►
From YouTube: Kubernetes SIG CLI 20211006
Description
Kubernetes SIG CLI bi-weekly meeting on October 6, 2021.
B
C
B
D
B
B
E
So
we
have
a
like
a
high
level
outline
of
how
that
might
look,
we're
going
to
put
together
a
q,
a
document
to
send
out
just
generally
for
community
visibility,
because
this
is
also
going
to
apply
for
cube,
adm
and
other
things
down
the
road
that
want
to
move
out
of
kk
and
still
take
advantage
of
the
the
release
process
and
promotion
of
artifacts
and
things
at
a
really
high
level.
What
that'll
boil
down
to
is
6
eli
will
be
able
to
own
the
release
process
for
building
artifacts.
E
However,
they
want
to
build
them,
but
those
artifacts
will
need
to
get
into
a
staging
bucket,
eventually,
that
staging
bucket
can
then
go
through
the
same
kind
of
image.
Promotion
process
that
we
do
for
everything
else,
some
of
the
cluster
api
things
do
that
now
cops
is
using
the
same
image.
Promotion
tooling,
to
do
some
of
the
things
they're
doing
so,
there's
a
lot
of
like
history
that
goes
into
that
we're
going
to
deep
dive
more
into
that
at
the
next
release
engineering
meeting,
so
the
sig
release
has
bi-weekly
and
then
off
weeks.
E
We
do
the
release
engineering
meetings,
which
is
the
more
technical
and
mechanical
things,
but
that's
going
to
be
canceled
this
week
because
of
kubecon,
just
as
your
bug,
scrub
is
going
to
be
canceled
because
a
lot
of
people
aren't
going
to
be
around.
So
what
what's
going
to
come
out
of
this
is
that
we're
going
to
end
up
building
a
cap,
that's
going
to
describe
what
this
process
looks
like.
E
I
think
you
had
one
already
for
moving
some
of
the
components
out
of
tree,
but
this
is
going
to
be
a
higher
level
kind
of
umbrella
thing
about
how
this
will
look
for
other
projects
as
well.
Just
what
the
mechanisms
for
building
things
out
of
the
tree
look
like
nabaroon
is
going
to
be
running
point
for
that,
but
I
will
be
a
good
liaison
between
six
cli
and
sig
release.
E
So
when
it
comes
down
to
actually
implementing
some
of
these
things,
I'm
totally
happy
to
pitch
in
and
help
and
get
like
the
image
promotion
working
things
like
that.
So
we're
going
to
try
to
send
out
the
like
a
summarization
high
level
q
a
sometime
this
week
and
then
start
working
on
the
cap
after
that,
and
then
we
should
have
more
concrete
steps
for
y'all
to
take.
But
I
wanted
to
come
and
give
an
update.
F
E
That's
another
thing:
that's
on
the
high
level
q,
a
we
think
that
it's
probably
better.
If
you
do
it,
however,
you
want
you
don't
need
to,
and
it's
probably
better
if
you
don't
adhere
to
the
kubernetes
release
or
versioning,
the
the
important
thing
will
be:
maintaining
the
matrix
of
the
versions
you've
released
to
kubernetes
versions
that
get
released
just
for
the
sku
policies
and
stuff,
probably
already
has
support
for
doing
all
the
skew
testing
that
we
would
need
to
do
so.
C
E
F
Cool
yeah,
I
I
totally
I
agree
with
mate
on
that,
but
I'm
glad
that
you
all
are
thinking
about
it
a
little
early
on
and
it
is
going
to
be
very
confusing,
so
we'll
definitely
have
to
put
some
thought
into
actually
like
I
I
don't.
I
don't
even
know
we're.
Gonna
have
to
figure
out
how
we
do
a
version
skew
the
right
way.
So.
B
E
C
So
we
should
be
good
on
that
front.
Yes,
there
will
be
obviously
always
some
edge
cases
where
I
don't
know
a
particular
api
was
removed
or
will
be
removed,
and
it
so
happens
that
in
a
cube,
cuddle
it'll
be
there
for
for
a
little
bit
longer
it'll
disappear
sooner
so
it,
but
we
should
be
good
most
of
the
time.
