►
From YouTube: Kubernetes SIG API Machinery 20180801
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
A
B
C
B
B
Stefan
had
a
couple
outstanding
questions,
requests
for
minor
updates,
Erik
tuned
and
reviewed.
It
said
it
looks.
Ok,
it
looks
ok
to
me
as
well,
so
this
is
your
last
call
to
go
into
that
design
and
make
comments.
I,
anticipate
it
merging
sometime
this
week
and
I
am
hoping
to
see
the
start
of
implementation.
Also
this
week,
maybe
you
gonna
be
able
to
do
that.
D
C
B
B
A
E
E
It
was
basically
presented
last
time.
One
big
change
is
that
there
wasn't
proposal
I,
think
by
Eric
already
that
we
8
skeleton
validation
schema
from
the
main
one
automatically
and
David
and
media
also
proposed
that,
and
we
discussed
that
of
line-
and
the
cap
represents
this
now,
so
we
don't
expect
the
user
to
specify
the
skeleton
skeleton
is
automatically
computed.
Just
defaults
must
be
outside
of
the
contours,
so
sort
of
the
logic
part
of
a
variation
schema.
Otherwise
it's
as
before,
but
it's
completed,
especially
from
the
experience
from
the
implementation.
E
C
B
D
E
So
I
had
the
pleasure
to
look
into
Q
builder
and
operator
SDK,
which
uses
control
event
I'm
preparing
a
workshop
and
I
notice
that
it
looks
like
we
have
four
kind
of
clients.
Today
we
have
the
client
go
client.
Everybody
knows
we
have
sex
line
code
and
Emma
client.
Most
of
you
know
as
well,
and
now
we
have
one
client
problems
operator
SDK,
which
is
a
lower
slide
there
and
the
upper
one
is
about
you
builder.
E
E
They're
very
similar
cupid
uses
a
scheme
to
translate,
go
type,
choose
a
coupe
version
kind
and
operated
SDK
doesn't
use
a
scheme
at
all.
It
requires
that
you
set
API
where
from
and
kind
which
gives
you
the
same
information,
so
those
can
be
combined
unified
into
one,
and
my
question
is:
do
we
want
to
support
a
style
inclined
go?
Should
we
have
a
third
line
which
somewhere
lives
between
dynamic
and
type
clients.
B
So
I've,
actually,
given
this
some
thought,
cuz
I
was
working
on
the
dynamic
client
and
one
thing
that
people
had
asked
me
for
was:
why
do
you
make
me?
Do
the
rest?
Mapping,
myself
and-
and
the
answer
is
in
the
case
of
a
conflict
where
you
have
ambiguous
information
that
can
be
mapped
to
multiple
resources.
B
You
have
to
decide
what
to
do,
but
if
we
were
to
take
a
constrained
use
case
and
build
on
top
of
the
dynamic
client
that
we
have
entry,
you
would
be
able
to
include
a
rest
mapper
and
automatically
do
a
mapping
for
someone
on
top
of
an
unstructured
type.
Now
you
talked
about
how
you
actually
use
the
external
types
given
an
external
type.
B
B
E
E
F
F
B
Do
think
it's
up
to
us
to
help
guide
people
to
what
we
think
are
best
practices,
and
if
you
don't
know
what
it
is
you're
dealing
with,
then
the
best
practice
would
be
a
dynamic
client.
If
you
do
know
what
you're
dealing
with
then,
the
best
practice
is
to
use
the
typed
client
that
gives
you
the
type
safety
for
your
language.
G
B
B
B
C
F
H
B
E
B
E
Looky,
so
if
he,
if
you
have
this
generic,
not
you
know
it's
a
static
line,
so
it's
the
same
for
all
objects.
You
could
implement
things
like
Y
updates
those
helper
functions.
This
is
something
we
cannot
do
without
generics
was
a
type
clients.
We
so
I
suspect
that
in
open
up
way
to
educate
and
cupola,
those
tools
will
start
being
created.
G
F
B
I
B
G
B
It's
part
of
the
resource
builder:
it
takes
a
generic
rest
client.
You
have
to
make
sure
that
you
specify
the
correct
scheme
that
you
want
to
use
as
you
construct
it
and
build
it,
and
then,
after
that
point,
if
you
were
to
wrap
a
rest
map
or
around
it,
instead
of
passing
the
mapping
you'd
have
the
rest
of
the
pieces
present.
Okay,.
B
I've
seen
a
whirring
pattern
of
people
like
when
I
created
the
when
I
fixed
it
a
name,
add
client
people
came
out
of
the
woodwork
because
they
had
created
their
own
dynamic
clients
around
our
dynamic
client
to
create
their
own
dynamic
Informer's
around
our
dynamic
and
client.
Instead
of
helping
client
go
develop
the
dynamic
list
watch
it
former
pattern
and
I.
