►
From YouTube: Kubernetes SIG CLI 20170118
Description
Kubernetes SIG CLI bi-weekly meeting for Jan 18th 2017.
A
A
Okay,
for
me,
this
tough
couple
weeks
has
been
all
about
the
general
stabilization
and
reduce
the
amount
of
open
issues,
a
story
so
having
mostly
working
on
bugs
and
a
lot
of
code
reviews
and
bug
triage.
Mostly.
You
know
this
kind
of
clean
up
for
our
backlog
of
ochem
issues
so
doing
that
I'm
also
working
on
a
per
request,
not
yet
open
for
related
to
the
reduce
dependency
of
Cobra
from
commands.
I
know
tony
has
also
been
working
on
on
that,
specifically
about
the
get
delete.
A
So
it
seems
to
me
that
we
are
pretty
advanced
in
terms
of
soup
commands
as
a
plugin,
but
we
still
have
a
lot
of
a
lot
to
discuss
about
supporting
third
party
resources
and
federated
API
researchers,
so
I'm
going
to
get
back
to
to
those
design
of
from
the
extensions
medicine
but
the
past
couple
weeks.
For
me
all
about
stabilization
of
initials
and
I,
think
like
that,
that's
it
for
me.
B
Some
concept
guides
for
coop
control
and
what
I'm
trying
to
do
is
break
up
a
period
of
overview,
guys
that
give
brief
introductions
around
like
general,
more
general
topic
areas
that
to
control,
tries
to
address
and
then
more
in
depth
guides
for
each
one
of
those
topic
areas
I've
seen
the
same
question
almost
every
week,
which
is
like,
what's
the
difference
between
coop
control
apply
and
the
other
command,
and
we
like,
we
do
have
some
documentation
on
it,
but
it's
clearly
not
written
in
a
way.
That's
addressing
that
knee.
B
There's
a
PR
out
to
fix
an
issue
with
apply,
which
is
two
oppositely
milling
fields
which
is
getting
pretty
close
to
submissions.
That's
great!
There's
a
third-party
resource
regression
with
apply
best
coffee
in
huge
fur
box.
This
regression
was
caused
by
fixing
another
issue
which
was
getting
the
wrong
version
of
the
resource
or
getting
an
arbitrary
version
of
a
resource
for
years.
Multiple
versions,
for
so
I
mean
he's
working
on
a
picture.
B
So
anyone
has
been
arresting
that
can
talk
to
me
about
it.
Online
I
mean
there's
two
sort
of
projects
getting
started.
One
is
wildcard
queries
to
do
like
gloss,
so
you
can
take
it
pods
and
then
say
in
your
next
start
to
something
and
that
it
gets
all
the
pods
that
match
that
cloth
like
with
file
and
then
the
other
is
for
all
to
print
out
a
warning.
If
a
service
doesn't
have
any
matching
in
place.
C
I
guess
I'll
introduce
myself
I'm
Joe
if
we
haven't
met
before
hello,
I'm,
the
CTO
of
hefty.
Oh
we're
going
to
be
focusing
some
of
our
efforts
on
is
in
general,
making
covering
these
more
useful,
and
I
think
that
the
experience
around
a
imperative
versus
declare
it
is
and
how
we
present
deaf
users
in
the
four
terminology
and
Dives
that
we
build
around
that,
I
think,
are
going
to
be
important.
C
So
I
was
talking
I
wrote
to
talk
a
little
while
ago
with
some
calmness
and
I'll
share
with
the
group
here
and
I
was
talking
with.
So
a
little
bit
seems
like
this
is
a
good
place
to
start
talking
about
some
of
this
stuff
and
then
on
the
globbing
thing.
I
think
it's
a
bad
idea.
Okay,
we
can
talk
about
that.
I,
don't
want
to
I,
don't
wanna.
Why.
B
C
Is
the
number
by
heart,
so
yeah
so
I'm
not
surprised
to
see
that
there's
a
lot
of
thinking
here
and
so
I
think
a
big
part
of
this
is
you
know:
where
can
I
add
value
versus
resetting
the
computation,
because
I
don't
want
to
I,
don't
want
to
be
bull
in
the
china
shop
and
reset
stuff.
So
there's
something
in
there.
