►
From YouTube: Panel: Linux in the Kubernetes Era: Does The OS Still Matter? - Tasha Drew, VMware; Kiko Reis
Description
Don’t miss out! Join us at our upcoming event: KubeCon + CloudNativeCon Europe 2021 Virtual from May 4–7, 2021. Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.
Panel: Linux in the Kubernetes Era: Does The OS Still Matter? - Tasha Drew, VMware; Kiko Reis, Canonical; Darren Shepherd, Rancher Labs; Dusty Mabe, Red Hat; & Vincent Batts, Kinvolk
With the end of life of the original “Container Linux” (CoreOS), what is the future for the key underlying component of any Kubernetes deployment: the operating system? While many are opting for general-purpose distributions like Ubuntu or CentOS, there is a healthy ecosystem of Kubernetes- and container-optimized distros, as well as new container-specific kernel features. This panel, bringing together representatives from the key Kubernetes Linux vendors, will review the current state of Linux for cloud native workloads, the areas being worked on by the community, and the pros/cons of different approaches from an end-user perspective. Discussion topics will include improved kernel support for containers, minimal vs general purpose distros, and loose vs tightly coupled approaches to distributing Linux+Kubernetes.
https://sched.co/ekDI
B
Sure
so,
I'm
in
the
middle
of
a
transition.
I
was
at
canonical
for
15
years
and
I've
about
to
to
join
a
new
company
in
brazil.
That's
through
an
acquisition
of
a
side
business
that
I
had.
This
company
is
a
big
retailer
called
magazine,
louisa
or
magalu.
They
have
a
huge
technology
team
and
they're
big
users
of
kubernetes
have
basically
every
single
technology
stack
across
the
company,
so
it'll
be
interesting
to
see
from
the
inside
what
the
user
perspective
is.
A
B
Yes,
I
think
so
yeah,
so
I
I
had
opinions
before
that
already,
but
I
think
my
time
at
canonical
cemented
that
as
well
is
the
os
relevant.
I
think
this
next
big
technology
transition
and
those
are
always
interesting
and
produce
kind
of
kind
of
unexpected
results.
The
the
big
questions,
I
think,
are
going
to
be.
B
What
how
will
the
distro
vendors
react
to
the
change
that
cloud
native
brings
if
platforms
and
distros
basically
have
built
this
value
of
being
a
bridge
between
upstreams
that
release
at
their
own
cadence
that
develop
on
their
own
timelines
and
priorities
to
what
users
expect
in
terms
of
maintainability
of
long-term,
long-term
support
or
predictability
of
release
like
how
does
that
value
translate
into
a
world
where
developers
are
picking
all
the
components
that
go
into
the
into
the
container
itself?
What's
what
about
the
interface
between
kernel
and
the
user
space?
B
Since
you
don't
have
a
kernel
inside
the
container,
you
have
to
use
the
host
container.
So
those
are
the
interesting
questions
a
little
bit
that
are
there.
I
will
and
I
won't
spoil
the
rest
of
the
panel.
I
have
some
advantages
of
the
big
existential
questions
for
people
that
are
building
general
purpose:
linux
in
general,.
C
C
Obviously
we're
getting
more
to
the
point
where
you
know
special
purpose,
os's
or
container
focused
os's
seem
to
you
know,
solve
some
needs
that
platforms
have
on
top,
but
at
the
same
time
I
don't
think
necessarily
general
purpose.
Os's
are
going
anywhere
soon.
I
mean
we
really
leverage
in
this
special
container
os's.
We
we
leverage
the
ecosystem
that
the
general
purpose
os's
provide.
In
our
case,
we
have
fedora,
we
have
rel,
but
there's
rarely
a
new
container
os
that
you
see.
That
starts
completely
from
scratch.
D
D
I'm
also
co-chair
of
the
multi-tenancy
working
group
for
kubernetes
and
co-chair
of
sig
usability
and
as
far
as
the
question
about
operating
systems,
I
think
that
there's
sort
of
like
an
interesting
historical
separation
of
operations,
teams
and
application
development
teams
that
resulted
in
this
drive
towards
having
like
this
golden
image.
That
may
not
have
anything
to
do
with
the
applications
being
deployed
on
it.
