubernetes
is
not
really
providing
you
with
a
framework
except
a
place
to
attach
that
plugin
right
right
and
so
as
a
result.
We
have
this
whole
Marketplace
of
database
operators
and
other
stateful
applications.
Operators
that
are
in
what
we
call
the
operator
framework
level.
B
A
B
A
B
A
B
Well,
there's
a
I
mean
there's
a
lot
of
sort
of
separate
areas
to
address.
You
know
one
of
the
ones
that
I've
personally
worked
on
is
Resource
Management,
which,
where
I'm
working
with
the
people
who
are
in
run
times
right
because
now
we
have
c
groups
V2.
We
have
more
control
over
Resource
Management
on
right,
the
individual
machines,
but
at
the
same
time-
and
this
is
another
meeting
at
the
contributor
Summit
right-
the
runtime
people
all
met
up
and
by
runtime
I
mean
like
cryo
containerd
Etc.
B
A
lot
of
how
kubernetes
interacts
with
the
container
runtimes
was
built
around
Docker,
and
a
lot
of
that
was
some
really
ugly
workarounds,
because
Docker
was
not
designed
to
be
controlled
by
an
orchestration
system
right
now
that
Docker
is
no
longer
supported,
they
can
get
rid
of
those
workarounds,
and
so
the
runtime
people
were
discussing.
Okay,
if
we're
getting
rid
of.
B
Yeah,
the
so
I
I
mean
it's
funny,
because
people
outside
kubernetes
look
at
it
and
they're.
Saying:
okay,
kubernetes
is
being
used
in
production,
a
whole
bunch
of
places.
You
can
use
it
to
run
a
cluster
with
thousands
of
nodes
Etc.
You
know.
Most
of
the
big
problems
are
solved.
People
are
moving
on
to
doing
platform,
engineering
and
developing
stuff
in.
A
A
I
think
that's
one
of
the
things
that
I
think
is
like
I
appreciate
about
the
communities
project,
and
it
is
one
of
those
problems
that
you
see
in
a
lot
of
mature
projects.
He's
like
okay,
when
you're
mature.
Do
you
recognize
the
fact
that
that
doesn't
have
anything
to
do
with
being
done
or
do
you
kind
of
just?
You
know
like
keep
making
your
mature
thing
faster
right,
like
I
I,
appreciate
that
I
think
kubernetes
in
a
lot
of
ways
is
is
saying:
okay,
we've
got
a
certain
level
of
maturity.
A
B
Well
and
I
mean
I
like
kubernetes
approach
right
and,
and
it's
honestly,
one
of
the
reasons
why
kubernetes
is
kind
of
won
out
over
misos
was
that
misos
immediately
went
for
the
hard
problems,
but
the
difficulty.
The
problem
with
that
was,
you
had
to
be
a
computer
scientist
with
a
PhD
almost
to
contribute.
B
A
And
and
oddly
I
thought
with
Bezos
actually
is
that
and
it
felt
like
they
were
targeting
the
individual
developer
in
their
early
stage.
Like
you
know,
here's
you
know,
here's
how
to
get
started
or
whatever
and
I
was
always
very
much
that
in
the
sense
of
I,
feel
like
I'm
I'm,
doing
a
really
heavy
lift
to
get
my
stupid
thing
to
run
yeah.
You
know
and
I
don't
know,
I
just
always
thought
that
was
kind
of
interesting,
but
yeah.
So
kubernetes
has
been
doing
some
interesting
stuff
there.
A
A
But
if
we
talk
a
little
bit
about
what
happened
before,
you
know
we're
at
least
I
was
joking
around
about
I've
been
doing
container
stuff
since,
like
2012
and
I,
think
it's
funny
that
you
don't
recognize
that
you
need
orchestration
until
you've
like
shot
yourself
in
the
foot,
a
bunch
of
times
with
containers
already
and
then
you're
like.
Oh
now,
I
get
it
you
know.
So
what
how
do
you
feel
about?
What's
your?
What's
your
interesting
stories
about
the
history
of
kubernetes.
B
A
B
A
B
It,
the
and
and
kubernetes
was
in
the
middle,
and
the
thing
is,
if
you're
in
the
middle,
you
never
know.
