►
From YouTube: sigs.k8s.io/kind 2019-07-15
A
B
Oh
hi
well
I'm
looking
to
contribute
so
I'm
sort
of,
like
you
know,
Jen
different
meeting
a
second
get
an
idea
of
what's
different
cigs
are
doing
and
then
see
where
I
can
start
from.
You
know
not
to
on
board
myself.
To
could
be
understanding
with
workflow
is
like
what
different
groups
are
there.
You
know
what
I
need
to
do
to
level
up
and
so
on.
Yeah.
A
A
testing
in
general
still
does
a
lot
of
discussion
in
the
meetings,
but
only
four
in
the
cig
testing
meeting,
but
only
for
like
some
of
the
projects
like
this
one's
kind
of
split
off
enough,
a
really
various
byte
cygnus
cig,
it
kind
of
comes
down
to
what
the
you
know.
What
the
people
it
that
group
find
useful.
A
B
A
A
So
this
is
the
bi-weekly
meeting
for
that
sub
project
which
at
this
point
is
mostly
redundant
to
discussing
other
places.
So
we
I
think
we're
reaching
the
point
buried
by
a
solution
like
it's
already
we're
already
like
by
the
time
we
get
to
the
meeting
we've
already
like
discussed
everything
so
nah.
A
And
it's
also,
it
was
really
hard.
We
have
people
kind
of
very
global
involved
in
this
project,
so
we
tried
pretty
hard
to
find
a
time
that,
like
technically,
everyone
can
attend,
but
it
means
it's
about
time.
For
most
people
is
we
needed,
like
New
Zealand
and
a
couple
of
countries
in
Europe
and
America,
and
a
couple
of
like
okay.
This
time,
Monday
is
a
time
that
people
can
attend
that
there
isn't
a
cig
meeting
somewhere
else.
They
need
to
go
what
it
means.
B
A
A
That's
been
a
real
mixture
of
crew
days
to
a
couple
of
groups,
have
like
two
meetings,
one
in
like
an
eight-pack
time
zone
and
one
in
like
more
americas
time
zone
and
some
projects
like
this
they're,
just
like
yeah,
10
or
11
ish.
On
a
weekday,
we
can
probably
technically
swing
everywhere.
They're
10,
11,
ish
Pacific,
but
it's
I
know
for
a
lot
of
people.
It's
a
pretty
terrible
time
but
like
if
you
try
to
pick
a
better
one.
It's
like
oh
well,
there's
some
other
meeting
at
this
time.
A
It's
being
used
for
some
of
the
testing,
so
we
also
spin
up
like
actual
cloud
clusters,
but
at
some
point
all
the
cloud
providers
are
moving
out
and
also
like,
for
you
know
some
random
contributor.
It's
not
necessarily
reasonable
to
say,
like
oh
yeah,
use
this
like
a
key
tool
that
makes
a
Google
Cloud
closure
and
oh
by
the
way
you're
gonna
need
to
pay
for
that.
So
it's
like
this.
A
A
But
you
can
test
a
lot
of
the
core
things
like
creating
a
pod
and
deleting
a
pod
and
using
cube
cuddle,
and
so
we
can
run
a
lot
of
the
tests
and
we're
working
on
being
able
to
run
more
so
like
it
passes,
conformance
rate
out,
for
example,
but
it's
not
like
everything
some
people
use
it
for
like
looking
development
kind
of
like
mini
cube.
Oh,
it
is
no
reuse
cases,
but
originally
it
was
like
we
wanted
something
that
we
could
run
in
our
kubernetes
based
CI.
A
That's
what
this
meetings
supposed
to
be
for,
but
I
think
as
of
about
the
last
meeting
or
two
we're
kind
of
realizing
that,
like
you,
might
just
want
to
spend
this
one
down
and
we
have
the
slack
channel
and
stuff
yeah.
But
if
you're
interested
in
reading
to
kubernetes
in
general
I
would
definitely
say
it
in
some
state
meetings.
Maybe
the
community
meeting
might
get
a
feel
for
where
those
groups
are
headed
and
see.
If
there's
any
efforts
that
they
have
interests,
you
they
also
it's
kind
of
a
lot
just
a
lot
of
sprawl
there's.