I
think
the
biggest
thing
and
I
think
you
know
Tim
is
is
probably
my
biggest
order.
C
We
want
to
make
this
super
easy
for
a
user
and
come
up
with
some
sort
of
synthetic
scenario
and
write
code
that
will
work
in
that
that
world
and
then
you
sort
of
like
go
down
that
path
and
you
get
to
the
end
and
you're
like
wait
a
second.
This
is
kind
of
crazy
and
I
kind
of
feel
like
we're
there
with
the
three-way,
merge
and
applies
I,
don't
know
what
the
options
are
moving
forward
here.
I
think
you
know
talking
to
Tim.
C
You
know
there's
some
thinking
around.
Ok
as
we
move
to
the
v2
API,
do
we
make
it
multi-layered
by
by
default
with
like
different
owners
and
different
layers
and
have
explicitness
is
sort
of
like
what
over
right
sweat?
What
another
option
is
that
we
use
you
know
annotations
or
some
other
mechanism
to
actually
have
people
leave
bread
crumbs
of
which
programs
or
which
entities
own
which
parts
of
the
resource
using
some
sort
of
mini
language.
C
So
that
way,
the
tooling
can
warn
you
if
you
try
and
do
and
apply
and
the
thing
that
you're
applying
actually
wax
something
that
somebody
else
is
controlling
right,
Jeff,
mom
or
if
you
have
an
auto,
you
know,
if
you
have
somebody
I
mean
skit.
Auto
scaling
is
not
the
best
example,
because
there's
sort
of
that
bird
that
scaled
it,
but
in
some
ways
like
if
you're,
using
the
scale,
verbs
imperative,
lisset
stuff.
C
That
means
that
that
you
know
it's
an
imperative
owner,
vs
vs,
actually
setting
the
stuff
by
by
doing
a
push
right
and
I.
Think
scale
is
one
of
the
best
examples
to
look
at
this
stuff
where
the
three-way
merge
can
really
start
to
fall
apart.
So
I
don't
know
I
don't
have
any
solutions
there,
but
that's
one
of
the
things
that
I'm
looking
at
oh
and
thinking
about
Oh.
C
B
C
B
A
C
Flow
of
like
I
check
something
in
and
then
somebody
maybe
who's,
not
me,
maybe
sometimes
down
the
line
after
somebody
approves
the
dip
or
whatever
says
yes,
and
then
that
thing
gets
applied
and
boom.
It's
like
you
know,
make
it
so
right,
yeah
and
that's
the
model
and
and
there's
another
interesting
thing
around
around.
Can
we
do
a
directory
of
the
mo
files
as
sort
of
sort
of
a
a
mock
API
server
that
we
can
then
use
imperative
commands
against
to
actually
write
those
so
can
I
do
like.
C
Does
that
make
sense
at
all
so,
like
they're
sort
of
this
mirror
of
like?
Can
you
use
imperative
to
actually
modify
the
EML
files
that
are
in
my
git
repo
check,
those
in
and
then
have
apply
just
whack
those
things
out?
Yeah
I
can't
do
a
scale
other
resource
against
your
directory
updates
the
config
exactly.
C
Like
okay
I'm
you,
the
scale
command
can
either
use
against
mine,
my
spec
or
it
can
be
used
against
the
live
server.
It
doesn't
matter,
and
that
means
that
the
sort
of
workflow
and
the
lessons
that
people
use
from
imperative
you
don't
throw
all
that
stuff
away.
What
you
say
is
like:
oh,
you
can
use
your
imperative
command
against
your
local
copy.
You
can
use
that
check
that
in
visit
replicated
across
multiple
clusters,
what
have
you
I'm
going
to
write
that
down?
Actually
that's
in
the
doc,
okay,
I'm
trying
to.
B
Yeah
and
you
know
what
this
I
just
reading
adaptive
and
I
don't
know,
entertain
everything
in
it
and
kind
of
actually
have
to
copy
it.
Just
to
remember
it
is
weird
but
yeah,
so
I
agree,
I
think,
there's
so
there's
a
number
of
things.
I,
don't
think
are
mutually
exclusive,
with
apply
like
which
is
like
the
thing
you
said
about.
B
B
C
C
No
I
want
to
reset
my
face
to
be
the
thing:
that's
really
there
and
I
really
want
to
do
too
right.