F
E
Well,
if
there's
any
other
questions
happy
to
answer
them
offline,
happy
to
answer
them.
If
you
all
want
to
stop
by
sig
release
the
the
next
meeting
will
be
in
two
weeks
and
it'll,
be
the
regular
bi-weekly
sig
release
meeting
and
the
one
after
that
will
be
the
release
engineering
meeting,
where
I
think
we'll
have
more
in-depth
discussion
on
things
that
will
happen.
F
E
A
lot
of
meetings,
a
lot
of
meetings
for
everybody
all
right,
so
I
have
a
second
topic
on
the
agenda
too,
and
I'll
just
jump
into
that.
We
katrina
and
I
have
been
meeting
with
the
kept
team,
and
I
see
some
of
them
on
the
line
as
well,
mostly
collaborating
around
krm
functions
so
plugins
or
functions
that
you
can
use
in
customize
same
thing
for
kept
kpt
kind
of
built
around
the
specification
that
lives
in
the
customized
repo
we
would
like
to.
E
We
have
a
couple
of
caps
that
are
in
flight
right
now,
one
to
define
a
catalog
of
what
these
things
might
look
like
one,
the
piggybacks
off
of
that
and
defines
a
like
a
official
repository
of
plugins
and
then
probably
some
follow
up.
That
comes
from
that,
and
as
we're
talking
through
these
things,
we
thought
that
it
would
be
good
to
move
these
to
a
more
official
thing,
probably
sanctioned
by
the
sig.
E
E
That
would
serve
some
of
these
things
and
we
really
wanted
to
come
and
just
bring
this
topic
up
before.
We
actually
made
an
official
request
to
kick
off
a
subproject,
so
I'm
happy
to
to
engage
in
some
discussion
here.
If
anybody
else
wants
to
underline
from
the
group
wants
to
mention
anything
else
or
or
comment
happy
to
have
you
jump
in
as
well.
B
H
Oh,
it's.
The
the
design,
isn't
the
exact
same
as
for
crew,
because
the
plugins
are
not
installed
and
managed
the
same
way
in
in
customized
in
particular.
So
the
the
plan
is
somewhat
different.
It's
outlined
in
our
caps,
but
we
would
still
need
a
repo
and
some
hosting
infrastructure
to
drive
it,
which
I
guess
is
similar
to
what
career
needs
as
well.
From
that
perspective,
I'll
add
a
link
to
the
cap
in
the
agenda,
the
one
that
actually
requires
the
infrastructure.
I
B
Okay,
thanks
everybody,
so
so,
actually,
before
we
move
on
to
clint,
I
noticed
that
there
was
a
couple
of
announcements
that
also
got
added
a
little
bit
tardy
and
I
just
want
to
quickly
mention
those
that,
in
addition
to
the
talk
at
kubecon
for
the
six
eli
updates,
katrina
and
jeff
are
also
doing
a
talk
that
has
to
do
with
customize.
Is
that
correct
katrina.
H
B
A
A
A
B
I
B
D
Sure
so,
basically,
what
we're
doing
is
we're
trying
to
replace
the
netcon
that
gets
created.
The
opencvd
project
needs
to
own
the
connection
that's
created
it
does
that,
so
that
it
can
do
all
the
zero
trust
good
stuff
that
it
wants
to
do
and
then
oftentimes
you'll
find
apis
will
expose
some
way
of
providing
that
function.
Oftentimes
through
a
dial
function.
Callback
that
you
can
call
it's
not
really
obvious
in
this
pr,
because
the
way
it's
used
and
consumed
is
not
inherently
obvious
in
the
pr,
but
basically
down
below.
D
There
is
a
wrap,
config
function
that
happens
down
in
this
if
yeah
right
here
and
that
wrap
config
is
a
function
that
is
able
to
be
specified
in
the
cube,
config
flags,
it's
it's.
It's
kind
of
a
few
level
layers
deep.