If
it's
the
Hat
you
only
history
tells
you
is
that
the
happy
medium
or
the
undesirable
right
right,
because
because,
if
you
look
at
it
a
different
way
like,
for
example,
if
you
compare
this
to
my
history
of
databases
right
when
the
new
non-relational
databases
emerged
on
the
scene,
yeah
like
mongodb
Etc
right,
it
put
MySQL
in
a
bad
position
where
they
were
kind
of
sandwiched
between
mongodb
yeah.
A
I
think
that
people
who
are
in
deep
dark
inside
of
like
building
you
know,
kubernetes
or
whatever-
is
that
it's
so
difficult
to
wrap
your
head
around
when
you're
talking
when
you're
building
distributed
systems
not
having
a
visualization
is
very
very
difficult
because
you,
the
kind
of
the
goal,
is
that
you
don't
really
know
how
the
whole
thing
works
and
sometimes,
as
an
engineer
you,
you
kind
of
need
to
get
a
tighter
picture
than
that
and
but
generating
that
from
a
bunch
of
ammo
files
is
very
difficult
to
do.
Conceptually
yeah.
A
A
B
And
and
it's
like
Schrodinger's
Box
yeah
well-
and
we
have
this
problem
all
the
time
with
relational
databases.
Right
is
that
people
would
use
entity
diagrams
right.
You
know
or
other
orm
tools
as
their
way
into
the
database,
and
that
was
nice,
but
that's
not
how
the
data
is
actually
being
stored,
right,
right
and
and
sometimes
the
gap
between
your
representation
and
how
it's
actually
happening,
and
particularly
particularly
there's
a
tendency
of
abstractions
to
reduce,
distributed
computing
to
sequential
to
procedural
logic
right,
which
it
really
isn't.
It
very
specifically
isn't.
B
Yeah
and-
and
this
actually
goes
back
to
one
of
the
hard
things
to
teach
people
which
is
you
know,
how
do
you
teach
people
to
really
understand
concurrent
execution
right
yeah,
because
that
was
always
it
was
in
the
database
world.
That
was
a
battle
for
us,
because
all
the
time
people
would
be
saying,
oh
we're
trying
to
debug
it,
and
this
transaction
is
doing
this
and
this
other
transaction
doing
that
next
and
I'm
like
no.
No,
there
is
no
next
right
right.
You
have
16
cores
in
this
machine.
B
Those
are
happening
at
the
same
time
right
right
and
you
can
never
actually
determine
the
order
in
which
they
happen
in
any
reliable
way
right,
the
so
the
and
then,
when
you
throw
that
across
a
whole
bunch
of
computers
like
we
do
for
kubernetes.
It's
really
true
right,
because
you
don't
even
have
unless
you
have
a
very
expensive
setup,
you
don't
even
have
clocks
that
you
can
rely
on
for
with
any
level.
B
A
Their
heads
around-
and
this
is
the
funniest
thing
I
find
about
this
too-
is
that
people
also
have
a
really
hard
time
getting
their
head
wrapped
or
an
event
driven
architecture.
Yet
event,
architecture
actually
simplifies
most
of
this
conversation
right
because
you
know
the
thing:
the
order
of
things,
because
it's
by
when
the
event
occurred.
So
you
don't
necessarily
know
whether
this
event
occurred
before
that
event.
But
you
know
that
when
this
event
fired
this
thing
happened
and
then,
when
that
event
fired
that
thing
you
know,
like
so
I,
don't
know.
A
B
A
B
B
A
B
Is
really
not
where
we
should
be
in
terms
of
developer
interface,
right,
right
developers
should
be
writing
stuff
and
their
own
languages,
and
the
yaml
file
should
be
produced
by
a
tool
right
right,
the
and
and
we're
getting
that
right,
because
we
now
have
all
kinds
of
uis
and
code
ready
and-
and
we
have
now,
people
are
getting
into
serverless
stuff
and
you
know
which
provides
a
more
developer
friendly
layer.
There.
A
A
A
You
know,
but
it's
like
I've
had
all
of
these
relationships
with
this
one
person,
but
I
still
remember
him
like
giving
me
a
super
hard
time
because
of
I
had
a
racing
thread
problem
in
some
code.
I
had
written
and,
and
it
was
like-
and
he
was
like
what
why
you
know.
Why
did
you
do
this?
Whatever
I
was
like
yeah
and
I'm?