A
C
A
Fine
yeah,
so
you
really
fine
Cuban
emissions
is
often
in
the
main
kubernetes
repo.
It's
kubernetes,
/q
batum
that
repo
they
like
track
their
issues.
They're
like
reach
group.
If
you
like
to
attend
one
of
their
meetings
or
something
you'll,
usually
catch
them.
Talking
about
that
and
then
then
you
can,
you
know,
check
their
well.
There
are
other
options
is
if
you
find
something
try
to
find
the
slack
channel
related
to
it.
I'm
just
like
send
a
note,
be
like
hey,
yeah,
I'm
interested.
Would
it
you
guys
have
for
me.
A
A
Is
that
the
the
component
work
group
stuff
yeah
yeah
lay
is
really
nice.
I
haven't
seen
how
that
particular
onboarding
goes
yet
I.
Think
that's
new,
but
I
really
like
that
idea.
I
yeah
I
would
definitely
recommend
ran
out
there.
There
that's
a
stealthy
boxes
of
good
good
people
and
that's
a
that's
a
that's
an
understaffed
project.
I
would
say
like
it's
a
very
important
thing
to
do
from,
like
the
future
of
like
everything
should
have
component
config,
but
it's
one
of
those
things
it's
hard
to
get
people
excited
about.
A
B
A
That'd,
be
great,
I.
Think
right
now
we're
about
to
do
some
big
kind
of
overhauls.
Will
we
get
a
big
one?
There's
just
you
know,
doctor
many
things
better.
We
get
people
asking
questions
all
the
time
lots
of
new
users
and
we
often
find
out
that,
like
people
know
the
answer,
but
you
can't
find
it
anywhere.
A
A
Like
it's,
it's
not
or
like
I,
don't
know,
maybe
there's
some
way
of
using
it
even
just
figuring
out
how
to
organize
all
of
it
like
we're
realizing.
We
had
this
quick
start,
but
it
kind
of
covers
everything
under
the
Sun
and
people
just
keep
adding
stuff
like
okay,
but
this
isn't
gonna
be
quick
anymore,
like
break
it
up.
Somehow
you
know
a
lot
of
projects,
that's
an
easy
way
to
get
started.
They
tend
to
overlook
Docs
and
they're.
A
Usually
not
too
hard
to
you
know,
contribute
some
improvements
to
and
then
that
also
kind
of
gets
you
to
like
interview
I
think
that's
a
good
way.
Another
way,
it's
like
just
looking
for
some
like
small
bug,
because
it's
like
you
know
with
the
bigger
changes.
Then
they
want,
like
you,
know,
discuss
how
they're
gonna
do
it
and
stuff,
and
that
takes
time,
but
like
those
smaller
ones,
usually
it's
kind
of
more
obvious
and,
like
you
can
just
fix
it
like
sup
it.
You
started
yeah
yeah,.
A
Been
in
kind
of
a
bad
place
like
yeah
I,
keep
trying
to
open
issues
that
are
like
look.
This
is
a
nice
small
bug
get
started
and
they
just
get
like
pounced
on
recently
by
it's
like
okay,
but
you've
already
resents
interferes
like
leave
some
for
new
people,
so
I
haven't
figured
out
how
to
solve
that.
Yeah
I've
been
discussing
I
think
that's
a
problem
throughout
the
project.
We
have
some
people
that
are
like
they're,
not
super
experienced
yet
they're,
also
not
new.
A
A
A
A
C
B
Awesome
to
using
geeky
than
it
appears
like
it's,
it's
like
fighting
a
war
trying
to
string
up
the
clothes
line
it
up
beautiful.
What
you
see
is
just
different,
like
you
know,
just
described,
would
like
your
normal
Yama
file.
You
know,
yeah
kakazu
is
just
enough.
Whatever
is
you
need
to,
of
course,
in
to
have
these
that
I'm?
Just
it's
a
level.
A
A
Yeah
well,
definitely
go
snowboarding,
and
you
know,
if
you
do
anything
there.
I
would
also
like
if
you're,
just
looking
for
a
stuff,
even
just
like
stealthy
box,
is
a
good
person
to
just
reach
out
to
they're
pretty
plugged
into
like
cluster
lifecycle,
and
they
will
probably
have
a
bunch
of
other
projects
too.