So
in
some
ways,
if
we're
going
to
do
a
fly
with
a
three-way
merge,
there
also
has
to
be
a
sink
mechanism.
Where,
essentially,
you
say
you
know,
I
I
wanna,
like
I,
am
now
based
off
of
this
other
thing.
Instead
of
last
thing
and
I
applied
right,
there
there's
a
sink
mechanism.
That's
missing
here,
yeah.
B
I
think
one
thing
we
need
is
a
set
of
tools
to
define
management.
So,
like
the
situation
you
just
described,
the
user
experience
I'd
like
to
see
is
that
every
tool
we
have
a
consistent
view
of
who
owns
which
field
you
know.
If
they
don't,
if
what
they
see
differs
from
what
they
think
it
should
be
in
terms
of
ownership,
then,
by
default,
the
tool
doesn't
make
the
change,
but
there's
a
set
of
commands
that
can
be
used
to
reset.
You
can
feel.
C
C
If
you
want
to
really
like
feel
ownership
of
the
replicas
thing
then
go
ahead
and
do
that
but
you're
going
to
break
your
autoscaler
and
then
you're
autoscaler
is
going
to
start
throwing
warnings
because
it's
like,
oh,
oh,
you
want
to
need
a
scale
and
I'm,
not
the
owner
of
the
thing,
so
I'm
going
to
be
mucking
with
somebody
else
with
stuff
and
so
I'm
not
going
to
do
that
right.
So
an
annotation.
B
C
Yeah
I
guess
I'm
not
arguing
with
them
with
apply
as
the
the
name
of
the
thing
I'm
arguing
with
like.
Can
we
make
the
semantics
more
concrete
more,
you
know
totally
I
know
what
s
and
n
deck
there's
a
trade-off
between
easy
between
implicit
and
easy
and
explicit
and
and
clear,
right
and
I.
Think
we
like
right
now
the
current
apply
model.
It
is
a
little
bit
too
implicit
means
that
you
know
you
do
hit
these
confusing
cases
and
it's
like
it's
not
doing
what
I
want
I
Iran
apply
and
scales,
not
updating.
B
Think
I
think
different
ways
of
approaching
that
problem
like
I'm,
not
sure,
if
is
a
single
bullet
I
think
one
thing
we
can
do
is
add
lenders
and
dip
tools
to
make
it
more
clear.
What's
going
on
right,
so
just
run
like
apply
lid,
and
then
it
tells
you
hey,
here's
all
the
things
that
are
different
that
are
managed
by
other
things.
Here's
all
the
things
we're
not
going
to
update,
but
we
noticed
our
out
of
things
is
not
well.
C
You
know
and
if
you've
used
terraform
before
and
I,
don't
know
if
you
have
I
mean
there's
a
plan
mode
there
right.
So
you
to
terraform
plan
and
it
lays
out,
like
hey,
I'm,
going
to
recreate
this
VM,
because
you
change
this
property
here
and
that's
actually
a
recreating
property,
and
so
there's
a
lot.
I
mean
it's
not
perfect
and
it
can
get
kind
of
or
both
but
like
at
the
end
of
the
day.
B
So
there's
one
other,
so
what
we
discussed
so
far
like
I,
said
compatible
with
applying
additions
and
things
that
can
be
composed
together
and
we
can
decide
the
defaults
later,
but
we
can
at
least
get
the
user
experience.
We
want
by
composition
without
tearing
apart,
apply,
there's
another
approach
on
which
I'm
not
endorsing,
but
we
should
talk
about
anyway,
which
is
like
you
could
get
rid
of
the
three-way
merge
all
together
and
just
have
like
gifts.
B
C
What
you
would
do
is,
you
would
have
a
you
it.
You
would
have
a
layer,
cake
model
right
and
then
you'd
have
different
tools
for
editing
different
layers
of
the
layer,
cakes
right
and
you
would
merge
these
together
after
the
fact,
and
then
there
becomes
the
problem
of
okay.
Who
gets
what
order
right,
because
your
apply,
you
may
think.
Well,
you
know
most
users
are
going
to
set
the
base
layer
and
then
there's
going
to
be
all
these
other
things
on
top
of
it,
but
those
tools
themselves
have
to
actually
figure
out.