D
So
a
lot
of
things
have
gotten
baked
into
the
operating
system,
layer
that
the
applications
deployed
on
it,
don't
actually
need
and
being
able
to
have
more
purpose,
purpose-built
operating
systems
that
serve
the
needs
of
the
applications
running
on.
It
makes
a
lot
of
sense
in
terms
of
just
minimizing
your
footprint
having
fewer
attack,
vectors
and
not
shipping
a
lot
of
stuff
in
that
lower
level
that
you
don't
actually
need,
and
so
I
think,
a
lot
of
it
kind
of
comes
down
to
the
build
the
build
tool
systems
that
we
have.
D
I
still
think
there's
a
ton
of
value
in
the
people
who
are
doing
all
of
the
back
breaking
work
of
maintaining
the
base
operating
system,
components
that
everybody
leverages
and
the
community
that
works
on
maintaining
those
as
well
and
proliferating
them.
So
yeah,
I
kind
of
have
a
a
little
bit
of
a
mixed
bag
opinion.
A
Yeah,
that's
a
split
view
on,
but
it
is
interesting
that
that,
even
you
know,
saying
ops
versus
ops
versus
apps
teams
does
kind
of
give
two
completely
different
like
focal
points
of
the
same
set
of
software
darin.
That
probably
leads
pretty
well
into
how
you
have
operation
teams
versus
enabling
apps
teams.
Why
don't
you
go
next.
E
Yeah
yeah
I'm
darren
shepard
cto
and
co-founder
at
rancher
labs
rancher.
We
do
multi-cluster
management
kubernetes
and
we've
also
done
some
container-optimized
os's
yeah.
So
that's
my
background.
So,
regarding
this
question,
gonna
like
I
could,
I
could
probably
like
you
know,
monologue
for
the
next
half
an
hour.
I'm
gonna
answer
this
question.
E
It
yes,
so
I'm
gonna
keep
it
short,
but
does
it
make
sense
to
run
a
general
purpose
os
in
the
cloud
native
world?
I
would
say
no,
but
it's
actually
it's
significantly
more
nuanced
than
that,
because
you
know
it's
like
what
do
you
call
a
general
purpose
os
and-
and
you
know
it's
like
well,
if
I,
if
I'm
saying
the
answer,
is
no,
does
that
mean
that
the
death
of
a
ubuntu
or
red
hat
or
something?
E
No,
that's
not
true,
because
it's
really
all
of
the
definitions
of
these
things
are
changing
like
the
assets
that
come
from
a
linux,
the
ecosystem
or
vendor
are
all
completely
still
relevant,
but
they're
just
packaged
and
delivered
in
a
completely
different
way
because,
like
I,
don't
even
really
like
the
term
of
container
optimized
linux,
because
the
reality
of
when
you
look
at
these
things
is
they
almost
have
nothing
to
do
with
containers,
except
for
the
fact
that
they
have
docker
or
podman
or
kubernetes
on
it.
It's
like
a
container
optimized.
E
Linux
is
still
a
general
purpose
os
because
it
can
run
basically
in
any
any
workload.
It's
just
how
the
workload
is
packaged
and
delivered,
so
I
think
there's
a
huge
transition
that
we're
going
to
see
as
we
go
into
the
cloud
native
world
of
like
how
users
interact
with
the
linux
distribution,
what
they're
expecting
from
it,
and
I
also
see
a
lot
of
interesting
opportunities
for
kubernetes
and
kubernetes
distributions
to
effectively
replace
what
is
linux
distributions
today
and
sorry,
that's
a
little
off
topic
I'll
just
say
something
real
short
and
I'll.
E
Stop
it's
just
that.
The
the
linux
distribution
today
is
really
a
kernel
and
a
set
of
user
space
right
and
so
they're
a
bunch
of
user
space
packages
that
run
on
the
kernel
right.
So
when
you
look
at
kubernetes,
our
view
of
a
kubernetes
distribution
is
very
naive
right
now.
I
think
it's
very
myopic.
It's
because
we
have
kubernetes,
but
then
we
have
a
whole
host
of
applications
and
software
that
run
on
top
of
it
and,
like
you
see
that
from
the
cncf
landscape,
you
know
this
gigantic
slide
of
all
the
scrap.