That
they're
working
with
I
think
they
also
work
full
time
from
somewhere
else
for
getting
going.
A
You'll
see
more
like
Jordan
Liggett
is
someone
that
you
will
just
see
pop
up
everywhere
over
time,
their
their
number
youth
suspects
like
there's.
Also
some
people
that
stick
like
they
say
gang
super
involved
like
the
people
in
the
storage.
Sig
are
like
really
like.
They
mostly
do
storage,
but
like
they're,
really
involved
there
and
they
run
that
sig
really
nicely
like
that's
a
good
one.
A
If
you're
looking
for
stuff
to
start,
they
actually
gotten
awarded
one
of
the
last
cue
cards
for
this
like
they,
they
run
a
tight
ship
and
they
have
like
you
know
they
have
that
issue
thrashing
and
they
have
like
contributor
on
board
and
I
think
they
do.
Like
example,
code
review
and
stuff
like
I
mean
I
would
definitely
go
for
whatever
you're
interested
in
but
like
there's
a
fallback
here,
just
six
storage
that
are
like
very
big.
There
do
a
very
good
job,
making
it
like
accessible
to
get
involved.
A
B
I
mean
I
was
I'm,
am
still
interested
communities,
but
then,
when
I
go
into
the
slack
Channel
and
I
was
like
Casey
and
then
I
saw
the
roopu
the
committee
of
on
a
like
kid,
so
many
groups
so
and
then
literally,
if
I
becoming
clear
to
me
that
okay,
you
know
this
special
interest
group
sort
of
makes
it
easy
for
you
to
you
know,
get
things
done
faster
supposed
to.
You
know
attacking
everything,
the
Mundell
afraid.
A
Way,
cuz
like
if
we
had
like
a
single
group
that
made
decisions
about
literally
everything,
we'd,
never
get
anything
done.
So
a
bunch
of
things
are
delegated
like
how
the
testing
and
infrastructure
is
managed,
is
sig
testing,
like
kind
of
the
framework
for
testing
and
like
the
like,
see,
IM
friend,
stuff
or
if
you're
talking
about
like
the
api,
is
that
the
api
machinery
groups
you
kinda
finds
like
this
is
how
you
do
a
p.
I
is,
and
this
is
how
you
do
the
tooling
for
api.
A
I
think
it's
a
good
pattern
for
bigger
projects,
but
it
does
seem
super
overwhelming
get
started
in
there's
like
just
just
like
even
trying
to
I,
don't
know
if
I
can
even
name
all
the
things
they're
a
few
that
are
like
not
super
active
but
like
still
exist
as
well,
so
yeah,
it's
a
lot
to
keep
up
with
I
would
definitely
recommend
kind
of
picking
an
area
and
focusing
more
on
that.
Just
because,
like
you
know,
subscribing
to
everything
going
on
in
the
whole
project,
I'm
gonna
be
overwhelming.
It's
just
overwhelming
yeah.
B
A
There's
also
the
contributor
experience
group
infants
and
some
respects
supposed
to
deal
with
like
wrangling
all
this,
so
hi
James,
just
China
you
just
hear
about
like
marine
kubernetes
I,
think
this
meeting
is
kind
of
like
then,
mostly
it's
a
terrible
but
technically
possible
time
for
almost
everyone,
and
we
discuss
almost
everything
else
also.
So,
thankfully,
the
attendance
is
about
zero
yeah,
oh
yeah,.
C
A
A
It's
got
some
rough
spots
like
you
have
to
be
root
to
invoke
it
right
now,
because
he
uses
mouth,
but
also
it's
missing,
like
exec
I
might
go
in
from
that
by.
Basically,
you,
you
do
have
to
pick
like
a
VM
size
and
stuff,
but
it
imports
kernel
using
a
docker
image
and
they
imports
new
docker
image,
and
if
your
docker
image
is
based
around
running
in
an
it
system
like
system
D
it
just
that
just
worries
and
yeah,
they
start
pretty
fast,
I
think
not
as
fast
as
they're
advertising.
A
C
A
Something
like
that,
and
it
looks
like
it
even
does
some
stuff
with
the
networking,
I'm
police
hood,
but
anyhow
doing
that
you,
you
can
just
kind
of
take
your
image
and
run
it
and
I
time,
starting
there
bone
tree
image
at
something
like
eight
seconds
once
you
have
it
pulled.
