►
From YouTube: sigs.k8s.io/kind 2019-10-07
A
A
So
I
just
there's
the
usual
spiel
about
this-
is
a
CNC
F
community
meeting
/
under
the
kubernetes
community
code
of
conduct.
So
basically
don't
get
jerked
anyone
this
meeting
possibly
will
be
recorded
and
uploaded
publicly
to
YouTube
hi,
so
I
don't
think
we
actually
had
any
particular
agenda
items
today.
We've
actually
been
missing
this
one
for
a
little
bit,
the
attendance
dropped
and
we've
mostly
handled
discussing
things
through
the
issue
tracker
and
a
little
bit
in.
A
B
B
B
Yeah,
we're
I
mean
I,
guess
we
have
had
conversations
before
in
the
testing
group,
maybe
through
issues
but
I'm
from
the
kima
project
nice
you
heard
of
that
or
have
we
using
prowl
and
we're
looking
into
using
kind
for
testing
as
well?
Oh
so
I'm
just
trying
to
stay
on
top
of
what's
coming,
what's
going,
can
we
help
with
features
that
we
need,
or
something
and
I
just
try
to
stay?
On
top
of
that
at.
A
The
moment
we've
got
a
few
features
up
in
the
air.
Besides,
just
some
like
you
know,
maintenance
we're
keeping
communities
and
dependencies
and
whatnot.
So,
let's
some
big
ones
we're
almost
done.
Landing
the
coupling
most
of
the
code
base
from
docker,
or
at
least
the
parts
that
involve
cluster
manipulation,
the
image
lock
and
all
over
we're
going
to
handle
that
separately
and
the
goal
there
is
to
make
it
a
little
bit
more
testable
internally
and
to
start
finally
rolling
out
some
limited
support
for
things
like
pod
man
as
alternative.
A
A
It's
really
important
that
you
can
run
it
locally
as
well.
One
of
the
main
benefits
versus
some
of
the
alternative
ways
that
we
test
communities
is
that
you
can
replicate
this
on
your
machine
without
any
cloud
resources.
That
said
for
I,
frequently
get
people
asking
you
know:
does
this
replace
mini
cube
and
not
necessarily
using
VMS?
Has
better
isolation?
You're
not
sharing
a
kernel,
so
you
can
do
interesting
things
like
loading
drivers.
A
You
can
enforce
that
swap,
is
off
and
ensure
that
memory
learners
work,
though
I've
heard
that
you
may
actually
have
swap
enabled
I'm
not
sure
why,
but,
but
like
that
sort
of
you
can
run
into
problems
where
some
some
of
the
more
advanced
features
really
depend
on
good
host
level,
support
and
you're,
not
getting
a
real
host
in
the
container.
So
it's
pretty
good
for,
like
I
need
a
kubernetes
api
I
want
to
test
some
things.
This
is
a
temporary,
throw
a
cluster
yeah,
it's
a
bit
weak
at
the
moment
at
long-lived
clusters.
A
A
Yeah
I'm
not
sure
I,
don't
know
on
that.
One
I,
don't
personally
this
too
much
because
again
that
wasn't
really
the
original
intended
use
case
and
we've
still
got
problems.
We
need
to
tackle
on
the
main
use
case,
but
long
term.
We
do
want
to
try
to
improve
that
sort
of
thing.
That
said,
I,
don't
think
kind
is
doing
anything
special.
There
I.
B
B
A
Yeah
I
suspect
as
well.
I
would
guess
there
are
things
that
are
trying
to
watch
at
4:00
and
keep
updated
and
I
would
bet
that
most
of
that
is
somewhat
time-based
and
that.
A
B
B
A
A
For
now,
because,
like
it's,
it's
pretty
reliable
and
it
doesn't
depend
Network,
and
if
you
need
to
do
things
like
you
know,
authenticated
access,
you
have
to
worry
about
getting
them
and
that's
a
cluster.
You
just
pull
it
to
your
host,
which
is
like
you
already
understand.
It's
a
user
like
I
need
to
authenticate
my
daugher
there's
good
UX
for
that,
whereas
it
tends
me
especially
for
new
users,
it
has
to
be
trickier
to
get
them
to
understand
things
like
you
know,
using
imageable
secrets
or
something
like
that.
A
I
mean
so
I've
been
working
more
and
collecting
bread
like
profound
kind
itself,
but
as
far
as
like
communities
and
things,
the
standard
tends
to
actually
be
the
logs
for
the
components.