C
C
Those
annotations
ends
up
being
a
diff,
you
know
or
partials
that
get
applied
on
top
of
it
using
some
some,
you
know
straightforward,
merge
rules
when
you
call
apply,
if
you
don't
specify
anything,
then
it
isn't
being
you
know
a
layer
called
0
dash
space
or
something
like
that
or
0
use
or
other
things
that
run
on
top
of
that
end
up
editing
disk.
So
maybe
the
imperative
stuff
ends
up
being.
You
know,
you
know,
100
dash,
imperative
or
1
dot,
dot
dash
comparative
right,
and
then
you
could
have
an
admission
controller.
C
That
actually
says
hey.
You
know
what
users
aren't
these
users
aren't
able
to
set
anything
above
you
know
level
50
right
and
now
I'm
going
to
leave
the
other
levels
available
for
policy
agents
that
get
layered
on
top.
So
that
means
that,
no
matter
what
the
user
does,
they
can't
go
ahead
and
override
scale.
You
know,
or
they
can't
go
ahead
and
override
this
attribute
and.
B
C
At
the
end
of
the
day,
you're
not
doing
any
smarts
or
maybe
what
you
do.
Is
you
write
these
layers
in
and
then
and
then
maybe
there's
a
there's,
a
controller
that
comes
through
and
actually
looks
at
the
annotations
and
then
rewrite
the
dancing,
or
maybe
that
happens
client-side
you
know,
maybe
it
ends
up
being
into
mission
controller.
That
does
the
merge
right.
So
you
right
in
the
end
of
the
day,
you
write
an
annotation.
You
never
actually
write
the
spec
and
what.
C
I
think
applies
better,
but
oh
well,
if
we're
going
to
do
that.
That
starts
getting
towards
the
v2
API
stuff.
That
Tim
was
talking
about,
and
it's
like
I
really
hate
to
like
put
that
much
crap
and
annotations.
That's
a
little
bit
scary.
To
be
honest,
so
the
way
I
like
the
things
you
suggested
that
apply
I,
don't
know
I,
don't
know
which
of
them
are
my
dry.
So
I
just
know
that,
like
I
look
at
apply
and
I'm
like
okay,
logical
decisions
have
led
us
to
a
crazy
destination.
Yeah.
B
C
B
C
C
B
C
So
I
think
it
would
be
worthwhile
to
actually
separate
out
the
different
commands
into
groups,
whether
they're,
actually
imperative
or
declarative,
or
at
least
come
up
with
some
sort
of
names
for
different
tracks,
so
that
people
realize
these
commands
work
well
together.
These
commands
work
well
together.
Okay,
if
you
try
and
mix
these
things,
you're
going
to
have
a
bad
time
and
then
here's
how
to
convert
from
one
mode
to
the
other
write.
C
It
I
think
if
we
just
play
this
off
as
modal
and
we
actually
and
maybe
we
use
in
an
annotation
to
say
hey.
This
is
a
this
resource
is
being
managed
declaratively
and
so
then,
if
somebody
goes
through
and
they
actually
try
and
muck
with
it
with
a
set
operation,
you
get
a
warning,
saying:
hey.
This
thing
seemed
like
even
shorter
doing:
full
ownership,
just
some
sort
of
hint
to
the
UI,
saying:
hey
you're,
mixing
modes
here,
you're
going
to
have
a
bad
time,
yeah.
B
C
B
C
B
Read
it
and
the
second,
so
that's
the
first
pieces
we
just
need
to
be
upfront
with
users
can
tell
them
like
here's
cheers
of
different
modes
and
yeah.
You
can
switch
back
aboard
notes
or
think.
The
second
thing
we
need
to
do
is
we
need
to
do
what
you're
suggesting,
which
is
make
the
plans
hard
enough
so
that
they
don't
interact
with
one
another.
If
you
try
and
run
apply
on
a
resource
that
doesn't
have
the
annotation,
it
says
just
in
the
wrong
mode,
you
can
even
use
rice.
C
So
there's
other
stuff:
that's
like
wacky
here,
wackadoodle,
there's
the
and
it
doesn't
work
well
across
these
things.