F
Think
I
think
part
of
this
may
be
that,
for
whatever
reason,
there's
a
number
of
people
that
find
some
of
the
constructs
in
Incline
code
to
be
more
complicated
to
use
than
they
wish
and
some
like
I
I,
think
it.
If
some
of
some
of
this
may
be
like
it
may
be
worth
considering
like
people
who
aren't
necessarily
main
kubernetes
developers.
What.
F
F
B
I
I
wouldn't
recommend
drift
for
doing
that,
though
right.
So,
if
you
look
at
operator
SDK,
they
created
their
own
list
washing
former,
and
then
there
was
something
from
Anthony
and
a
sock
that
did
something
similar.
They
both
did
it.
Similarly,
they
both
did
it
but
they're
slightly
different,
and
neither
one
of
them
took
it
to
the
upstream
and
it's
a
thing
that
we
be
interested
in
having
yeah.
F
I
mean
it's
less,
that
we
believe
that
a
generic
Informer
wouldn't
be
useful
in
general
and
more
that
we
wanted
to
put
together
a
prototype,
see
how
actual
users
used
it.
You
know
make
tweaks
and
feedback
and
then,
when
we
had
something
that
we
were
very
confident
and
we
could
come
back
and
say:
okay,
we
have
we
have
this
usage
information.
We
have
users
saying
this
is
useful
to
work
with
yeah.
B
B
And
like
I
know,
I
know
this
because
I
change
the
dynamic
client
and
then
all
of
a
sudden
there's
three
people
who
are
like
everything
just
broke:
why
did
it
break
and
then
what
do
you
mean
it
broke?
It's
usable
right
here.
Well,
I
built
these
pieces
on
top
of
it
well
you're,
like
the
third
person
that
I've
talked
to
that,
did
that
you've
all
built
the
exact
same
thing
and
you're
all
broken
in
different
ways.
B
G
B
B
We're
gonna
pick
on
Sally,
though
Sally
also
knows
me
and
Daniel
and
ciao
and
yeah,
and
explaining
the
use
case
and
then
saying
how
do
we
do
this
I
mean
I
at
least,
would
be
able
to
point
at
Q
control.
Do
it
like
this
see
if
you
can
just
move
it
and
collapse,
the
two
cases
and
we'd
have
it
in
client
go
today
on
the.
B
B
F
F
Think
sometimes
I
I
still
maintain,
though,
that
sometimes
it's
it's
it
can
be
very
useful
to
kubernetes
is
sometimes
like
kubernetes
core
is
sometimes
slow-moving
and
I.
Think
sometimes
we
also
put
stuff
in
there
that,
like
we
end
up
regretting
a
little
bit
and
so
I
think
there's
room
for
prototyping
outside
of
core,
especially
if
we
can
say
alright
we're
working
on
this
thing.
That's
new,
it's
not
necessarily
as
long-lived
or
it's
it's
not
as
strict
of
with
compatibility
guarantees,
and
then
we
move
it
into
core.
When
we
have
more.
I
I
B
So
we
own
both
repos
right,
it's
a
sub-project
of
API
machinery,
and
so
now
we
have
different
clients
and
different
sub
projects
and
we're
gonna
break
one
of
them
and
was
our
life
better
because
of
that
maybe
I
guess
maybe
fewer
people
use
it.
So
we
break
it
to
your
people
wine,
but
in
order
to
be
able
to
collapse,
they're
gonna
break.
I
H
G
H
G
H
J
J
What
we
had
like
I
wanted
to
provide
this
free
bit.
We
had
when
we
started
building
something
around
the
curated
API
and
go
client.
Is
we
had
exactly
that
problem?
We
wondered
we
did.
Everything
was
keep
control
before
and
we
expected
the
same
behavior
and
the
same
things
to
do,
and
we
ended
up.
Ok,
no,
we
can't
one
thing
was
the
apply
stuff,
I
think
I
discussed
on
the
select
a
bit,
but
the
behavior
was
quite
confusing
and
we
were
frustrated
then
that
people
taught
told
us
to
well
call
cube
control
in
the
shell.
B
On
a
pie
might
be
a
little
bit,
that's
gonna
be
a
hard
one
to
pick
apart
at
this
particular
moment.
Given
the
effort
that
is
has
been
underway
now
for
some
time
to
move,
apply,
server-side
yeah
that
feature
branch
is
well
underway.
I,
don't
see
us
trying
to
pull
the
existing
apply
logic
out
into
something
like
a
client
go
I
think
we
would
just
let
that
existing
apply.
J
Nope--Like
heart
beside
that
or
something
else,
but
the
weight
had
to.
How
do
you
get
those
objects?
If
you
get
if
you
get
young
move,
for
example,
and
push
them,
we
had
to
do
a
lot
of
work
around
with
the
dynamic
stuff
and
and
actually
how
to
find
out
which
were
version
kind.