A
Looking
at
it
and
I'm,
like
you
know
and
and
passed
me,
was
apparently
an
idiot
because
I
looked
at
it
I'm
like
yeah,
there's,
obviously
a
race
condition
right
here
and
it's
just
always
like
wrapping
your
head
around
actual
stuff.
That
is,
you
know,
truly
multi-threaded
or
simultaneously
executing
or
whatever
is
it's
super
hard
and
it's
just
so
prone
to
error
and-
and
it's
also
very
difficult
to
test-
you
know
so
I
think.
That's
also
part
of
the
challenge
so
you're
enjoying
Bell
out.
Did
you
notice,
I?
A
B
A
B
B
A
Yeah
yeah
it's
funny
as
the
more
I
get
into
something
I
start
to
be
much
more
interested
in
the
more
esoteric
stuff,
yeah
I.
Think
because
it's
like
that's
the
stuff,
I
haven't
even
thought
about.
Yet
you
know
I,
don't
know
it's!
It's
cool,
I,
completely
agree.
I
gotta
check
this
out.
I,
don't
think
I
knew
about
that.
Yeah.
B
A
A
B
B
B
A
A
B
A
A
B
Like
that,
and
it
gets
really
important
when
we're
entering
whatever
we're
we're
on
right
now
year,
seven,
something
like
that
of
kubernetes
right
where,
like
we
know
that
there
were
a
bunch
of
assumptions
built
into
kubernetes
originally
that
were
based
that
were
circumstantial
right
and,
and
you
know
we
had
talks
in
contributor,
Summit
and
stuff
about
you
know
ripping
some
of
those
out,
but
we're
so
used
to
them.
We're
not
going
to
recognize
all
of
them
right
for
sure,
yeah
and
and
having
a
new
person
show
up
and
saying
you
know
hey.
A
Like
you
know,
even
if
it's
even
if
it's
past
me
versus
future
me's
perspective
yeah,
it's
still
a
different
perspective
yeah
but
yeah.
When
you
have
I
mean
say
you
know,
the
kind
of
the
beauty
of
Open
Source
right
is
the
you
know
the
many
eyes
thing
you
know,
but
you've
gotta
there's
a
little
bit
more
to
it
than
just
saying
you
know
many
eyes.
A
You
know
you
actually
have
to
make
it
possible,
you
know
for
for
them
to
do
it
and
you
know,
and
the
problem
with
big
you
know
industry
conferences
right
is
they
want
to
draw
a
crowd?
You
know
whether
whether
they're
making
a
profit
or
not?
It's
still,
like
you
know,
a
business
right.
You
want
to
have
when
you
run
a
conference,
you
want
to
have
a
group
of
people
who
want
to
come
right
and
the
way
you
get.
That
is
by
having
big
name
speakers.
B
You
can
never
get
any
more
big
names
well,
and
one
of
the
things
I've
liked
about
kubecon
is
even
as
much
as
it
becomes
sort
of
a
large
commercial
conference
and
stuff.
They
have
stayed
dedicated
to
inclusion
and
and
trying
to
attract
new
speakers
and
everything
else
right
like
to
the
new
speakers.
I
was
coaching,
are
actually
going
to
be
giving
Keynotes.
Oh
wow,
cool
nice,
so
yeah
the
and
you
know,
and
and
that's
not
taken
for
granted
right.
B
A
Well,
it's
you
know,
it's
like
how
do
you
get
the
next?
You
know
big
name
speaker
if
you
don't
give
any
opportunities
to
new
people
right
yeah,
it's
that's
one
thing,
I
think
kubecon
is
really
interesting
for
I
mean
you
know,
you
even
see
it
in
the
cfp
process
of
like
you
know,
hey,
you
know
they
ask
you
a
lot
of
kind
of
relatively
detailed
questions
about
how
you
know.
How
is
this
going
to
widen
the
kubernetes
community,
essentially
before
they're,
going
to
choose
your
stock?
B
A
A
Know
you
know
it's,
you
see
it
it's
mostly
in
Java
these
days,
but
it's
the
same
kind
of
concept
with
wasam
or
candy.
One
of
the
that's
one
of
the
models
right
like
Envoy
uses,
is
basically
they're,
essentially
doing
aspect
programming
except
using
wasam
to
deliver
it
kind
of
along
the
wire,
but
kind
of
awesome
being
everywhere
has
some
really
interesting
possibilities
and
characteristics
like
outside
of
the
kubernetes
world
as
well?