So
we
have
a
little
bit
of
tooling
for
collecting
those,
but
most
of
the
kubernetes
components
themselves
are
not
instrumented
in
this
way
and
that's
been
the
primary
focus
you
know
going
forward.
We
have
some.
A
You
know
we
have
some
other
users,
for
example,
from
where
I
work.
Seo
is
a
is
a
big
user
that
I
try
to
you
know
give
some
of
my
time
to
keeping
up
with
what
they're
using
they're,
you
believe
using
it
for
all
their
priests
limits,
and
so
you
know,
they're
testing
an
application
on
top
of
crudités
instead
of
congrat
itself,
but
I
think
this
sort
of,
like
instrumentation
thing
is
usually
done.
B
A
A
Other
things
that
see
we're
working
a
little
bit
on
the
UX.
You
know
the
export
team.
Config
thing
has
been
a
real
tripping
point.
It
is
different
from
other
clusters.
I
think
we're
going
we're
headed
towards
the
writing
to
the
default
one
by
default.
Unless
you
specify
Thank,
You
config
back,
you
want
and
contact
merging
and
that
sort
of
thing
it
will
take
a
bit
of
work,
because
key
betim
doesn't
actually
handle
this
very
well,
but
we
can
rewrite
it.
We
can.
A
We
can
take
the
output
from
Cuba
and
rewrite
it
into
a
format
that
will
work
and
it's
actually
not
terribly
involved.
It's
mostly
been
sub
decision-making
around
that
and
I.
Think
that's
why
you'll
find
with
most
of
China
in
general
is
that
the
easy
problems
are
solved
and
the
really
hard
problems
are
usually
really
hard
things
that
are
not
actually
like
an
issue.
A
What
kind
their
upstream
like
when
we're
dealing
with
now
that
you
haven't
had
they
were
still
investigating
is
that
it
appears
that
docker
exec
as
a
mechanism,
is
slightly
flaky,
especially
I'm
busy
O's.
It's
somewhat
unlikely
that
we're
gonna
go
fix
that
upstream,
but
we
might
like
engineer
some
alternative
to
it.
But
it's
not
going
to
be
like
a
quick
change.
A
Similarly,
you've
got
some
things
like
no
people
would
like
it
to
work
on
Chrome
OS.
That's
like
well
looks
like
there's
some
security
restrictions,
that
kind
of
block
that
or
like
that,
that
sort
of
thing
a
few
other
ones
like
it
would
be
great.
If
we
had
automated
releases
of
communities,
note
images
but
there's
two
problems
there.
One
has
just
infrastructure.
You
know
tossing
this
on.
Prowl
isn't
secure
enough.
A
We
need
to
think
about
this
a
little
bit
more
carefully,
since
people
are
going
to
run
these
as
somewhat
effectively
controlling
their
hosts,
and
just
that
like
we
need,
we
need
to
think
more
about
how
we
communicate
with
users
that,
like
there's,
no
guarantee
that
this
node
image
works
with
this
other
version.
So
far,
we've
solved
that
by
if
we.
A
A
A
That
one
there's
a
number
of
options
to
work
in
the
ecosystem
and
improved
options
for
local
storage.
That's
another
one,
we're
like
mini
cube,
has
some
tricks
that
they
use,
that
are
pretty
straightforward,
but
they
are
straightforward
because
they
only
work
for
one
node,
which
is
somewhat
limiting
for
kind
and
not
an
option
that
we
really
want
to
ship
we'll
want
to
get
it
working
for
multi
node.
It's
pretty
important
to
kubernetes
testing
and
I.
Think
it's
somewhat
important
to
robust
application.
A
Testing
I
think
there
are
classes
of
wanting
to
handle
things
like
rolling
over
deployments
and
things
where
you
you
want
to
know
that
it
behaves,
and
it's
a
little
bit
more
straightforward
to
do
that.
If
you
tell
kubernetes
that
you've
got
it
coming
across
and
nodes
and
you
want
to
cycle
it.
Things
like
stateful
sets.
A
So
it's
probably
the
most
interesting
Oakland
to
look
into
right
now,
but
it
it's
going
to
involve
some
like
ecosystem
work
like
the
the
the
solution
in
kind
is
pretty
much
just
got
to
be
start
deploying
one
of
these
instead
of
using
one
of
the
built-in
parishioners,
but
the
only
reason
we're
not
doing
that
yet.
Is
that
the
things
to
start
applying
it
work.