So
that's
yeah,
and
this
is
not
on
the
beefiest
machine.
A
This
is
like
some
spare
box
at
home,
so
I,
it's
still
like
a
reasonable
speed
and
like
we
don't
need
to
make
like
some
weird
VM
image
format
or
anything
like
that
and
most
of
the
command
line.
Kinda
looks
like
docker
ish,
so
thank
you.
I
think
the
biggest
thing
is
gonna
be
awkward.
Is
we're
gonna
have
some
node
fields
that
don't
work
between
the
two
like
you
can't
have
extra
mounts.
There
are
no
mounts:
firecrackers
security
design,
the
only
kind
of
storage
you
can
have
it
all.
A
C
A
So,
like
I,
don't
think
anybody's
going
to
support
that
I.
Think
that's
probably
okay,
but
it
does
mean.
We
already
have
a
feel
that,
like
just
won't
work,
I
think
we
can
just
try
to
document
that
and
have
it
validated.
The
other
thing
will
be
you'll
need
a
way
to
specify
the
resources
for
the
node,
which
won't
make
sense
for
docker
like
the
disk
size,
but
we
can
figure
that
out.
That
doesn't
seem
like
a
you
know,
killer
problem.
A
A
A
A
Let's
see
so
some
test
blindly
just
decide
to
ssh
to
nodes
without
checking,
if
that's
possible,
so
that
doesn't
work.
There
is
some
tests
that
check
if
they
can
ssh
and
the
check
is
basically
just
is
the
provider
in
a
whitelist
so
like
is
it
GCE
and
by
that
I
mean
the
flag?
You
pass
the
test
code,
what
changed
as
a
test
behavior?
So
that's
never
going
to
work
because
yeah
we're
not
so
when
I
had
to
change
that.
A
There's
some
stuff
that
uses
a
host,
exec
pod
and
said
with
like
NS,
enter
and
privilege
and
stuff,
but
then
they're,
invoking
sudo
on
the
host.
For
some
reason,
I
want
to
understand
why
they're
doing
that
and
then
possibly
remove
it
or
make
it
optional
instead
of
installing
sudo
in
kind,
because,
like
you're
you're
I
mean
you're
becoming
rude,
why
do
you
need
to
use
like
you
are
rude?
A
What
are
you
using
sudo
for
so
there's
that
oh
there's,
some
hard-coded
code
that
says
control
plane,
nodes
which
we
want
to
collect
metrics
from
them,
or
something
like
that?
They
will
assume
that
the
way
you
detect,
that
is,
it
does
the
node
name
in
master
and
there's
like
two
or
three
different
implementations
of
that
check
that
are
varying
levels
of
wrong
and,
of
course,
our
nodes.
Don't
so
it
doesn't
think
the
master
is
registered
because
we
don't
call
it
that,
because
no
one's
supposed
to
be
using
that
name
anymore.
A
A
A
If
we
can
fix
that
without
like
ignite
or
whatever
I
just
have
to
not
run
pond
security
policy
tests,
but
I
think
other
than
that,
if
it's
not
a
lake
GC
specific
test
or
like
a
load
balancer
or
something
or
a
balancer
ingress,
I
think
we
can
actually
get
the
rest
of
the
pre
submit
test
from
kubernetes
running.
We're
just
going
to
need
to
do
something
about
all
that
SSH
access.
A
No
I
mean
so
when
you
create
the
pod,
you
need
to
like
enforce
like
a
farmer,
something
like
that
or
like
SELinux,
and
that
gets
weird
and
also
dependent
on
the
host
and
which
again
would
not
be
a
problem
with
like.
We
could
definitely
work
around
that
and
make
sure
it
works
great
with
like
any
night
but
with
actual
containers.
I'm,
not
sure
if
we
can
do
like
app
bar
/
enforcement
and
I
like
third
level
down
container
or
whatever.
A
We
can
either
see
if
we
can
get
everything
to
use
host
exec
pods
instead
or
we
need
to
somehow
give
it
access.
We
could
just
install
us
and
say
sure,
is,
but
then
the
problem
is
like
networking.
It's
assuming
you
know
port
22
on
every
node.