E
Good
software,
sorry,
not
crap,
and
so
you
know
how
you
know.
There's
there's
still
the
fundamental
need
that
linux
distribution
is
always
providing,
which
is
like
delivering
packages
that
run
on
the
kernel.
Now
we
kind
of
need
the
same
thing
in
the
cloud
native
space,
so
it's
like,
but
those
those
packages
that
you
know
run
on
top
of
kubernetes.
They
still
have
linux
bits
inside
of
them.
They
still
have
glibc
and
they
still
have.
You
know
java
runtime
they're
all,
but
you
know
so.
This
is
just
all
changing
that
that's
basically
mm-hmm.
A
No,
I
do
think
lots
of
lessons
learned
of
software
packaging
in
general
that
have
been
invented
and
reinvented
in
so
so
many
ways
and
we're
just
going
to
reinvent
them
again
in
the
container
space.
So
good,
good
good.
So
then,
with
that,
I,
like
you,
you
you're
talking
about
you-
know
everybody's
kind
of
mentioned,
like
the
way
that
these
things
are
changing
and
even
to
some
extent,
kind
of
implying
that
obviously
there's
enough
players
in
the
scene
that
are
interested
in
seeing
it
play.
A
You
know,
play
out,
and
I
think
one
of
these
things
that's
interesting
is
that
it's
like
yeah.
I
don't
think
that
any
of
the
general
purposes
are
going.
Os's
are
going
away
and
I
don't
think
they
need
to,
but
in
some
ways
the
pie
is
just
getting
bigger
so
like
what
what
purpose
they
did
is
fine,
but
the
cloud
native
space
is
expanding
out,
and
so,
with
that
you've
seen
you
know,
different
cloud
providers
come
up
with
either
their
own.
A
B
So
I
think
that's
something
which
is
absolutely
a
a
likely
trend
to
come
that
they'll
optimize.
This
deliverable,
which
is
they're
saying
like
how
much
is
that?
Actually
linux,
you
know:
okay,
it's
linux.
From
the
kernel
perspective,
what
else
is
there
like?
There's
not
a
lot
there?
You
know
that
that
just
serves
enough
scaffolding
for
you
to
boot,
the
the
containers
off
there's
one
wild
card
on
that,
though,
which
is
that
as
you
move
into
enterprise,
everyone
says.
B
Oh
hold
on,
but
I
need
to
run
my
asset
tracking
or
I
need
to
run
my
my
ids
pieces
on
this
layer
as
well.
So
I
think
that's
the
only
piece
where
I'm
not
sure
what
the
future
actually
will
look
for
that,
but
that
something
specific
to
be
delivered
for
that
that's
small,
atomically,
updatable,
safe
to
basically
roll
back
and
forward.
I
think
that's
definitely
coming
the
question.
Is
this
thing
like
what
do
you
do
like
when
somebody
says?
B
E
A
E
Like
isis,
isv
certification
is
a
big
deal
too.
You
know
it's
like
a
lot
of
people
are
like
I'd,
have
to
run
this
os,
because
that's
what
you
know
this
vendor
has
certified
on
which
completely
changes.
When
you
know,
as
vendors
are
starting
to
package,
you
know
their
third-party
software
on
top
of
kubernetes.
What
does
it
now
mean
to
certify
you
know
and
that's
kind
of
interesting
space?
I
mean
all
right
question
a.
B
E
Yeah
and
and
well,
and
what
does
it
mean
you
know?
Is
there
even
honestly?
Is
there
even
such
thing,
as
can
I
certify
for
kubernetes
in
general?
You
know,
or
am
I
actually
certifying
for
eks
or
or
you
know
like
vmware
suite
or
or
whatever
openshift.
You
know.
Those
are
interesting
questions
so,
but
going
back
to
because
I
already
forgot
vince
like
vincent.
What
was
that?
What
was
the
original
question?
A
It
was,
it
was
more
like
is:
is
it
a
useful
trend
that
we
see
these
kind
of
optimized
derivatives
because
there's
a
few,
you
know
players
in
the
scene-
and
you
know
like
you've,
even
had
experience
with
having
kind
of
an
optimized
yeah.
