►
From YouTube: Kubernetes SIG API Machinery 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).
B
D
C
A
A
We're
sort
of
still
working
out
our
client
strategy.
We've
got
a
document
with
the
escapes,
like
you
know,
make
clients
in
the
kubernetes
client
repo,
with
the
generator
repo,
we're
still
sort
of
staging
the
generator
repo
and
getting
it
ready.
And
so
the
the
question
came
up
of
Brendan
or
a
team
that
Brendan
notes
as
a
dotnet
client
that
they
generated
with
an
alternative
generator,
and
so
there's
some
discussion
on
the
mailing
list
about
whether
that
means
we
should.
A
E
Can
just
clarify
why
I
felt
it's
on
the
mailing
list
thread,
but
basically
there's
a
version
of.net
that
is
cross-platform
that
works
on
Linux
and
Windows
and
Mac
OS
generator
that
we're
using
generates
the
code
for
that
version
of
the
language,
these
distinct
bi-directional,
and
we
would
prefer
to
use
the
one
that
is
cross-platform
concerned
raising
Nettie's
concern.
That
was
on
the
thread.
E
Just
to
summarize,
everybody
was
effectively
around
consistency,
but
I
think
that
if
we
set
goals
around
just
outcomes
we
want,
rather
than
specifying
the
tools,
then
I
think
that
we
will
probably
get
where
we
need
to
be
and
be
more
flexible
as
a
community.
And
we
can
say
you
know
what
we
we
value,
consistency
or
e
value.
You
know
things
like
it
has
to
all
come
from
the
same
open,
API,
spec
and
things
like
that.
It
will
achieve
yeah.
A
C
C
A
You
know
we
started,
we
gotta
document
started.
It
doesn't
cover
this
case
super
well,
because
we
didn't
anticipate.
We
didn't
anticipate
this
question,
so
I
think
we've
been
we've
been
talking
about
this
quite
a
bit
here
and
I.
Think
something
we
need
is
what's
the
criteria
for
for
dropping
something
in
the
kubernetes,
client,
repo,
org
and
and
I.
A
Think
we're
going
to
look
at
our
client
requirements,
talk
and
see
if,
if
I
don't
know,
if
we
want
to
require,
like
you
know,
alpha
and
bronze-level
or
if
we
want
to
invent
some
slightly
lower
level
as
a
threshold,
but
we
need
to
document
will
be
what
we
want
like
I.
Don't
care
what
goes
into
the
days
incubator,
but.
C
Except
that
everything
is
going,
all
the
clients
are
outside
the
create
or
Gabonese
client
get
it
works,
I
think
people
are
people,
look
at
the
kinetic
blanket
and
what
we're
going
expected
to
host
like
the
official
clients.
Well,
regardless
of
whether
we
consider
an
official
or
not
official
or
whatever
I.
We
need
to
document
things
like
how
stable
should
they
expect
to
be
from
an
interface.
That's
what.
A
C
There's
part
of
it
is
the
badges,
but
also
part
of
it
is
deciding,
for
example,
you
know.
Maybe
we
need
a
dotnet
library,
that's
out
there,
so
people
can
get
feedback
or
or
whatever,
but
if
we're
not
that
confident
to
function,
signatures
will
say
the
same,
because
maybe
we'll
change
the
names
of
the
ones
in
the
dollar
client
or
what
timing
or
whatever
I'm
going
to
next
I
mean.
E
E
C
E
A
E
A
A
C
E
E
F
E
G
F
E
F
That
that
is
a
great
hope
but
being
realistic.
This
is
the
unexpected
thing
to
expect.
This
is
something
that
will
come
up
in
almost
every
release
for
the
next
year
or
two
I
be
shocked.
If
that
doesn't
happen,
so
something
new
that
we
did
not
receive
will
cause
it
to
be
the
case
that
your
thing
will
build
right.
Somebody's
claim,
yes,.
E
That's
what
I
mean
I
think
that's
fine
too.
That's
part
of
it
taken
on
as
part
of
the
main,
its
Bergen
just
like
bug,
fixes
and
everything
else
like
that.
I
I,
wouldn't
call
it
out
separately.
I
guess
I
feel
like
it's
a
bug
fix
like
anything
else
and
that
there's
a
la
expectation
of
a
maintainer
on
a
bug
fix.
Okay,.
A
Sure
so,
to
the
timothy's
comment
on
chat,
I
think
yeah.
We
need.
We
need
an
SDK,
API
thing,
I,
don't
think
we're
quite
there.
Yet
I
think
we're
going
to
get
there
and
1.73
these
that's
going
to
be
the
goal,
but
you've
got
a
community
dedicated.
If
you
client
answer
your
components
and
like
set
it
up
for
you
and
you
know
whether.