So
on
Mac
we're
not
it's
that's
not
gonna
work.
We
can't
expose
the
same
port
on
every
note.
What
we
could
do
instead,
maybe
is
I,
was
thinking
so
it
does
shell
out
to
SSH.
A
We
can
have
a
mode
in
the
kind
binary
where,
if
dollar
zero
is
SSH,
it
mimics
the
SSH
CLI,
for
example,
King
commands
but
uses
like
docker
exec
under
the
hood,
and
then
we
can
have
like
cube
test
or
whatever
do
put
a
symlink
two
kind
called
SSH
in
path
before
it
invokes
the
test
and
then
when
they
try
to
SSH
they'll,
actually
just
like
dr.
exec
yeah,
it's
a
little
hacky
but
like
that
will
actually
work
with
like
I'll.
Look
like
whatever
random
no
thing.
We
want
right
because
kind
of
pretty
fundamentally
exact.
A
A
So
I
think
it's
actually
gonna
do
like
Association.
Unless
we're
super
invasive
in
the
code
I
mean
we
could
also
tell
it
to
like
run
some
command
instead,
but
we
probably
wanted
to
be
a
sub
command.
So
it's
just
thinking.
Instead
of
making
it
a
sub
commander
than
doing
some
awkward
thing
to
allow
you
to
specify
like
multiple
art,
V
components
for
the
e2b
test,
we
could
just
like
just
set
up
a
kind
binary
that
does
that,
because
one
way
or
another
you're
gonna
have
to
pass
something
to
the
test
code.
A
A
A
This
is
this
is
good,
but
they
kidding
all
the
e2b
tests.
Working
great,
it's
gonna
be
fun.
We're
gonna,
we're
gonna
have
to
do
some
version
of
that
SSA
track
one
way
or
another.
That's
there's
a
lot
of
code
that
wants
to
go
and
ensure
that
things
look
right
from
the
host
site
and
it's
kind
of
hard
I
realize
like
some
of
them
actually
kind
of
can't.
A
One
of
them
is
mount
propagation
right,
so
you
can
have
mounts
propagate
to
the
host
or
and
the
note
and
the
pod
or
you
can
have
them
only
propagate
down,
or
vice
versa.
So
there's
some
code
that
needs
to
like
go
check
from
the
host
like.
Can
it
see
these
mounts
or
not
and
like
kind
of
fundamentally
the
behavior
is
defined
by
if
the
host
can
see
them
so
I,
don't
even
know
if
you
could
same
Lee,
pull
that
off
with
a
host
exact
point.
I
think
you
kind
of
just
need
SSH
or
equivalent.
So.
A
A
Only
thing
I,
don't
like
is
that
for
basically
all
of
them
I
can
think
of
you
do
need
to
do
something
when
running
the
tests,
like
we're.
Probably
gonna
need
to
provide
this,
for
you
like
whether
it's
cube
test
or
whatever
you're
gonna
need
something
that
does
some
setup
magic
I.
Would
it
would
almost
be
nice
if
the
e2b
tests
did
it
itself,
but
I?
A
Don't
really
like
the
idea
of
the
eme
test
code
being
kind
to
where,
like
we
can
add
kind
as
a
provider,
and
we
could
just
say
ya-
know:
ssh
is
now
actually
like
doctor
exec,
but
then
hey
that
makes
the
test
code
awkwardly
aware
of
kind,
existing
and
B.
It
means
that
like
say,
we
went
coded
it
there
whenever
we
add
ignite
or
something
it's
like.
Oh,
not
docker,
exec
a
night
executor
whatever.
A
C
A
Only
thing
is:
we've
been
rejecting
adding
providers
there's
like
like
you,
should
not
be
baking
in
providers.
It
should
just
be
the
like
skeleton
provider.
That
does
nothing,
and
you
should
just
either
have
access
to
these
things
or
not.
But
one
of
the
things
is:
does
your
node
have
like
SSH
keys
and
connects
the
sage
so
yeah,
because
we're
basically
everything
else?
It
actually
genuinely
doesn't
need
to
be
aware
or,
like
conformance
testing,
which
can't
do
as
a
sage.
Wait.
Don't
do
that
I,
don't
know.