That's
right
focus,
but
is
it
a
useful,
useful
trend
or
is
it
well.
E
Know
so
well
I
mean
it's
disruptive
for
sure.
Like
so
there's,
there's
like
the
bottle
rocket
and
like
google's
container
optimized,
you
know
microsoft.
I
they
haven't
announced.
I
don't
think
a
linux,
you
know
container
distro
or
whatever,
but
I
imagine
one
would
come
at
some
point
honestly.
I
think
it
makes
total
sense
for
these
cloud
providers
because
it's
like,
what's
the
touch
point
of
the
customer,
it's
like
the
customer
or
the
end
user.
They
really
want
kubernetes.
E
So
why
do
they
need
to
care
about
the
nodes
and
what's
running
on
the
nodes?
The
interface
is
kubernetes,
and
this
is
why,
like
it's
it's
tricky,
especially
if
you're
a
linux
vendor
of
like
well.
What
does
this
mean
now
because,
effectively
that
linux
layer
is
the
the
kernel
itself
or
the
container
linux?
Whatever?
Is
it's
fairly
commoditized
from
us?
You
know
it's
like
people,
don't
really
want
to
pay
for
it,
that
much
and
and
they,
since
they
don't
interact
directly
with
it.
E
C
A
C
I,
as
far
as
you
know,
different
cloud
providers
coming
up
with
their
own
distros.
Is
that
a
useful
trend?
I
I
think
you
could
go
both
ways
right.
It's
useful
because,
anytime,
you
bring
new
ideas
to
the
market,
there's
things
that
can
be
learned,
but
at
the
same
time
it
also
increases
fragmentation
a
little
bit
like
there's
now
more
options,
and
if
you
happen
to
be
rolling,
you
know
trying
to
set
up
kubernetes
yourself.
C
You
know:
here's
a
new
option
to
consider
which
is
options
are
good,
but
they
also
might
confuse
people
or
lead
them
down
the
wrong
path.
At
some
point,
I'm
not
sure,
but
I
think
darren
had
a
good
point,
which
is
it
depends
on
who
you
are
right.
Are
you
just
clicking
a
button
and
having
the
entire
cluster
set
up
for
you
and
it's
completely
managed?
Not
you
know
you're
not
managing
the
cluster.
C
If
that's
the
case,
then
you
know
the
role
of
the
distro
does
have
a
lot
less
of
a
role,
at
least
that
you're
concerned
with
right.
The
the
os
is
still
very
important.
We
see
that
every
time
there's
a
new
security
issue
that
comes
out
right,
but
who
cares
who's
responsible
for
it
right
is,
is
the
big
part
and
we've
seen
the
the
the
lines
move
more
and
more
towards
you
know
more,
is
being
taken
care
of
by
something
that
you
outsource.
C
You
pay
for
it,
it's
a
cloud
platform
and
you
care
just
about
your
app
right
and
that
service
that
you're,
providing
to
your
customers
or
to
your
users.
That
is
giving
them
value,
and
so
you
know
darren
is
right.
If
you
are
a
cloud
platform
and
you're
offering
that
abstraction
to
your
users
and
your
users
don't
care,
then
it's
probably
a
useful
trend
for
them.
C
E
Yeah,
wonderful
because
the
you
it's
like
it's
kind
of
like
as
we
move
into
this
cloud
native
world,
it's
all
kind
of
reducing
the
burden
on
the
the
operator
or
whatever.
Even
if
you're
running
yourself,
you
know
data
center
or
whatever
you
know
it's.
You
know,
I
think
you
know
core
os
in
the
early
days
and
now
fedora
core
west
really
kind
of
pioneered
this.
E
This
kind
of
what
is
it
a
self-driving
car
kind
of
mentality
of
of
you
know
it
just
you
treat
it
more
like
a
phone
or
whatever
it
just
updates,
and
you
know
reboots
or
whatever
it's
so
it's
all
about
reducing
the
the
maintenance
which
which
a
lot
of
that
also
is
about
re,
reducing
the
variability
in
the
os
so
like.
If,
if
you
say
like
that,
the
trend
is
going
towards
re,
reducing
the
maintenance
and
the
variability.