Also,
I'm
trying
to
remember
what
is
it
there's
the
thing
that
actually
stores
in
an
annotation
the
lap
record,
the
dash
dash
record
record
current
to
control
commands
in
the
resource
annotation.
This
is
used
for
for
deployment
so
that
actually,
when
you
do
a
deployment
sort
of
you
know
rollout
list
or
whatever
the
thing
that
actually
shows
you
your
history
of
deployments,
it
can
actually
have
oh
here's
what
actually
caused
this
deployment
using
a
declarative
workflow.
C
All
of
those
commands
that
get
recorded
our
to
control
apply
to
control,
apply
cube
control
of
life
right
because
that's
the
command
that
caused
the
change
and
so
I
think
it's
worthwhile
to
think
sure.
I
think
that
that's
another
feature
where
it's
just
you
know
it
wasn't
thought
through
with
these
sort
of
different
modes.
So,
as
we
add
this
stuff
up,
I
think
it's
more
than
just
you
know,
documentation.
We
also
have
to
figure
out
which
of
these
features
works
well
in
which
of
these
different
modes.
I,
don't
know!
B
B
So
over
yeah,
with
the
record
like,
for
instance,
I
think
there's
things
we
could
do
like
we
do
the
commit
hash
and
I
think,
when
we're
guiding
people
on
the
fly
like
we,
we
want
it
to
be
close
to
a
git
workflow.
But
if
you're
using
apply
one
and
be
like
change,
your
configs
or
check
your
configs
into
an
SDN
probably
get,
and
we
ask
you
to
make
a
commit
message
and
then
we
can
build
the
tools
that
when
we
call
apply
it
includes
the
last
commit
message
that
blank
all
the
fun
stuff,
but.
C
So
it
turns
out
that
the
way
that
it
works
in
it
and
I
figure
out
a
happy
workaround,
that's
probably
in
a
break
at
some
point-
is
that
if
you
don't
actually
set
the
annotation
in
the
deployment,
but
you
actually
set
it
in
the
template,
then
it
actually
doesn't
get
whacked
as
it
creates.
The
replica
sets
from
it
that
that
thing
comes
through
and
then
it
shows
up
in
all
new
I.
So
there's
some
work
around
here.
C
B
Helpful
Joe
I
think,
like
you,
said
some
of
the
stuff
with
discussed
already
and
had
some
more
ideas,
some
of
its
new
and
maybe
coming
in
a
play
at
a
different
angle,
which
is
helpful
like
a
big
problem.
We
have.
Is
that
there's
a
lot
of
work
to
do
here
and
a
lot
of
it's
like
basic
work
to
make
some
things
not
broken
like
Fox
be
like
our
applies
breaking,
and
we
can't
you
know,
do
a
production
push
like
that's
that's
kind
of
the
level
we're
at
right
now
is
like
just
survival.
C
Mode
yeah,
so
I
personally
am
NOT
going
to
be
able
to
spend
a
lot
of
time.
Hacking
code
here,
I've
already
committed
to
doing
things
and
other
cigs
that
I'm
falling
behind
on
it's
very
embarrassing,
but
we
are
going
to
behind
to
be
hiring
some
folks
up,
and
so,
as
we
bring
people
on
board,
I
want
to
be
able
to
help.
Have
them
help
out
with
some
of
this
stuff.
You
know
I
think,
and
it
sounds
like
you
know
you
have
some
folks
starting
to
who
are
going
to
be
kicking
on
some
some
starter
projects.
C
B
So
I'm
getting
I'm
having
one
person
full
time
on
this,
I'm
hoping
to
get
another
person
full
time
on
this
shortly
six
months
and
so
we'll
make
some
progress.
I
just
want
to
set
the
expectation
that,
like,
if
we're
not
working
from
the
user
experience
that
it's
not
because
we
don't
have
ideas
that
don't
think
of
a
folding.
C
Think
it's
worthwhile
for
us
to
push
to
have
a
group
of
people
who
are
really
interested
in
what
is
the
user
model?
What
does
these
are
experienced
around
this
review
new
features
to
make
sure
that
they
hang
together?
Well,
there's
the
API
machinery
group
they'll,
review
api's,
but
they're
really
looking
at.