C
A
A
Should
at
least
you
know,
work
together
for
Theo
yeah,
okay,
how
we
sort
of
beat
the
score
stood
up
I.
Do
we
have
a
path
forward?
I
I'm
inclined
to
accept
this
I
think
we
need
to
better
define
what
the
criteria
is.
So
we
don't
have
a
big
conversation
next
time,
but
given
the
assurance
that
I'm
going
to
have
two
owners
on
it
and
kind
of
manage
tomorrow
and
the
owners
know,
they
sign
up
they're
signing
up
for
some
level
of
automation,
then
double
yoga
yeah.
C
C
E
A
C
E
C
We
have
a
so
right
now,
the
we
don't
have
a
documented,
reportable
process
for
the
project
in
general.
That's
something
that
needs
to
be
developed.
We
have
been
following
the
procedure
of
sending
PRS
to
the
design
proposals
directory
I.
Think
we're
going
to
need
more
structure
around
that,
because
we
have
no
way
of
auto
assigning
reviewers
of
those
or
anything
like
that
yeah.
We
need
to
come
up
with
something
this
Google
Docs
or
way
more
convenient
for
the
time
to
use
them.
A
A
F
F
F
C
A
You
turn
some
of
them
for
to
markdown
and
store
them.
Yeah
I
think
I,
think
I
think
it
is
in.
A
A
Okay,
so
control
plane,
no
traffic,
egress
follow-up.
You
discussed
this
last
week,
Walter
made
a
document.
David
made
a
bunch
of
comments
on
it
and
I
haven't
had
a
chance
to
look
at
it
since
then
so
I
don't
know
as
David.
Is
it
people
remember
this
topic
from
last
meeting
or
do
we
need
another
recap?
I.
A
D
I
think
we
tried
to
outline
most
of
the
major
points
having
to
do
with
you
know
with
how
would
you
like
a
Grenadier
configured
and
what
are
the
networking
requirements
for
it?
So
there
are
a
couple
of
cases
where
there
are
two
major
issues.
One
is
whether
requests
coming
out
of
the
API
server
should
be
going
to
the
service
IP
address
or
to
individual
hope.
Pod,
IP
addresses
or
I
mean
dividual
user
API
server,
and
the
other
has
to
do
with
the
use
of
dialers
I.
D
Think
I
made
a
reasonable
set
of
arguments
as
to
why
both
of
these
things
are
desirable,
depending
exact
configuration,
our
kubernetes
cluster
and
and
I,
hopefully,
I
think
David
agreed
that
you
know
all
of
these
cases
need
to
be
supported.
So
it's
mostly
a
matter
of
how
we
deal
with
sad,
sound
about
right,
David.
It.
B
Does
I
thought
it
was
a
great
docket
laid
out?
The
explanation
quite
well
had
some
comments
where
I
didn't
quite
agree
with
all
the
goals,
but
in
the
end
yeah
you're
right
in
terms
of
the
feature
it
was
largely
a
I
can
see
what
you
wanted
and
I,
even
though
I
don't
think
that
most
people
will
use
it
or
set
up
that
way.
It
is
reasonable
implement
if
we
can
keep
it
contained
and
I
think
we
can
I.
F
One
thing
I
thought
was
a
goal:
was
people
may
want
to
use
these
aggregated
a
people,
the
aggregated
API
servers
and
objects
that
are
stored
within
them
as
part
of
the
master
or
cluster
bootstrapping
process?
And
so
it's
important
that,
whatever
implementation
we
come
up
with,
not
make
it
impossible
to
do
that.
In
other
words,
we
want
things
to
be
able
to
happen
and
to
reach
aggregated
API,
so
user
API
servers
with
a
minimal
amount
of
other
infrastructure.
A
D
Service
or
entice
resources
and
I
think
the
key
part
there
and
I
think
it's
a
great
point.
That
is
that,
at
some
level
there
are
certain
parts
of
the
core
system
in
certain
of
these
use
cases
which
are
needed
to
bring
up
full
functionality
of
the
aggregator.
So
it
needs
to
be
true
that
the
aggregator
and
that
the
core
functionality
API
server,
needs
to
be
able
to
come
up
without
successfully
of
pulling
out
anything
of
the
end
users
API
server.
If
that's
true,
it
should
all
come
up
right.
F
A
Yes,
I
guess
I
think
there's
a
couple
things
here:
one
that
I
want
to
say
one
is
we're
in
VP
mode,
so
skip
the
basic
functionality
of
working
as
as
soon
as
possible.
It'll
remain
a
mr.
roat,
the
other
one
is
we
should
take
it
like
Brian
has
this
layers
document
we
should.
