►
From YouTube: Kubernetes SIG CLI 20170510
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
B
A
A
B
C
C
D
C
C
We
are
kind
of
lumping
the
two
together
as
like.
We
want
to
fix
emerge
keys
for
certain
things
that
require
multiple
fields.
Is
the
merge
key
but
they're?
Really
like
two
orthogonal
implementation,
I
mean
we
could
fire
a
file
feature?
It's
not
user
visible,
it's
kind
of
an
internal
thing.
We
don't
even
publish,
merge
Keys
publicly
besides
the
code
but
I
know
what
do
you
think
I
mean.
C
D
I,
just
you
know
I
yeah
here's
my
take
on
it
is
that
you
know
I
I,
don't
track
this
stuff.
You
know
as
closely
as
I
probably
wish
I
could
and
so
there's
a
lot
of
conversations
that
are
going
on
and
so
I
think
having
some
way
to
write
this
down
so
that
no
other
folks
can
share
it.
You
know
in
proposals
because
I
mean
is
this:
there's
there's
over
well,
oh!
Well,
I'm!
Sorry!
D
D
D
C
Yeah,
the
clarification
was:
there's
there's
a
proposal
for
something
relatively
simple:
that
I
think
no
objection
was
very
little
objection
to,
but
in
order
to
change
existing
things
to
use
that
that
are
currently
broken,
it's
complicated
set
of
steps,
and
so
we're
scaling
back.
The
proposal
did
not
say
we're
not
going
to
figure
out
how
to
change
things
that
are
currently
broken.
We're
going
to
introduce
this
capability,
so
future
things
don't
get
broken
and
then
the
future
will
figure
out
how
to
fix
the
current
broken
stuff.
E
Okay,
I
just
have
some
small
TR
e
to
be
revealed
the
first
one
and
the
second.
Why
is
it
only
for
monkey
first
line
refectory
of
function
in
edit
option,
cope
just
make
three
function,
use
life
arguments
and
the
second
one
is
a
is
a
add,
a
simplified,
Chinese
translation
book
control.
It
just
sounds
just
the
TRD
to
be
revealed
the
only
people
who
know
about
Chinese.
So
this
is
the
second
one
like
I.
E
E
E
A
B
B
You
okay,
so
next
I
wanted
to
give
you
guys
a
status
about
plugins,
so
I'm,
pretty
sure
you
noticed
we
merged
v-0
of
plugins
still
without
too
much
noise,
and
you
know
it's
there
already
it's
in
keep
CTL
already.
If
you
take
a
look
at
keep
CTL
from
master,
you
already
had
because
keeps
it
out
plug
in
there,
but
given
that
it's
v-0
and
we
don't
have
really
any
documentation
or
other
than
the
proposal,
so
we
didn't.
B
You
know
advertise
that
on
the
public
mailing
list
or
anything
like
that,
but
anyway,
basic
support
is
there
and
that
what
I'm
doing
right
now,
because
I'm
working
a
little
bit
with
the
Service
Catalog
stuff,
specifically
about
adding
some
new
commands
to
the
Service
Catalog
and
we
are
looking
to
our
plan-
is
to
have
these
a
new
new
commands.
First
search
catalog
available
as
plugins
so
I've
been
able
to.
B
You
know
actually
start
using
it
less.
We
already
have
four
plugins
and
while
doing
that
already
working
on
v1,
so
v1
is
not
our.
You
know.
1.0
version
is
just
the
next
alpha
version
of
plugins,
where
we
add
some
new
stuff,
so
I've
been
working
on
it
already
and
basically,
while
implementing
this
new
course
catalog
command.
There
are
a
couple
of
things
that
I
really
missed
from
v-0.
B
B
What's
in
my
cube,
config
without
the
need
to
actually
read
and
implement
the
entire
logical
around
reading
it
and
all
that
so
gypsy
gel
is
also
at
environment
variables.
For
anything
goes
from
the
cube
config
that
we
that's
the
current
useful
field.
We
think
like
the
out
authentication,
information,
culture,
information
or
anything
that
I
eventually
need
to.