A
We
should
probably
discuss
with
someone
the
provider
out
but
I've,
even
kind
of
suggested
against
people
adding
more
providers
to
like
hack
ow,
like
logging
happens
or
something
it's
like.
We
should
just
fix
that,
make
it
an
option
and
like
we
should
not
keep
like.
Oh
now,
we
need
like
eks
provider
or
whatever,
like
that,
I
think
long-term.
It's
not
super
sustainable
for
kubernetes
to
have
like
every
provider
in
the
UE
code.
A
Well,
I'm,
not
sure
where
that
link,
though
that
might
be
an
option
but
I
feel
like
even
then
we're
basically
gonna
wind
up
implementing
in
order
for
their
code.
To
be
saying,
it's
gonna
have
to
implement
something
that
more
or
less
looks
like
this
H
come
in
anyhow,
so
it
might
be
simplest
to
just
you
know,
drop
a
fake
ssh
and
pass
when
we
run
the
tests.
A
The
biggest
downside
to
all
of
these
options
is
sauna.
Gui
can't
I
have
not
founded
a
workable
one
for
both
sauna
GUI
and
not
sauna.
Buoy
sonna
booyah
is
mostly
used
for
doing
conformance
testing,
and
so
none
of
this
applies
you're
not
allowed
to
have
like
node
level,
exec
or
privilege.
I
think
so,
like
that's,
not
gonna
happen
anyhow
but
like.
A
If
we
have
this
like
kind
exact
thing,
the
pod
won't
be
able
to
run
that
and
we
could
like,
maybe
actually
run
necessay
servers
and
set
up
keys
and
all
this
stuff,
but
we
wouldn't
be
able
to
expose
them
in
a
way
that
worked
from
outside
on
some
systems.
I'm,
not
sure
if
we
can
do
it
at
all.
Let's
say
like
also
like
ignite
or
whatever
so
I
think.
A
The
kind
exec
is
probably
the
same
as
from
a
maintenance
perspective,
whether
or
not
it's
kind
exec
and
the
provider
knows
it
or
it's
just
fake
SSH.
Those
are
both
pretty
maintainable,
but
they
also
don't
work
from
sound
way.
They
don't
like
I,
feel
like
it's
more
important
to
get
some
CI
up
and
running
with
these
tests.
In
some
way,
though,
then,
like
boil
the
ocean
on
that,
we
can
always
come
back
to
do
something
else
later
and,
like
I,
said
the
things
that
need
this
are
banned
from
conformance
anyhow,
because
you
cannot
improve.
A
A
We
could
also
expose
it
as
some
kind
of
kind
exact
command
if
we
want
for
Indy
users,
but
like
same
thing
like
you
really
shouldn't
need
to
do
that.
You
should
just
use
kubernetes
for
your
app
or
whatever
it's
pretty
much
specifically
kubernetes
eating
me
testing.
That's
like
how
do
I
make
sure
kubernetes
itself
is
working,
I
need
to
go,
poke
it
and
to
poke
it.
You
need
to
get
out
of
the
pod
and
on
to
the
hosts
and
stuff.
So
the
one
remaining
one
is.
A
A
So
storage
is
the
other
one.
We
need
dynamic,
TVs,
probably
something
like
rancher
localhost
past,
but
a
little
bit
more
robust
and
multi
arch
I,
don't
know
if
we
want
literally
that
or
something
else
but
I'm
guessing.
You
know.
Certain
manager
hasn't
needed
that,
but
a
lot
of
other
stuff
kind
of
breaks
down
when
it
can't
make
volume
claims
yeah.
A
A
I'm,
seeing
like
people
are,
try
to
install
home
chart
have
this
for,
like
the
using
kind
for
testing
other
things
purposes
and
for
some
kubernetes
ci
testing
its
kind
of
similar.
It's
like
okay,
make
a
generic
PV
and
then
do
some
stuff.
That's
like
well
right
now,
let's
started
default
storage
classes,
just
the
gates,
IO
host
path,
which
is
really
not
good
enough
for
that.
So.
A
We
might
want
to
solve
ow
and
going
for
it
and
that
may
get
more
confusing
with
other
platforms.
I
think
we
probably
want
to
stick
to
something
that
in
vaguely
winds
up.
Mapping
like
you
know,
approximately
an
empty
door
who's
back
later.