C
Is
this
correct
from
sort
of
a
long
term
model
point
of
view
and
it's
very
easy
to
have
the
right
api
and
with
a
shitty
user
experience,
and
so
right
now
with
Cooper
Nettie's
for
good
or
ill,
the
rubber
meets
the
road
with
the
CLI
sort
of
the
interface
with
the
user.
You
know
we
don't
have.
The
GUI
is
definitely
secondary
to
that
and
I
think,
probably
as
it
should
be
a
lot
of
this
stuff.
You
know
the
most
successful
people
in
this
world
in
terms
of
building
awesome
user
experience.
C
They
don't
think
about
a
feature
until
they
actually
think
about
what
is
the
command
line?
What
is
the
stuff
that
we
type?
What
does
it
do
and,
as
we
were
doing
cube
avidin?
That
was
definitely
something
that
we
were
taken
into
mind.
Also
right.
We
were
just
really
focused
on
look.
You
know
we
can't
have
people
copy
around
a
token.
That's
like
you
know,
hard
characters,
long
have
to
be
short
and
that
drove
a
lot
of
the
sort
of
architectural
decisions
that
we're
still
working
through
here.
C
That,
from
a
sort
of
pure
API
point
of
view,
would
seem
silly,
but
from
a
user
experience
down
at
the
command
line,
it
ends
up
being
critical,
so
I,
don't
you
know
I
think
we're
going
to
have
to
work
to
get
that,
but
it
makes
sense
to
make
that
user
experience
a
developer.
Experience
of
you
we'd
be
something
that
gets
taken
into
account
early
on
as
features
and
aspects
of
the
system
of
being
developed,
yeah.
B
B
B
I
agree,
I
think,
there's
there
are
some
things
I've
been
thinking
about
in
that
aspect
like
I
I
do
think
the
lending
files
is
really
important.
I
think
the
disk
being
able
to
dip
or
do
like
you're
saying
a
plan
is
really
important.
I
think
right
now
we
don't
have
aa
one
thing
I'd
like
to
see
is
we
don't
have
a
config
file,
/
directory
to
in
like
flags
or
to
pain
like
certain
important
things,
and
it's
like
like
for
prune
I
know
if
you've
seen
the
prune
command
yeah,
the
problem
was.
C
Pruned
is
how
do
you
make
sure
that
you're
pulling
everything
in
you
don't
delete
stuff
on
accident?
Exactly
that's
the
problem.
You
know,
and-
and
you
know,
and
eventually
and
I
understand,
it's
super
early
days
here,
like
I,
do
think
that
we
need
to
like
bill
some
like
sort
of
higher-level
language
into
this
stuff
that
gets
compiled
down.
We're
never
going
to
get
everybody
agreed
upon
that
right.
C
So
it's
like
it's
just
not
going
to
happen
right,
but
maybe
if
we
actually
use
this
sort
of
like
manifest
idea
as
a
starting
point
for
saying,
okay,
here's
an
extension.
We
have
a
language
that,
instead
of
declaring
the
mo
files
directly,
you
actually
have
a
yamo
manifest
manifest
right
that
actually
points
to
other
files,
and
that
ends
up
being
one
of
these
early
sort
of
plug-in
sort
of
extension,
e
types
of
things
right
we're
now,
instead
of
actually
naming
the
files
directly,
you
name
this.
C
C
I
hate
to
make
that
so,
first
of
all,
okay,
so
here's
the
thing
you
know:
I,
we
brought
on
a
tech,
writer,
editor
and
she's
starting
to
use,
get,
and
the
only
thing
she'd
used
before
was
like
the
the
GUI
get
stuff
and
I
was
showing
her
like
the
pr
workflow
and
all
those
stuff
works
and
really
base.
Is
it
and
it's
like
it
really
drives
home?
The
point
I
get
is
a
piece
of
crap
right
from
a
usability
point
of
view.
C
It's
just
awful,
so
I
like
the
idea
of
being
able
to
tie
into
some
sort
of
SCM
my
I
get
a
little
bit
scared
when
you
fake
it,
because
I
just
don't
want
to
special
case
it
right.
I
mean
it's
obviously
going
to
be
a
big
part
of
it,
but
I
think
like
having
a
way
such
that
you
know
there's
this
can
plug
into
some
some
sem
some
sort
of
version
makes
a
time,
although
he
say.