B
You
know
talk
to
the
server
so
in
the
plug-in,
I
need
to
access
the
Service,
Catalog
client,
so
I
need
to
know
you
know
service
out
in
for
anything
like
that,
so
the
tips
tell
is
going
to
provide
that
information
for
further
point.
So
I
expect
that,
probably
time
around
tomorrow
and
Friday
I'm
going
to
open
a
request
with
at
least
part
of
v1,
and
that
adds
that
kind
of
informations
also
basically
keeps
Dell
adding
more
context
to
the
plugins.
B
It
calls
right
so
context
based
on
a
global
Flags
configuration
and
even
some
some
other
informations.
That's
writing
in
Kyrgyz
right
now,
and
with
that
once
we
have
that
in
place,
we
already
have
you
know
stuff
enough
to
to
implement
some
kind
of
complex
plugins
like
the
Service
Catalog,
so
Service
Catalog
use
kind
of
complex,
because
there
are
a
number
of
new
commands
and
it
has
to
talk
to
a
Service,
Catalog
API.
You
know
do
from
from
read-only
stuff
like
get
HTTP
request,
but
also
updates
and
creating
new
stuff.
B
We
have
I,
think
three
versions
copy
of
CTL
plugins
before
going
final,
so
that
that
would
be
for
actually
starting
with
zero,
so
zero,
one,
two
and
three
versions
before
going
final
with
plugins
for
the
planets.
You
know
opened
up
this
v1
Pro
request
tomorrow
or
Friday,
and
right
right
after
that,
I
work
on
some
documentation,
so
that
even
not
being
final.
Yet
we
already
have
some
pretty
nice
support
for
for
plugging
lighters
to
to
use.
C
Different
ones
can
be
pruned.
Something
just
do
a
search
and
replace
for
like
client'll
versions.
Others
actually
are
more
complicated
and
require
moving
logic
out
to
the
server.
But
once
we
have
that
list
together,
we're
hoping
we
can
chart
it
across
the
sig
and
maybe
split
it
like
five
or
six
ways
to
get
it
all
done.
C
So
I
linked
the
pr,
and
that
means,
if
you
introduce
new
dependencies
and
control
on
new
kubernetes
libraries
available
break
yay.
The
other
thing
is
open.
Api
is
now
available
include
control.
What
we
do
is
we
now
hit
the
open
api,
swagger
endpoint
mr.
burnett
server
and
then
download
the
file
t
realized
into
a
binary
format
and
cashed
it
on
disk
sort
of
like
we
do
for
swagger.
A
big
difference
here
is
the
swagger.
C
Socks
were
broken
up
by
api
group,
which
were
much
smaller.
Then
what
the
open
API
has
open
API
is
everything,
and
so
the
consequence
takes
about
a
full
second
to
parse,
just
stupid,
CPU
JSON
parsing
to
binary,
which
is
actually
pretty
long.
You
do
notice
it
the
first
time
you
need
control.
So,
that's
why
we
see
realize
it
to
binary
format,
and
then
that
means
that
in
subsequent
times
is
like
10
milliseconds
to
read
the
cached
version
of
this.
A
D
I
think
this
is
one
of
those
where
it's
about
communication,
not
necessarily
about
looking
for
it
like,
because
you
know
it's
it's
honestly.
It's
a
localized
cute
control,
design,
change
right.
It
doesn't.
It
doesn't
leak
outside
its
sight,
acute
control
at
all,
but
the
behavior
change
of
that
that
you
know
a
lot
of
data.
One
second
ish
cause
at
the
start,
you're,
assuming
that
people
are
going
to
have
good
caching,
which
is
useful
when
you're
talking
about
the
typical
workstation
view
of
the
world.
D
But
there
are
people
who
do
things
like
clean
cube,
control
inside
a
container
and
spin
up
a
new
container
every
time
you
are
in
Q
control
and
then,
like
you
know,
just
throw
right
by
cube
config
in
there,
and
so
that's
going
to
slow
that
down
pretty
dramatically.
Similarly,
if
you're
running
a
server-side,
you
know
you
know
as
part
of
a
workload
there,
it's
going
to
slow
things
down
pretty
dramatically.