So
probably
the
rancher
thing
is
at
least
the
right
general
direction.
A
Well
like
currently,
we
couldn't
adopt
it
by
default
because
it
its
not
multi
arch
I'm,
not
sure,
if
there's
anything
else,
that
we
want
or
need
there,
but
we
probably
want
to
explore
some
something
for
that,
because
I
think
this
is
one
of
the
one
of
the
remaining
areas
like
what
now
the
head
port,
forwarding
and
stuff.
It's
possible
I
think
to
even
get
like
a
load
balancer
installed,
at
least
on
Linux,
and
that
sort
of
thing,
but
you're
kind
of
still
fairly
out
of
luck
for
storage,
I.
A
Think
that's
one
of
the
few
areas
that
really
just
doesn't
work
that
and
pods
here.
You
policy
anything
that
relies
heavily
on
isolation
is
still
sort
of
problematic,
but
ignite
may
give
us
an
option
that
we
can
offer
for
that
for,
like
it's
still
cheap
and
local
and
lightweight,
but
like
now
you
need
KVM
and
this
pignite
binary
thing
and
it's
a
little
bit
slower,
but
shouldn't
be
drastically
more
heavyweight.
And
if
you
really
need
to
test
one
of
those
things
that
should
work.
A
C
C
A
At
some
point,
we
probably
need
to
figure
that
out,
but
I
think
that's
gonna
be
a
little
bit
further
down
the
line
past
other
clean
up
stuff
like
we
reach
I,
have
made
this
quarter.
We're
gonna
have
unit
tests,
I'm
gonna
decoupled
docker
from
most
of
the
packages,
not
B,
not
because
of
ignite,
but
it
will
also
help
us
do
that
kind
of
thing,
but
because
we
need
to
have
the
ability
to
like
isolate
this
provisioning
logic.
Talks
to
a
fake
note
come:
let's
see
it
I
test
it
that
kind
of
thing
right
now.
A
A
So
my
big
plan
currently
is
like
make
it
unit
testable,
as
that
happens,
try
to
make
like
pod
men
and
ignite
and
stuff
feasible,
slash
imp
with
them
and
then
later
we
will
want
storage
and
otherwise
I'm
trying
to
tackle
like
the
original
mandate
of
testing
kubernetes.
Let's
get
more
of
that
working
right.
A
A
A
C
Yeah
I
mean
I,
think
yeah
I
guess
we're
dependent.
It
depends
on
what
so
what
the
priorities.
If
it's
sticking
with
like
trying
to
kill
the
end-to-end
testing
working,
they
keep
me
at
ease,
then
I
mean
I'm,
not
too
sure
I'm
not
I
have
myself
but
I.
It
sounds
like
it's
more
about
which,
when
they
come.
A
A
A
So
if
we
put
that
behind
it
in
your
face
and
then
we
have
like
the
docker
one
and
the
pod
man
one
and
a
big
night
one,
and
because,
like
pod
men
in
theory,
you
could
just
alias,
but
in
reality
it's
like
slightly
broken
versus
doctor,
like
they're
they're
command-line
tries
to
be
the
same
CLI,
but
it's
not
exact,
so
we're
gonna
wind
up
dealing
with
some.
You
know
little
incompatibilities.
So
while
we're
at
it,
we
can
just
have
other
darker,
like
things.
A
So,
like
we
probably
wouldn't
add
like
a
GCE
driver
but
kind
of
similar
to
like
the
mini
team
has
vm
drivers,
we
can
kind
of
have
I,
don't
know
node
providers
or
something
I'm
not
really
sure
what
to
call
it
yet.
But
that
name
seems
okay
little
clunky,
but
accurate
and
like
docker
would
definitely
be
default
everywhere
and
not
change
behavior
but
like
there
should
be
some
way
to
opt
in
to
like
that.
A
A
at
all,
then
like
we'll
look
for
a
pod
man
or
something
I,
don't
know
and
ignite
and
then
have
some
kind
of
explicit
way
to
force
an
environment
variable
or
flag
or
something
like
actually
use
this
one
for
kind
to
like
set
which
provider
then
cuz
like
actually
that
one
for
the
foreseeable
future.
If
you
have
ignite,
you
also
have
docker
ya.