A
A
A
lot
of
people
that
are
you
know,
major
Linux
power
users
will
do
something
like
run
their
system
with
CFS
and
tweak
docker
to
run
with
his
DFS
and
that
works
okay
for
docker,
but
it
doesn't
even
without
kind.
It
doesn't
really
work
or
kubernetes,
adding
like
nested
container
runtimes.
That
definitely
doesn't
work.
So
we
tend
to
see
that
with
all
of
the
like
non
default
file
system,
drivers,
most
of
them
don't
work
for
this.
Okay,
probably
the
biggest
tripping
point,
and
it's
again
like
there's
not
any
there's,
not
really
anything.
A
B
A
Certainly
intended
for
usage,
but
I
will
say
there
are
definitely
classes
of
testing.
Where
you
know,
I
would
tell
people,
you
know,
use
mini
keyboard
or
looking
your
options,
maybe
Mike
rotates
or
something.
The
thing
I
like
about
kind
is
that
it
works
so
many
places
easily
like
when
it
does
break
down.
It
tends
to
be
painful,
but
that
is
usually
some
kind
of
esoteric
host,
like
I
said
like
docker
running
on
ZFS,
which
is
generally
problematic,
yeah.
B
A
Kind
or
better
of
us,
but
there
are
cases
where
you
need
that
VM
level,
isolation
which
we
can
offer
at
the
moment
we
might
in
the
future,
was
one
of
the
container
runtime
options
that
is
actually
backed
by
Tavian
or
similar,
but
it'll
probably
be
less
portable
and
maybe
not
quite
as
well
supported
at
least
initially
and
the
things
like
GPUs.
A
lot
of
applications
don't
need
this.
But
if
you
do
need
this,
it
is
possible
to
sort
of
feel
a
local.
A
A
To
just
make
sure
that
you
know,
maybe
even
like
performance
testing
or
something
to
make
sure
that
things
behave,
Indians
actually
expect
to
be
deployed
to
because
there
are
going
to
be
some
level
of
difference,
I
mean
even
just
communities
itself
has
so
many
knobs
can
but
I
think
it's
wise
to
separate
out
testing.
Okay,
we've
tested
that,
like
you
know,
our
controller
behaves
and
can
do
interactions
between
controllers
and
and
customer
resource
instances
and
whatnot,
but
then
there's
also
be
like
you
know:
does
it
behave
fully
as
expected
in
the
prod
environment?
A
A
So
so,
given
that
I
think
it
leaves
room
for
like
kind
of
use
like
the
whole
spectrum
of
there's
a
lot
of
options
for
deploying
kubernetes
and
just
sort
of
finding,
you
know
what
works
for
what
you're
trying
to
test
and
where
you're
trying
to
test
it
and
how
quickly
and
whatnot
and
I
think
kind
is
a
pretty
good
option
for
quick
local
testing.
They
run
pretty
much
wherever
you
can
run
docker,
but
it's
certainly
not
perfect,
and
some
of
the
flaws
are
not
they're.
Not
trivially.
Solvable
like
like
that.
A
You
can't
necessarily
guarantee
that
memory
limits
will
work.
It
depends
on
the
host
and
some
tests.
Some
some
classes
of
testing
really
want
to
check
that
sort
of
thing.
You
know
what
happens
when
items
or
something,
and
it
is
possible
to
write
testing
like
that,
but
there's
only
so
much
kind
can
do
to
make
sure
that
that
will
actually
happen.
If
your
host
has
swap
enabled
that
there's
not
really
anything,
we
can
do
yeah.
A
We'll
see
I,
like
said,
I,
think
a
lot
of
the
discussions
that
moved
to
like.
If
we
want
to
discuss
some
particular
figure,
we're
not
waiting
every
two
weeks
and
we
haven't
had
you
know
big
high-level
stuff
for
a
while,
because
we're
kind
of
behind
tackling
all
the
things
we
already
planned
out
like,
for
example,
we've.
A
A
B
A
Like
we're
like
were
most
of
the
way
to
the
doctor,
decouple
it's
not
early
there,
and
that
leaves
a
few
things
a
little
awkward.
It
also
did
some
pretty
major
rework
of
blogging,
but
it
it's
not
fully
complete,
so
I
think
the
biggest
one
is
I,
probably
want
to
land
the
keep
config
and
make
changes
and
take
a
stab
at
that
this
weekend,
I
just
assume
and
all
of
the
command
line
it
like
changes
in
a
release
and
say:
okay,
we're
moving
closer
to
beta.
B
Makes
sense
all
right,
I,
gotta
jump
and
leave
you
to
it.