C
B
One
of
those
or
the
last
thing
is
labeled
and
that's
what
they're
using
right
now
and
it
scares
Nicole
age
that
they
do
it
now,
because
it's
Flag
driven
for
each
time.
You
run
the
command.
You
have
to
supply
the
right
flag,
so
so
having
a
config
at
the
beginning.
That
says
always
apply
these
labels
to
everything
you
create
and
then
always
be
able
to
find
things
that
you
deleted,
like
this
sort
of
thing,
I
think
might
yeah.
C
And
it
smart
I
mean
it's
I
mean
we
could
use
this
to
sort
of
like
try
out
some
of
the
sort
of
third-party
templating
ideas
and
like
build
some
extension
points.
There
I
think
you
know
my
mind
that
makes
a
ton
of
fun
speaking
okay.
So
it's
like
circling
back
to
the
glob
thing.
I
go
out
for
a
bad
idea,
because
then
that
encourages
people
to
put
structures
in
to
name
the
name
shouldn't
have
structures.
C
B
C
Way
we
put
the
right
thing
to
do
in
that
way.
I'm
going
to
say
right
here
is
that
we
could
have
cube
controlled,
yet
you
can
do
pods
and
then
you
could
actually
have
you
know
labels
from
and
then
you
could
actually
named
your
deployment
there.
So
that's
essentially
saying
get
me
all
the
pods
for
this
deployment
and
then
behind
the
scenes
that
should
actually
go
through
and
actually
look
up.
The
right
label
query
does
to
make
that
make
at
make
that
happen.
Yeah.
B
C
C
C
You
know
alpha
and
then
I'll
have
another
thing
called
alpha
beta
and
then
somebody's
going
to
glob
and
they're
going
to
forget
to
dash
and
they're
just
going
to
do
alpha
star
instead
of
alpha
dash
star
and
then
alpha
alpha
star
will
pull
out
alpha
beta,
also
right
and
so
now
you're
like
oh
well
dashes
are
special
and
so
will
make
the
glob
to
actually
be
dash.
Aware
similar
to
directories
and
slashes,
and
then
it's
like
well,
you
know
the
world
ones
who
rolled
over
right.
C
I
mean
like
better
now
right,
I
mean
like
there's
going
to
be
educators
that
are
going
to
lead
people
in
the
wrong
direction
that
will
lead
to
outages,
because
now
I
have
something
that's
actually
acting
on
all
the
Alpha
stars,
sometime
down
the
road
somebody
does
now
for
alpha
beta,
you
know
deployment
and
then
the
world
breaks
right
and
it's
a
total
land
mine,
that's
waiting
for
them
if
they're
actually
using
logging.
This
way,
oh.
B
B
B
Yeah,
that's
fair
I,
don't
know,
maybe
tab
completion,
I
don't
know,
but
how
about?
If
you
want
to
just
comment
on
the
issue,
then
maybe
we
can
resolve
it
there
and
okay
yeah
I
in
the
label
things
fine!
Maybe
you
can
just
post
like
here's,
here's
the
UX
ionization.
Then
we
can
go
sat
if
it's
everyone
agrees
is
fine.
C
A
Joe
I
get
I'm,
sorry,
you're,
gonna,
say
something
probably
now
I,
don't
think
so.
At
least
India
agenda
I
didn't
have
anything
for
today.
Oh
they're
dama
that
was
so
yeah.
It
was
actually
great
tool
to
have
do
yer
and
to
fuel
our
agenda
with
some
actual
topics,
but
its
twenty
work
to
do
like
don't
worry,
absolutely
absolutely
yeah.
So.
C
Going
like
you
know,
I'm
I'm
slammed
so
I,
don't
know
how
much
road
work
I
can
do
yet,
but
but
eventually
you
know.
I
am
interested
in
having
some
folks
work
on
this,
and
so
you
know,
hopefully
we
can
sort
of
pave
the
way
and
have
some
a
good
backlog
of
ideas
and
stuff
so
start
designing
working
on
stuff.
So.
B
I
do
like
I,
like
your
feedback,
I'd
like
to
continue
to
get
it,
but
I
don't
want
you
to
feel
like
you're
I
know,
as
you
said,
your
slam.
What's
the
best
way
to
get
like
the
right
then
get
level
engagement.