Those
cube,
config
flags
are
created
right
up
above
on
line
278
right
there.
So
you
see
what
would
happen
is
every
time
we
would
call
new
command
whatever
the
constructor
function
is
called
there,
it
would
basically
new
one
of
these
structs
up,
and
then
we
had
no
opportunity
to
replace
the
dial
function.
D
D
So
what
this
will
do
is
it
lets
me
specify
the
cube
config
flags
when
I
create
the
new
command.
So
if
you
scroll
up
just
a
scooch,
there's
a
function
called
new
right
right
there,
new
cube,
seedle
command
with
config
flags
and
all
it
does
is.
It
allows
me
to
send
in
that
generic
cli
options.config
flags,
pointer
that
I
constructed
from
outside
of
the
command
and
then
the
old
new
cube,
seedle
command
gets
def
forwarded
over
to
this.
D
One
just
passes
nil
in
so,
if
it's
not
nil,
then
it'll
use
whatever
flags
you
pass
in
it'll
modify
them
like
on
line
280
there
you
see
it
adds
the
flags
it'll
do
all
the
same
stuff
that
it
used
to
do.
The
only
thing
it
won't
do
is
it
won't
override
the
wrapper
function,
and
I
think
maybe
mache
was
wondering
if
the
headers,
because
it
downline
440
there
you'll
see
crt.headers
is
kind
of
added
in
in
some
way.
Whatever
that
is
doing,
I
will
absolutely
admit.
D
B
So
so
this
especially
the
wrap
config
and
it's
using
a
kind
of
a
closure
for
that
headers
and
we
had
to
when
I
fixed
a
bug
here,
I
had
to
create
a
new
version
of
this
command
round
tripper.
Otherwise
there
was
kind
of
a
hang
under
a
run
minus.
I
t
so
this
a
lot
of
this,
especially
around
wrapping.
The
config
is
very
subtle,
and
so
I
think
we
should
be
very,
very
careful
about
doing
any
changes
here.
D
So
the
beautiful
part
is
the
cube
color
command,
which
you
are
providing,
doesn't
change.
You
you're
not
going
to
be
passing
in
these
flags,
the
I
we
have
to
fork
the
cube
cuddle,
and
so
I
have
a
fork
of
it
that
you
know
sprinkles
in
a
couple
lines
here
and
there
to
actually
set
this
cube
config.
So
the
the
existing
cube
cuddle
won't
change
at
all.
It'll
hit
the
if
block
and
if
it's
nil,
it'll
go
through
the
round
the
wrap
function
and
everything
like
it
used
to
right.
B
D
Sure
sure
it
I
don't
know
who
uses
the
functions
that
are
provided
by
the
library.
So
this
is
the
library
that
the
command
line
actually
uses.
So
I
was
excited
to
be
able
to
change
the
library
and
only
maintain
the
the
fork
of
the
binary,
which
is
like
10
lines
as
opposed
to
maintaining
my
own
fork
of
the
kubernetes
library
that
isn't,
as
as
great
in
my
opinion,
I
don't
want
to
have
to
keep
merging
and
maintaining
that
I'd.
Much
rather
not
have
to
do
that,
but
that
is
an
option.
C
I
literally
am
copying
this
entire
new
cube,
cuddle
command
and
injecting
where
wherever
I
need
to
for
for
openshift
specific
bits,
especially
that
we're
adding
extra
commands
here
and
there
as
well.
But
then,
at
the
same
time,
if
you
invoke
cube
cuddle,
ask
cube
cuddle,
we
will
provide
a
full
compatibility
mode
which
will
just
run
as
if
you
would
normally
work
with
cuddle.
C
So
that's
why
I
was
I'm
probably
in
a
similarly
hesitant
camp
as
sean
is
by
saying,
if,
if
you're
providing
something
that
is
kind
of
like
your
cuddle
butt
is,
I
don't
know
on
steroids.
Let's
call
it
that
way.
Maybe
you
should
go
all
the
way
to
the
point
of
just
create
the
the
command
as
you
need
it
to
to
be
the
rest
of
the
functions.