B
E
A
A
D
Mean
yeah
in
an
ideal
world
they
wouldn't
have
to
but
like,
as
we
were
saying
earlier,
when
you
start
getting
into
what
software,
what
your
software
is
certified
to
run
on
from
a
container
and
host
perspective.
Like
you
from
an
isv
perspective,
you
really
need
to
like
minimize
the
possible
number
of
combinatorial.
D
You
know
combat,
like
the
matrix
like
like.
Let's
make
this
achievable
like
to
actually
say
we
support
our
software
so
yeah.
I
would
say
that
it
becomes
just
a
supportability
problem
for
the
isvs
versus
as
an
application
developer.
Do
I
wanna
care?
I
really
don't
right,
like
I
kind
of
just
wanna
care
about
the
exact
changes,
I'm
making
my
to
my
application
and
have
it
be
as
atomic
as
possible,
but
then
I
always
end
up
relying
on
something.
D
That's
at
a
lower
level
that
ends
up
making
things
complicated
from
the
cloud
vendors
perspectives.
I
think
releasing
their
own
linux
distributions
makes
a
lot
of
sense
just
from
cost
reduction
and
optimizing
for
their
exact
use
case.
They
have
a
more
limited
number
of
hardware
that
they
need
to
support,
so
they
don't
need
like
something
they
could
really
run
on
the
huge
swath
of
things
you
find
in
on-prem
data
centers.
They
don't
want
to
necessarily
pay
a
vendor
every
time
they
spin
up
linux.
D
You
know
they
want
to
keep
all
of
the
money
from
the
customer
instead
of
just
a
smaller
amount,
so
yeah
I
get
it.
I
don't
know
that
it
really
translates
to
on-prem
use
cases
where
people
are
really
consolidating
their
entire
security
posture
around
a
limited
number
of
linux
and
windows
operating
systems.
They
need
to.
You
know
whether
we
agree
with
it
or
not.
They
want
to
run
their
virus
scanner
against
every
single
node.
A
Yeah
gosh
that
security
checklist.
So
that's,
that's,
probably
pretty
perfect
segway.
Are
there
different
kinds
of
security
considerations
that
you'd
have
like
for
linux
in
general,
when
you're
specifically
running
kubernetes
so
like
in
this?
Like
we,
you
know,
we
said
cloud
native
by
and
large,
that's
meaning
like
schedulable
workloads
that
can
be
put
on
a
cluster
somewhere
like
kubernetes.
B
B
That
went
into
that
statically
linked
thing
like
okay,
like
where
do
they
come
from,
and
so
now
you've
got
this
huge
providence
manifestation
problem
that
the
distribution
is
very
neatly
solved
for
you,
because
we
gpg
signed
every
single
binary
that
comes
out
there,
and
so
you
know
this
binary
was
built
by
a
robot
inside
a
system
that,
in
fact
I
I
was
part
of
building
a
canonical.
You
know
that
robot
built
that
binary.
B
Guys
do
whatever
you
want,
you
know,
go
to
get
github
find
this
thing
like
get
close,
get
clone
away
or
even
worse,
like
npm
install
your
way
into
this
mess,
and
now
like
like
that
is
the
actual
security
problem.
That
distributions,
ironically,
are
very
well.
I'm
set
up
to
solve
because
again,
the
dna
of
this
distribution
is
to
serve
as
a
bridge
between
the
madness
which
is
upstream
and
the
predictability
which
end
users
expect
right,
and
I
use
npm
as
an
example,
because
it's
like
the
the
lsd.
A
A
B
The
first
ten
minutes
of
anybody
using
npm
like
if
you're
like
for,
like
an
old
school
person
like
what
the
hell
did.
I
just
do
you
know
like
this
thing,
imported
this
dependency,
which
is
like
10
lines.
What
the
hell
is
that
so
anyway,
but
the
point
is
that
that
madness
has
to
be
resolved
some
way
right.
They're
gonna,
like
somebody,
is
going
to
have
to
give
these
people
that
are
developing
on
the
side,
a
set
of
components
that
they
can
use
safely.
B
You're,
saying
statically
linking
that's
cool,
but
somebody
has
to
be
able
to
press
a
button
in
ci,
cd
and
say:
hey
bring
in
the
new
versions
of
these
things.