Coubertin
country
have,
and
all
that
looks
very
different
than
it
seemed
to
be
in
cube
cgl,
and
we
wanted
to
have
this
behavior
yeah.
B
Yeah
I
mean
if
the
goal
is
I
want
to
have
a
cube
control.
Client
in
client
go
I
would
try
to
move
the
cube
patrol
client
to
client
go
and
things
like
you
know.
It's
shaped
in
a
particular
way.
Our
clients
are
shaped
in
a
certain
way
around
namespace.
So
you
choose
your
namespace
to
choose
you're,
saying
bigger
thing
and
building
a
visitor
pattern
around
it.
Maybe.
B
By
the
time
you
get
to
something
like
parsing
logic,
from
your
animal
you're,
usually
talking
about
something
more
like
a
CLI
tool
and
so
to
control.
Six
CLI
is
trying
to
create
a
library
for
use
for
building
CL
eyes
that
hasn't
yet
been
split
out
into
staging,
but
it
is
a
thing
that
could
be
built.
B
H
I
Chow
yeah
so
could
not
have
this
long,
long-standing
problem
that
we
cannot
defecate.
A
version
of
the
API
from
the
main
component
is
Rico,
because
some
store
well
because
the
storage
like
sed
might
store
the
API
seeing
an
old
version,
and
if
you
remove
the
old
version,
the
API
server
won't
be
able
to
reach
those
anymore.
I
So,
for
example,
when
we
migrated
to
deployment
from
extensions,
we
don't
beta,
1,
2
x,
v1
y
sed,
they
might
be
still
encoded
as
extensions
we
want
might
have
one,
and
you
cannot
just
remove
the
extensions
beyond
beta
1
from
Nadia's
code,
because
otherwise
HSM
I
won't
be
able
to
read
it.
I.
Have
this
talk
evaluating
the
existing
tools
and
designs
targeting
at
solving
this
problem?
I
So
we
saw
the
current
currently
I
plan
to
take
open
ships
approach,
which
will
be
a
command
line
for
that
for
how
how
people
to
upgrade
their
storage
when
they
upgrade
their
cluster
I
plan
to
upstream
the
staying
into
a
repo
in
the
coconut
is
odd,
or
maybe
the
coconut
is
saqqaq
and
might
get
alpha
in
the
next
quarter
and
when
it's
appropriate,
we
can
move
it
back
to
kubernetes
memory.
So
please
take
a
look
at
the
dog.
If
you
have
any
you're,
also
interested
in
upgrade
their
problem.
B
B
B
B
D
B
We
do
it's
not
that
I
do
not
like
this
I
want
to
make
that
clear
right.
It
says
this
is
building
on
a
tool
that
I
helped
build.
So
it's
not
that
I,
don't
think
the
design
is
good.
It's
the
fairness
aspect
of
it
because
I'm
assuming
there's
people
on
this
call
who've,
never
even
heard
of
the
tool.
This
is
built
on.
B
Yeah
yeah
I
think
you're
right
us.
You
know
you
think
about
things
like
you
would
expect
as
something
develops
fully
in
lifecycle.
There'd
be
some
sort
of
feedback
loop
with
our
cube
CI.
As
we
start
cycling
versions
in
cube,
you
would
expect
this
tool
to
have
to
be
run.
I,
don't
know
that
I
want
to
try
to
force
that
in
a
112
timeframe
on
people
and
112
right
so.
A
B
A
D
Interesting
things
that
are
not
done
before
future
phase
I
think
we
do
some,
because
this
is
the
life
cycle
of
this
one
is
completely
decoupled
from
kubernetes
and
we
do
interesting,
spec
clients,
Python,
client
and
Java
clients
that
has
their
own
life
cycle.
Okay
could
be
treated
like
that
know
when
it's
ready,
we'll
just
design
their.
You
know,
review
it,
and
if
it's
ready,
yeah.
B
I'm
worried
about
how
it
interacts
with
the
CI
pipeline
for
you,
we
have
upgrade
tests
right,
like
I,
guess,
I,
don't
know
if
everyone
on
the
call
is
aware,
but
there
are
a
significant
number
of
upgrade
tests
they
go
through.
They
try
to
run
through
the
prescribed
or
some
reasonable
upgrade
process
and
as
we
approach
release,
the
great
tests
will
block
the
release,
wrap.
G
B
B
E
B
J
B
A
Know
what
one
know
what
to
know?
What
three
okay
looks
like:
that's
it
for
today,
I
thank
everybody
for
participating
today.
I
will
upload
the
recording
later
today
and
send
a
link
as
usual,
with
seasons
by
videoconferencing
two
weeks,
hopefully,
Daniel
will
be
back
by
them
he's
doing
very
well
by
the
way
regard
and
a
good
rest
of
your
day.
Okay,.