So
I
think
that
one
thing
to
consider
here
is:
is
there
a
trade-off
to
be
made
in
terms
of
like?
D
D
So
I
think
I
think
it's
just
a
matter
of
communication
of
like
look
in
these
scenarios,
we're
going
to
see
a
huge
pause,
we're
going
to
see
a
big
download.
You
know
and
I
guarantee.
What's
going
to
happen.
Is
that
you'll
go
out
with
this
and
nobody
else
cream
and
then
like
as
soon
as
you're
released
you're
going
to
see
like?
D
Oh
my
god,
like
all
of
a
sudden,
this
thing
got
slower
and
I
like
if
you
can
show
that
you
did
due
diligence
there
as
much
as
you
can
it's
like
can't
we
just
you
know:
Chris
implausible
I,
think
it's
a
it's
the
good
step
forward
to
actually
make
this
stuff
we
be
generated
and
stop.
You
know,
stop
making
the
some
API
special
and
some
API
is
not
special
I.
Think
that's
that's
a
huge
step
forward,
but
yeah
the
one
megabyte
downloading
compiled
sucks
yeah.
C
D
C
Can
that.
D
D
So
you
got
to
worry
about
version
skew
for
that
thing
across
different
protein
cube
control
because,
like
there's
going
to
be
people
here
on
a
control,
not
cube
control,
so
if
you're
using
like
like
any
sort
of
like
language
specific,
you
know
object
serialization
as
a
total
show
right,
since
we
realize
protobuf
with
just
the
stuff
that
you
need.
That's
been
precompiled,
maybe
a
same
way
to
do
this
and
they
are
they
download
the
protobuf.
D
So
now
what
we
say
is
that
we
have
a
compiled
version
of
open
API
that
can
be,
you
know,
generated
server
side
and
then
downloaded
versus.
Actually
you
know
PPI.
But
logically,
it's
like
here's.
The
chunk
of
code
that
you
can
run
server
client,
so
I
did
generate.
Do
that
compile
the
open,
API
down
yeah.
D
Know
and
again
like
this,
is
you
know
this
is
some
of
the
stuff
that
like
like,
if
we
had
you
know
a
good
way
to
get
a
design
review
on
those
stuff,
because
it's
like
look
this
is
you
know,
it's
not
officially
a
feature
yet
get
me
more
eyes
on
this
and
thinking
about
a
little
bit
and
trying
to
understand
the
full
implications
of
something
like
this
I
think
worthwhile.
We
don't
have
a
lot
of
good
processes
around,
although
no.
D
Working
on
I
think
you
definitely
want
you
know,
but
this
is
like
this
is
a
nasty
surprise
that
you
know
that
will
hit
some
people
harder
than
others.
Yeah.
D
So
officially
you're
not
breaking
anybody,
but
like
okay,
so
here's
the
thing
so
like
you
know
in
you
know
what
having
trouble
switching
to
go
1.8
because,
like
HTTP
slowed
down
dramatically,
and
that
means
that
we
no
longer
passed
the
5,000
node
scalability
targets
right.
So
this
didn't
break
anybody
as
soon
as
breaking
but
like
in
some
ways.
If
you
change
perk
that
dramatically,
it's
a
breaking
change,
yeah.
C
That's
true
on
the
client
side,
it'll
be
different,
but
I'm
sure
there's
some
weird
stuff.
We
could
run
into.
E
D
The
point
zero
seven
second
startup
was
what
I
heard
complain.
I,
don't
know
for
outside
for
the
pods
use
version,
previous
versions,
yeah
and
the
flag
is
definitely
a
good
place
to
start
but
like
if
we
look
at
sort
of
like
you
know
the
discussions
around
well,
that's
yeah,
I,
don't
know
so
okay.
So
let's
do
yeah.
There's
no
good
answer
here,
but
communication
communication.
We
should.
C
Like
once
at
a
minimum
we
can
do
is
like
have
a
disabled
by
default,
but
introduced
the
flag.
So
at
least
then
we
get
you
like.
You
can
opt
in
to
not
operating
yourself
and
then
figure
out
if
the
server,
if
it's
easy
to
introduce
from
the
server
right
like
right
now,
the
server
can't
parse
its
own
swagger
back,
which
is
kind
of
weird,
but
I
think
we
can
make
that
possible
all
right.