After
so,
we
could
have
some
kind
of
opt-in
mechanism
to
switch
to
like
one
of
the
alternate
ones
and
then
like
for
crazy.
A
A
Most
of
that
code
already
doesn't
know
about
docker
directly.
It
uses
this
like
node
object
to
do
like
get
files
from
it
and
write
files
to
it
and
exact
commands,
and
that
sort
of
thing,
so
the
main
blocker
is
actually
that
ignite
doesn't
have
a
good
interface
for
exacting
right
now.
It
has
good
interactive
commands,
but
I'm
gonna,
probably
like
PR
a
lake
ignite
exec,
and
then
we
can
just
almost
one-to-one
map
what
we're
doing
with
docker,
but
we'll
need
some
bit
of
logic
to
like
look
at
the
config.
A
A
Okay
give
them.
These
nose,
run
an
it
over
them
and
join
and
stuff,
and
that
doesn't
need
to
care
about
I
mean
it
could
very
well
be
like
ec2
machines.
We
should
probably
not
do
that.
That's
a
bit
out
of
scope
but
like
if
you're
writing
code
to
stand
up
kubernetes
most
of
that
code
really
should
not
care
even
for
the
stuff.
That's
Glu
incubate
them
together
right.
Just
like
you,
Batum
shouldn't
really
care
what
machine
it's
running
on!
I!
A
Don't
think
most
of
the
kind
code
really
cares
like
the
fact
that
it's
docker
or
not
most
of
that
stuff
happens
just
when
we
stand
up
the
machine
and
then
how
you
talk
to
it
and
we've
even
implemented
most
of
the,
how
you
do
stuff
against
it.
The
helper
functions
are
actually
just
using
exec
like
if
you
read
a
file,
it's
cat,
you
write
a
file,
I
think
it's
also
cat.
A
A
C
It's
I
mean,
if
you
otherwise
only
ever
use
its
for
converting
between
the
others.
That's
the
whole
point
so
yeah
you
can
have
all
of
the
packages
utilize
the
latest
API
version,
but
if
they
do
and
someone
provides
a
older
version,
you
need
some
way
to
convert
to
v3
like
the
new
one
from
reg.
One
couldn't.
C
You
could
but
then
each
time
like,
if
you
do
that,
then
each
time
you
have
a
new
API
version,
you're
having
to
write
like
how
did
you
go
from
one
to
the
other?
If
you
know
if
you
had
like
a
before
you've
updated
things
as
well,
not
before
necessarily,
but
if
someone
provides
an
older
one,
you
might
need
to
then
convert
so
like
you'll
have
to
rewrite
those
conversions,
because
the
ones
from
v1
yeah.
A
I
think
rewriting
conversions
is
an
overestimated
expense,
at
least
for
kind
like
we
can.
Most
of
the
conversion
is
automated
anyhow
and
the
bits
that
are
manual
are
extremely
manual
and
we
should
just
handle
them
with
care
and
just
write
manual.
Conversions
between
the
versions
I
can
see
where
maybe
this
doesn't
scale
for
kubernetes,
but
I
think
for
kind
that
isn't
a
super
strong
and
as
far
as
I
could
tell.
That
seems
to
be.
The
only
argument
right
is
that
you
want
to
avoid
rewriting
conversions.
That's.
A
C
I
think
the
big
difference
for
sake
you
don't
have
to
be
able
to
write
like
you
have
to
be
able
to
serve
any
version
that
they
like
that
the
client
requests,
so
I
can
get
asked
for
a
really
old
API
version,
an
object,
that's
new.
You
know
as
kind
you're
not
really
gonna
get
that
because
kind
isn't
used
to
convert
right.
A
C
A
C
A
A
Well,
all
the
ones
I've
seen
so
far
they
that
the
like
the
internal
type
is
pretty
much
a
mirror
of
the
external
one,
but
they've
stripped
away
the
serialization
tags
and
then
the
conversion
code
is
for
that
one
I'm
not
going
to
tell
them
to
change
all
that,
but
I
think
for
kind.
Probably
the
simple
thing
to
do
is
just
like
outright
like
you
know.
The
create
call
should
just
take
the
latest
API
version
and
we
can
provide
some
helper
for
converting
something
like
that,
because
otherwise
I
can't
see
any
other
way.