That's
right
for
you
like
I,
can
I
can
put
all
the
features
in
the
agenda
meeting
notes,
and
then
you
can
just
look
at
them
every
week
and
see
yeah.
C
I
mean,
let's
I
mean
I'll,
keep
trying
to
make
the
meeting.
You
know
whenever
I
can
and
so
I
think
that's
a
good,
that's
a
good
way
to
do
it
as
I
bring
folks
on
board
I.
Think
we're
going
to
have
to
do
our
own
prioritization
exercise
here
and
see
where
we
want
to
put
them.
I
think
this
is
a
critical
place
to
actually
start
start
leveraging.
People
the.
C
Hopefully
you
know-
and
you
know-
hopefully
you
know
as
towards
the
end
of
this
month.
You
know
work
on
the
book,
will
calm
down
and
and
some
other
stuff
will
calm
down
and
I'll,
be
able
to
actually
have
more
of
my
own
bandwidth
to
actually
bring
to
bear
okay,
yeah
I
do
think
this
cat
crosses
over
with
some
of
the
config
and
templating
stuff.
Also
and
I
just
don't
know
where
to
talk
about
that
stuff,
because
at
some
point
it's
like.
Oh
that's
part
of
API
machinery,
but
like
holy
crap.
B
Think
this
is
a
good
group
to
talk
about
the
stuff
that
something
my
teams
working
on
at
Google
or
will
be
along
in
his
take
on.
If
you
look
at
the
app
or
it's
already
I
think
it's
the
CLI
arm,
QR
sorry
2017
gold,
like
granite
souls,
is
looking
at
the
end
to
end
ownership
from
I
have
a
binary
jar
thing
to
running
in
Coober
nedick's
right,
and
that
means
some
part
of
that
is
factoring
your
configs
right
and
the
solution.
B
There
could
be
templates
in
the
API
machinery,
but
it
could
be
Jason
net
or
it
could
be
something
entirely
different
right,
and
so
our
team
definitely
needs
to
look
at
what
the
right
for
someone
structuring
their
configs
and
directories
and
factoring
them
so
that
there's
not
redundant
question
in
them.
Yeah.
C
As
you
know,
in
Google
is,
if
you
look
at
the
the
load
balancing
config
right
in
GTE
writers,
you
need
to
configure
three
or
four
objects
to
get
a
load,
balancer
up
and
that's
what's
exposed
in
g-cloud.
And
then,
when
you
go
to
the
UI
in
the
console,
all
the
the
UX
designers
are
like
I've,
too
complicated,
we're
going
to
fix
it
right
and-
and
so
you
have
the
sort
of
streamlines
view
that
fixes
this
and
then
there's
an
advanced
but
well.
C
The
problem
with
the
streamlined
view
is
that
if,
if
you,
if
you
do
stuff
that
doesn't
fit
in
that,
then
then
it
breaks
right.
So
if
you
actually
do
have
these
unique
needs
and
then
it
breaks,
but
the
other
important
things
is
that
we're
teaching
users
one
model
of
the
world
and
then
as
they
move
to
something
like
g-cloud
or
as
they
get
more
complex.
C
All
that
knowledge
all
that
under
scene
and
all
that
intuition
that
they've
built
up
they
essentially
have
to
throw
away
and
start
all
over
it
and
so
I
think
the
problem
with
config
is:
how
do
we
actually
sort
of
do
that
thing
where
we
make
the
easy
thing
busy
in
the
hard
thing
possible
and
make
it
work
in
a
bunch
of
different
contexts?
So
that's
that's.
The
sort
of
you
know
world
view
on
bring
into
this
yeah.
B
C
Because,
oh
don't
say
right
now
I
mean
you
know,
I'm
writing
this
book
and
it's
like
you
know
you
have
to
get
like.
You
know.
Eight
chapters
in
before
you
have
a
good
working
understanding.
Other
thing
yeah.
Well,
I
think
it's
something
that
we
start
out
with,
like
you,
control
run
and
it's
like
oh
and
that'll
create
a
deployment,
but
we're
talking
about
pods
now
so
ignore
the
whole
deployment
thing
right
mean
it's
like
hell,
yeah,
like
I,
don't
even
know
how,
like
you
know,
I.