What.
D
I
mean
so
one
of
the
things
that
I
you
know,
I
can
give
you
a
little
bit
of
a
preview
of
sort
of
some
of
the
ideas
that
have
been
floating
around
is
we
look
at
the
are
vac
example
41.6
and
how
that
got
turned
on
with
cube
up
and
then
got
turned
off
again
and
so
we're
viewing
changing
defaults
as
being
sort
of
something
that
gets
tracked
in
and
of
itself.
So
so
you
know,
the
general
idea
is
that
new
features
come
in
they're,
disabled
and
then
sort
of
they're
so
sort
of
like
ok.
C
C
B
Yeah,
just
just
want
it
to
add
to
you
guys,
no
not
sure
if
I
ever
mentioned
that
before
in
the
CLI
meeting,
but
something
we
faced
a
few
months
ago.
We
related
to
you,
know
serializing
and
deserializing
on
the
the
client
on
keep
CTL
specifically
to
Jason,
Lee
and
yeah
mo
so
at
some
point,
not
sure
if
you
you're
aware
tips,
detail
change
to
use
internal
structures
objects
in
yet
in
their
printers,
specifically,
so
until
until
that
we
have
each
using
the
actual.
B
You
know
the
actual
lobs
objects
and
instruct
this
serializing
base
on
to
the
actual
objects
and
because
of
GPRS
at
some
point,
we
see
switch
to
instruct
your
object
and
I'm.
Pretty
sure
that
you
know
we
didn't
face
any
performance
issues
at
first,
because
you
know
you
really
didn't
notice
on
small,
smaller
sets
of
objects.
B
However,
in
out
our
Kiwi
team,
when
testing
against
you
know
really
large
set
sets
of
objects
like
1000
and
IOC,
get
started
to
take
like
four
times
what
it
took
before
right
so
and
that's
basically,
I
did
some
research
after
we
got
that
report
from
our
tqe.
That's
basically
because
of
the
map
of
interfaces
used
in
the
unstructured
object,
so
that
really
has
poor
performance
and
every
JSON
API
for
go
out
there.
B
B
D
D
B
Yeah
I
guess
there
there
are
some
good
and
bad
parts
about
it,
but
you
know
this
is
a
really
bad
part
about
it.
Performance-Wise.
However,
we
could
think
about
some
other
strategy.
Clients
outside
anyway,
maybe
protobuf,
something
like
that
anyways,
but
that
that's
the
two
proposals-
I
just
added
them
to
the
chat.
Okay,
I
will
add.
D
It
to
my
list
of
stuff
to
review,
yeah
I,
would
say
like
anything
like
the
cube.
Control
gauge
should
definitely
be
a
teacher,
because
that
would
that
would
cross
things
right,
as
we
start
doing
this
stuff
and
it's
user
impacting
because
there's
new
API
evolve.
It's
you
know
yeah.
It
seems
like
that.
That
would
be
a
big
enough
thing,
because
I
know
this
has
been
floating
around
for
a
long
time.
I'm,
not
the
only
one
with
opinions
here,
yeah.
B
I
think
I
think
you
know
even
something
like
heavy
and
output
format
that
is
not
structured
but
is
well
structured
in
the
sense
of
not
using
interfaces
and
regular
maps
like
having
an
actual
object.
That
says
like
field
key
name
and
really
value.
Something
like
we
would
already
fix.
The
issues
with
you
know
dates
on
performance
on
the
climber
site,
so.
D
Anyway,
yeah
I
think
doing
server-side
transformations,
there's
something
to
look
at,
but
that's
different
from
saying,
server-side,
rendering
right
so
yeah.
It's
not
you
know
finding
a
way
to
sort
of
solve
the
problems
without
having
to
sort
of
you
know
do
like
you
know,
count
tabs
and
spaces,
we
run
run,
run
the
templating
stuff
or
whatever
that
you
know
on
server-side
seems
better,
but
oh
I'll
read
through
the
proposals.
Oh
it's
my
list,
so
then
I'll
be
informed
next
time,
I
about
it.