C
You
should
be
able
to
just
reuse,
and
I
can
probably
I
could
probably
see
us
restructuring
the
this
particular
file
to
make
those
constructs
easier
by,
I
don't
know,
maybe
creating
some
kind
of
factory
that
is
similar
to
what
we
do
with
generic
cli
options.
In
the
line
you
had.
Oh,
the
278
on
the
right
side
of
the
screen,
where
we
have
generic
cli
options,
new
contract
packs
with
deprecated
passwords
flag.
Maybe
something
similar
of
such
a
builder
pattern.
We
could
introduce
in
this
file
so
that
it
is
easier
to.
I.
D
C
C
So
maybe
just
restructure
this
entire
file
towards
some
kind
of
factory,
slash
builder
and
then
make
it
easier
for
everyone
to
reuse.
This,
oh,
oh
happy,
except
something
like
that,
because
it
will
make
my
life
easier
as
well
and
then
maybe
just
you
know
think
through
how
it'll
serve
to
other
folks
through
having
those
simple
bits:
more
configurable.
Basically,.
D
Yeah-
and
that
was
kind
of
the
the
idea
here
is
by
passing
those
flags
it
does.
It
exposes
more
for
better
or
for
us,
and
that's
the
kind
of
history
that
I
don't
have
as
to
you
know
what
what
are
we
you
know
allowing
for
somebody,
some
other
consumer
to
to
now
do
that
they
maybe
shouldn't
be
able
to
do
that.
They
now
kid
because
they've
exposed
this
sure.
F
That's
okay,
so
the
I'm
stoked
that
you're
using
this
as
live
library
code.
It
was
awesome
we
we
are
not
quite
at
the
point
where
we
are
versioning
this
as
a
library.
Yet
so
just
keep
that
in
mind,
as
we
just
talked
about
like
we're
happy
to
completely
change
and
break
that
file.
So
I
did
happen
to
pick
up
on
that:
okay
yeah.
So
just
just
that
caveat
like
we,
we
are
intending
for
this
to
be
consumed
as
library
code.
Eventually
we
have
a
bunch
of
refactoring.
F
We
need
to
do
to
like
pull
cobra
out
of
the
mix
for
all
the
library
functions
and
then
the
the
whole
new
returning
new
thing
is
like
definitely
the
the
go
way
of
not
breaking
compatibility.
So
I
totally
get
that
the
I
I
didn't,
take
a
big
look
at
the
pr,
but
as
far
as
unblocking
you
goes,
this
this
whole
wrap
config
function,
thing
that
should
probably
just
be
a
list
of
config
functions
right.
We
shouldn't
be
overriding
and
setting
a
single
one.
F
D
A
D
Unless
you're,
really
like
the
golang
standard
library,
actually
is
really
good
about
this,
because
z,
defying
ssh
was
cake
because
it
already
allowed
me
to
provide
a
dialer
function
back
so
any
way.
I
can
just
be
the
the
transport
creator.
That's
all.
I
really
need
from
any
of
this
and
then
like.
I
would
love
it
if
it
was
one
line,
just
one
one
deferred
to
the
dialer
function,.
B
So
actually
the
next
I
I
believe
the
next
topic
is
is
very
similar
to
yours.
You
want
to
have
control
over
the
dialer
over
kind
of
low
level
socket
stuff,
and
it
sounds
like
for
this,
so
I'm
gonna,
I'm
gonna
transition
to
the
next
item.
Clint,
if
that's
okay,
to
to
kind
of
connect
with
what
you're
trying
to
get
done.
Yeah.
D
B
No,
no
thanks
thanks
for
so
so
for
anybody
else.
If
there
is
a
change.
Probably
the
best
thing
to
do
is
to
bring
this
up
in
this
meeting
and
to
you
know,
try
to
do
it
over
email
or
don't
do
it
through.
Pr's
can
be
very
awkward
and
tedious,
and
just
getting
a
little
bit
of
face
time
is
is
worth
you
know,
hours
of
going
back
and
forth,
and
so
that
was
exactly
the
right
thing
to
do.