You
know,
just
like
you
know,
your
browser
being
running.
A
B
You
know
what
about
server
side
was
right,
you
know
Wazee
and
that
sort
of
thing,
and
then
how
do
we
run
that
on
kubernetes,
because
we
just
have
this
whole
like
discussion
with
somebody
rejects
who's
working
on
one
of
the
wasm
based
platforms?
Okay
and-
and
you
know,
and
and
they
were
talking
about
some
of
the
problems
that
they're
trying
to
solve
for
server
side
and
I'm,
like
you
know,
k-native
actually
already
does
all
of
that.
It
kind
of
feels
like
we
just
need
a
way
for
K
native
to
run
Blossom.
B
B
Right
yeah,
so
containers
are
the
abstraction
that
we
have
right
and
and
there's
a
lot
of
trying
to
reduce
it
to
something
that
looks
like
a
container
yeah.
But
but
that's
one
of
those
things
that
you'll
reevaluate
all
the
time
right,
we're
talking
about
reevaluating,
runtimes
and
stuff
is
like
well.
How
much
does
it
have
to
look
like
a
container
right?
The
and
you
know,
and
can
we
make
other
things
look
sufficiently
like
containers
right,
particularly
and
and
not
make
it
like
bespoke
right.
B
Unconfigured
right
the
and
up
until
now,
there's
been
a
lot
of
focus
on
doing
that
with
VMS
right
from
the
other
end,
because
there's
times
that
you
know
you
want
to
use
a
VM
rather
than
than
a
container
right,
you
know
not
just
I
mean
a
lot
of
folks
have
been
in
the
Legacy
use
case
right.
You
know
where
you
have
applications
that
are
already
optimized
for
VMS,
but
there's
also
other
use
cases
where,
if.
A
B
Nr
yeah
yeah,
yeah,
yeah
and
the
yeah,
and
for
that
matter
right,
if
you
want
to
link
into
CPU
level
encryption
right,
you
know
all
of
these
sorts
of
other
things,
but
there's
no
reason
that
we
can't
do
this
at
the
other
end
right
and
look
at
things
like
wasm
and
stuff
and
say:
okay.
Well,
on
the
one
hand,
we
can
go
sort
of
heavier
weight
than
containers
and
have
VMS
that
kubernetes
treats
like
our
containers
right
or
pods.
It
tends
to
treat
the
VMS
Molly
pods.
B
And-
and
you
know
have
kubernetes
orchestrating
that
as
well
right,
because
because
ultimately,
the
Goro
kubernetes
is
an
API
and
a
scheduler
for
a
cluster
right
and
a
lot
of
ways
you
can
think
of.
It
is
hey
we're
replacing
the
Linux
CPU
scheduler
and
we're
just
doing
it
across
a
cluster
of
machines
right
right,
I,.
A
Will
say:
that's
one
thing
that
I
am
at
least
that
I'm
aware
of
right.
There
could
always
be
lots
of
stuff
that
I'm
not
but
I,
don't
know
of
a
lot
of
people
who
are
working
on
kind
of
new,
interesting
schedulers
for
kubernetes.
You
know
taking
into
account
what
came
up
earlier.
Actually,
if
you
know
like
everybody,
starts
their
day
at
9am,
so
we
want
to
do
some
sort
of
predictive
modeling
so
that
we
spin
up
a
bunch
of
engines
before
that,
so
that
we
don't
have
a
Slowdown
during.
B
Logging
people
people
are
waiting
for
some
stuff
to
merge.
You
think
yeah,
there's
been
a
ton
of
work
on
What's
called
the
scheduler
framework.
Okay,
because
previously,
in
order
to
do
your
own
custom
scheduler,
you
were
writing.
You
were
forking
the
main
kubernetes
scheduler
and
and
writing
your
own
in
go
right,
right
and
and
the
scheduler
you
know
a
Sig
scheduling
said
you
know,
people.
A
B
Not
enough
of
this
yeah
that
we
really
want
to
make
it
possible
to
do
this
by
basically
plugging
in
a
bunch
of
algorithms,
the
and
changing
waiting
on
things
and
changing
configuration,
and
that
sort
of
thing
to
to
make
it
more
possible
to
modify
the
scheduler
in
small
ways.
And
hence
this
the
scheduled
framework.
Getting.