Okay,
the
the
server-side.
C
C
One
other
thing,
I
kind
of
want
to
ask
you
guys
about
I'm,
not
looking
for
an
answer
here,
even
the
brainstorm.
There
may
be
something
we
can
think
about
the
next
time
we
talk
is
I
notice,
a
lot
of
PRS
that
come
in
that
I
did
I,
don't
think.
Have
anyone
on
this
meeting
necessarily
reviewing
them
for
COO
control
or
a
lot
of
features
that
don't
get
discussed
here
and
they
typically
are
like
apps
introduced
in
your
resource
and
they
want
to
add
to
create
commands
for
it
or
whatever.
D
C
D
I
you
know,
we've
been
talking
about
sort
of
like
what
is
a
cig
and,
and
one
of
the
things
that
I
think
resonates
with
me
is
that
cigs
are
formed
for
long
term
ownership
of
processor
code,
and
so,
if
somebody
checks
something
into
cube
control,
this
jig
and
whoever
this
CB
is
it's
going
to
be
responsible
for
that
code?
You
know
over
time
right
and
so
there's
a
certain
amount
of
process
that
this
it
can
impose
as
part
of
taking
on
ownership
of
that
code,
and
so
I.
D
Don't
think
it's
insane
for
this
for
the
safe
to
actually
set
standards.
In
terms
of
like
you
know,
you
have
to
do
XY
and
Z
if
you
want
to
introduce
a
command
and
sort
of
justify
yourself
and
maybe
come
and
present
so
that
we
can
sort
of
do
a
UX
orgy
of
this
thing,
that's
not
an
insane
thing
to
do.
I'm
Eric
different
comments
on
the
on
the
use,
namespace
thing
because
I'm,
like
okay
house,
is
consistent
with
other.
C
B
Some
something
we
we
did
here
at
Red
Hat
at
some
point,
was
yeah,
but
I
think
that
that
makes
a
lot
of
sense.
That
I,
probably
don't
have
an
exact
answer,
but
something
we
did
here
on
open
chips
specifically
was
to
exactly
distribute
these
on
our
ship
on
some
commands
for
the
actual
groups,
where
the
subject
of
the
command
applies.
D
So
I
hope
one
thing
to
recognize
is
that,
at
the
end
of
the
day,
the
you
know
whether
we
like
it
or
not,
the
CLI
is
an
api
and
we
have
to
treat
it
like
an
api
yeah
hey.
So
that's
that
one
that
mine
it
doesn't
answer
the
test
flight
question
but
like
in
terms
of
like
ownership
and
review,
like
you
can't
just
throw
stuff
in
there.
I.
C
D
I
mean
one
thing
that
is
worthwhile
is:
are
there
generic
things
that
can
be
done
in
queue,
control
that
can
release
some
of
the
pressure
around
this
stuff?
Like?
Can
we
take
the
create
templates
and
have
a
way
so
that
you
can
have
a
directory?
You
can
install
the
create
templates
for
the
foo
bar
thing
and
it
essentially
becomes
which
version
templates
are
you
using
and
that's
not
compiled
in?
C
C
That
was
exactly
and
I'm
like.
How
can
we
do
that
cool?
C
B
C
That
is
the
actual
stuff
yeah,
and
one
thing
you'll
notice
is
a
lot
of
this.
Just
has
the
overall
link
and
that's
because
the
J
unit
tests,
XML
or
whatever
is
being
published.
It's
just
like
the
go
test
that
just
has
overall.
So
one
thing
we
want
to
improve
this:
we
need
to
make
it
so
those
publish
the
J
unit
style
and
we
get
the
breakdown
per
test.
B
F
Building
a
fast
system,
we
have
been
working
on
it
for
past
one
and
a
Sears
right
now.
So
what
we
are
doing
is
integrating
some
bass
components
and
has
to
build
a
general-purpose
pass
system
for
users.
So
I
was
looking
at
some
of
the
issues
we
work
in
the
cube
CTS
back
of
it
so
I.
We
kind
of
figured
out
one
of
the
issue
that
we
are
also
interested
in,
which
is
about
the
cube
control
port
forward
and
cube
control
proxy
issues.