D
C
Might
my
reasonable
answer
is:
if
we
just
have
to
figure
out
this
and
the
vr
reviews,
it
might
go
through
several
iterations
back
and
forth,
so
to
have
a
prompt
responses
when
you
have
your
part,
ready
paying
myself
katrina
or
eddie
or
sean
or
all
of
us
on
slack
and
we'll
try
to
to
do
it
sooner
than
because
github
notifications,
at
least
in
my
case,
I
already
declared
it
back.
I
get.
C
D
B
D
B
Yeah,
so
so,
so
I
think
this
next
item
actually
is
similar
in
that
for
this
particular
bug.
When
we
have,
we
end
up
throttling
on
the
client
side
when
there's
more
than
say,
100
crds,
because
we
have
a
100
gps
limit,
and
this
is
where
we're
probably-
and
I
think
that
maybe
there
are
others
who
want
to
get
down
into
the
lower
levels
of
the
qp.
You
know
set
the
qps
or
set
other
socket
or
lower
level
communication
parameters,
and
that
which
is
where
this
is
similar
to
clint.
B
Once
you
once
you
get
to
about
100
qps,
I
think
as
the
client
go
and
you
got
more
than
100
crds,
you
end
up
going
to
be
throttling,
because
we
have
to
do
so
much
discovery.
We
have
to
do
one
for
every
single
crd
and-
and
it
sounds
like
we
may
be
over
discovering,
so
this
this
the
only
reason
I'm
even
bringing
this
up
is
just
for
visibility.
I
don't
think
we
need
to
make
any
decisions
here.
B
I
would
just
like
to
bring
this
up
for
visibility
for
the
rest
of
the
sig
that
this
there
there
is
this
problem
with
over
discovery.
It
looks
like,
and
we
may
want
to
cr.
You
know
to
even
solve
clint's
problem,
and
this
problem
give
a
mechanism
to
be
able
to
dig
into
the
lower
level
parameters
of
right
right
now.
It
doesn't
appear
that
we
have
a
simple
way
to
set
the
qps
or
to
set
some
of
clint's
lower
level
socket
parameters.
C
So
yes,
and
no
to
answer
your
first
question.
Yes,
we
already
have
an
option
exposed
if
you
recall
the
generic
cli
options,
it
has
a
bunch
of
builders
and
since
we
struggled
with
this
one
in
openshift
a
couple
months
back
a
few
months
back,
I
don't
know
last
year
or
something
along
those
lines.
I
introduced
something
that
it's
called
with
discovery
first
and
my
first
link
to
the
cli
runtime
points
to
a
particular
function
inside
of
the
cli
runtime
and
contactlex,
where
you
need
to
set
it,
and
so
we
can't
set
the
qps
there.
C
It's
a
burst.
You
should
be
good.
The
problem
that
we
have
in
openshift
by
default
is
that
we
are
running
more
than
100
crts
by
default,
so
we
were
hitting
the
the
limits
pretty
frequently,
so
this
is
bumped
to
250
as
it
currently
stands.
The
last
line
is
basically
showing
how
we're
wiring
that
inside
of
oc,
so
that
that's
the
openshift
client
line
that
also
answered
the
questions
and
what
I
was
mentioning
to
clint,
how
we're
running,
how
we're
wrapping
the
cube
cuddle
but
yeah.
C
D
D
C
It
is
currently
hard
coded
in
cube
cuddle
to
be
the
default,
which
is
a
hundred,
if
I
remember
correctly,
but
there's
nothing
stopping
us
from
bumping
the
limits
a
little
bit
higher,
which
should
solve
your
problem
almost
immediately,
because
with
the
proliferation
of
crds
by
average,
I
would
expect
a
hundred
crds
not
to
be
uncommon
in
the
wild,
so
yeah
it
it.
The
fix
is
very
simple
and
we
can
do
it
and
keep
cuddle.
So
just.