Oh,
but
they
can't
be
api
breaking
otherwise
the
app
breaks.
You
know
like
this,
this
sort
of
things
which
distributions
are
very
known
for,
like
we
don't
break
the
applications.
If
I
break
your
app,
I
apologize
file
a
bug.
I'll
revert
back
that
that
breaking
change
in
the
api,
because
I
keep
stability
for,
however
long
I
promise
to
maintain
distribution
like
that
has
to
be
yeah.
B
E
E
Well,
curated,
that's
an
extremely
difficult
thing,
but
what's
different
what's
happening
now
is
like
the
kind
of
traditional
the
way
that
things
were
before
is
you
would
install
this
generic
thing
and
then
you
would
kind
of
mutate
it
there
into
what
it
was
going
to
be.
You
know
that's
the
way,
puppet
chef,
configuration
management,
salt,
all
those
things
work
you
know,
ansible
is
it's
kind
of
you
form
it.
You
know
it's
like
this
clay
that
you
form
into
what
it's
supposed
to
be
once
it
gets
there
and
we're
shifting.
E
All
of
that
earlier
into
the
pipeline,
and
so
it's
now
basically,
I
want
to
take
all
of
those
assets,
those
rpms
or
debian
packages,
and
I
want
to
do
one
of
two
things.
I
want
to
build
an
os
image
which
then
turns
into
an
immutable
container
optimized
linux,
or
I
want
to
build
a
docker
packet
or
docker
docker
image,
which
then
becomes
your
container
runtime.
E
So
it's
like
we're
moving
this
kind
of
mutability
and
assembling
of
all
this
stuff
earlier
in
the
stage
to
produce
these
like
reusable
assets
that
we
can
more
easily
track,
and
you
look
at
the
big
advantages
of
containers
in
general.
A
lot
of
it
has
to
do
with
distribution
management
pipeline.
That's
where
you
get
a
lot
of
the
value
out
of
them,
and
so
it's
like
by
shifting
everything
kind
of
earlier
and
more
predictable.
E
A
E
Into
security
is
that,
like,
I,
don't
think
necessarily
the
security
requirements
of
containers
change
but
security
and
the
models
that
we
have
are
changing
such
that
we
can
make
it
more
secure,
because
you
know
we're
building
like
we're,
we're
creating
immutable
assets
that
we're
deploying.
So
we
can
cryptographically
ensure
that
the
right
thing
is
is
put
there.
So
it's
like
what
I'm
seeing
with
containers
and
and
stuff
and
all
this
kind
of
immutability,
is
that
we
can
greatly
increase
the
level
of
security,
because
we
can,
you
know,
like
secure
boot,
fully
verify
things.
E
A
A
A
Yeah
yeah,
not
only
that
the
recent
xkcd
about
you
know
one
one
person
out
in
nebraska
that
broke
the
internet
or
whatever,
but
no
and-
and
it
is
crazy
because
often
when
people
hear
that
of
like
oh
you
should,
you
should
actually
be
familiar
with
the
lines
of
code
that
you're
importing,
whether
it's
golang
or
npm
like
and
it's
usually
it's
like
the
gut
reaction
is
that
it's
that's
such
an
overwhelming
reaction,
or
you
know,
ask
that
people
just
laugh
it
off
as
a
joke.
It's
like
that
is
the
joke.
A
C
C
Many
different
companies
all
over
the
world
coming
together
and
you
know,
I
think
I
think
some
of
the
cloud
providers
are
getting
to
a
point
where
they
they
realize
that
if
they
they
can't
just
survive
forever
without
heavily
investing
in
this
part
that
they
they
started
to
build
up
so
yeah.
I
I
think
it
it's
starting
to
get
a
little
better,
but
I'm
just
one.
E
Person
in
the
in
the
cloud
providers-
you
know
kind
of
you
know
it's
like
open
source
kind
of
started
with
this
ideal.
You
know
it's
kind
of
like
the
random
person
contributing
and
in
their
free
time,
doing
things
and
there's
still
a
lot
of
that,
of
course,
but
it's
been
tricky
over
the
years
has,
as
open
sources
also
become
the
business
model
of
so
many
companies
is,
is
people
they're
paying
people
to
work
on
open
source?