B
Just
like
that,
we'll
just
modify
the
qps
for
for
that
one
yeah,
so
yeah
daniel
smith
was
was
even
saying.
You
know
just
remove
it
all
together
the
qps
limit
there
there
is
server-side
throttling
as
well,
that
happens,
correct
and,
and
so
maybe
client-side
throttling
isn't
something
we
should
be
doing
anymore
anyway.
There's
I
I
didn't.
I
didn't
mean
to
to
get
into
this,
to
solve
it
right
here.
I
just
wanted
to
to
bring
it
up
for
visibility,
and
it
seemed
very
topical
for
clint's.
C
A
C
C
C
C
C
D
C
B
D
C
I
was
rather
thinking
about
restructuring
this
entire
file
into
something
similar
to
what
we
have
in
the
config
flags,
so
that
you
will
have
a
new.
Let's
say
we
would
do
new,
cube,
cuddle
and
then
would
be
different
with.
I
don't
know,
plugin
handler
with
a
generic
cli
options
and
then
eventually
construction
of
the
of
the
command
as
a
final
step.
C
Something
something
along.
Those
lines
I
think
would
be
would
be
good
and
would
make
your
yours
life's
easier
mine
as
well,
and
I'm
pretty
sure
that
it
will
be
other
a
couple
other
folks
who
will
benefit
from
from
this
change,
so
a
little
bit
something
a
little
bit
more
sophisticated
than
just
adding
yet
another
new
cubecardo
command
with
something
with
something
and
something
else,
because
that
will
be
really
a
pain
to
maintain.
F
Cool,
so
this
one
that
I
put
here
talking
about
the
krm
functions
made
me
think
about
it.
The
mark
has
done
a
fix
here,
and
this
is
something
we
talked
about
a
while
ago
when
we
were
promoting
the
bug
from
a
plug-in
to
a
built-in
command
of
how
plugins
can
be
shadowed
by
commands
and
back
and
forth.
F
C
F
L
Better,
I
think
all
the
all
the
other
commands
are
actually
are
are
checked
against,
so
that
you
can't
create.
You
know
you
can't
create
your
own
plugin
that
overrides
an
existing
command.
However,
you
can
create
your
own
plugin.
That
overrides
help
so
like.
If
you
name
your
plugin
right,
then
you
do
couple
help.
It'll
actually
invoke
your
plugin.
L
C
Okay.
Okay,
so
this
is
basically
to
ensure
that
we
are
protected,
because
since
we
are
reusing,
cobra
internals
there's
no
mechanism
that
will
ensure
that
cobra
will
pick
up
the
right
commands
and
since
all
the
others
we
are
just
registering
manually
within
cube
cuddle,
they
are
probably
a
properly
parsed
or
have
a
higher
precedence
than
the
plugins.
Okay,
yeah
sure
I'll
have
a
look
at
it.
After
the
call
end,
it
seems
pretty
straightforward
and
should
be
simple
to
to
tag.
B
G
The
only
thing
possibly
worth
mentioning
is
the
remember
we
mentioned
the
user
study.
The
survey
we
sent
out
a
few
weeks
ago.
We
have
some
initial
results
we
have
yet
to
set
up.
The
blog
cora
has
been
delayed
a
bit
from
her
grad
student
life,
but
so
I
was
debating
whether
we
should
start
discussing
and
sharing
the
findings,
or
should
we
wait
for
the
blog
to
go
out
and
get
more
results
before
biasing?
Anyone
with
the
results
we
have
so
far
happy
to
go
either
way.
G
G
I
can
maybe
maybe
I'll
wait
till
the
next
time.
Just
give
her
one
more
one
more
session.
I
can
I'll
send
you
the
link.
Anyone
who
wants
a
link.
That's
pinging
me
on
slack
on
I'll,
send
you
the
link
to
the
write-up
she
has
so
far
about.
I
figured.
I
don't
want
to
give
it
on
the
record.
You
know
yet
right
just
because
I
don't
want
to
buy
anyone.
You
know
in
case
we
want
to
get
in
case.
We
get
more
submissions
for
the
survey
got
it.