It
may
already
be
in
there,
but
we
should
take
a
look
at
that
and
see
where
you
were
to
user
API
service.
What?
Where
do
they
go
in,
because
that
that
basically
should
inform
the
bootstrapping
process?
Well.
C
So
in
the
architectural
roadmap
dock
I
put
the
aggregator
and
then
what
I
be
in
our
most
layer,
which
I
call
the
new
twist
yeah,
because
we
need
it
in
order
to
be
able
to
add
any
other
functionality.
So
today
the
system
is
multiple
binary,
but
effectively
final
with
it,
like
every
API
has
to
be
added
to
a
guy
server
and
controller
manager
right.
So
if
we
want
to
move
to
the
layered
model,
where
that's
not
true,
without
breaking
backward
compatibility,
then
we
need
the
aggregation
layer
in
the
innermost,
amine
or
most
layer.
C
C
Sighs
but
yeah
I
get
to
say
you
should
at
least
be
thinking
about
that
say.
Gay
should
be
thinking
about
that
I
thought
well,
the
way
I
have
it
right
now.
Is
this
core
API
machinery
and
extensibility
mechanisms
like
this
control?
Attention,
for
example,
are
going
to
be
in
the
nucleus
and
I
also
put
nodes
and
pods
and
the
things
that
they
depend
on
in
cubelets
in
the
nucleus,
because
kubernetes
isn't
for
Nettie's
without
the
ability
to
actually
execute
stuff
and
plus,
you
can't
run
any
extensions
without
being
able
to
execute
stuff.
C
The
next
layer
about
that
is
video
application
layer,
which
is
contains
both
the
api's.
You
need
to
do
things
anything
useful
and
to
have
portable
home
charts
or
other
resource
manifests,
for
example,
so
things
like
deployment
and
service
are
in
the
application
layer.
Services
non
nucleus
very
deliberately,
although
for
complicated
reasons,
endpoint
is
in
the
nucleus,
but
the
something
some
sort
of
routing
mechanism
attitude
I
disagree
with
that,
but
it
depends
on
what
you
mean
by
routing
and
then
I
mean
other
things
are
outer
layers
that
are
more
up.
You
don't.
C
A
A
A
C
A
G
Was
actually
I
got
four
hours
of
sleep,
and
that
was
all
that
really
mattered
and
I
wasn't
like
cramped
up
like
a
sardine
in
a
can
so
on
this
I
think
I,
don't
Jeremy's
on
the
call.
So
Jeremy
was
pulling
together.
The
API
changed
for
the
admission
control
webhook,
so
the
the
API
that
does
is
used
to
call
out
when
we
made
over
and
then
we
can
start
debating
that
age.
I
change
the
initializing
API
I
was
going
to
answer
some
of
the
remaining
comments
and
make.
G
A
G
You
cannot
do
parallel
validation,
so
you
can
do
linear
validation
potentially,
and
you
can't
do
updates
validation,
so
update
validation
is
actually
a
pretty
key
security
hole
like
today.
If
you
wanted
to
gain
beautiful
object,
audition,
okay,
now
not
okay,
that
I
buy
yeah
and
there's
no
way
to
do.
Update
initializers
without
forking
objects
with
sed,
because
two
simultaneous
people
coming
in
and
starting
an
edit.
You
have
to
create
two
objects
for
that
to
be
or
three
or
four
and
then
merge
them
back
in
and
that's
just
my
worship
wait.
G
G
We
we
briefly
did
we
briefly
discussed
the
fact
that
services
change
type
which
causes
problems
for
like
quoted
today,
because
quote
have
to
recalculate
on
an
update
and
quota,
can
do
that
because
it's
an
addition
controller.
So
the
the
thought
process
here
is,
we
will
probably
will
almost
certainly
need
mutating
emission
controllers.
We
will
start
with
non
mutating
emission
controllers
and
we
will
expect
to
find
a
reason
to
go.
Add
mutating
mission
controllers
for
update
the
kind
of
an
advanced
case,
yeah.
A
G
A
A
A
If
I
is
that,
like
I
would
expect
that
the
initializers
could
all
run
and
if
one
of
them
has
a
requirement
for
on
another
one,
it
can
just
wait
like
and
they
can
make
their
changes
with
patches
and
then
they
can
all
do
their
mutations
in
parallel
and
I
thought
that
that
would
be
the
whole
entire
point
of
this.
So
in
practice.
G
You
still
get
conflict,
so
we
basically
turn
updates
into
a
storm,
and
so
the
phone
would
be.
This
will
be
alpha
if
we
choose
to
relax
it,
we
can
and
actually
nothing
prevents
nothing
in
the
proposal
prevents
things
going
out
of
turn.
I
think
it
is
a
recommendation
and
you
can
decide
whether
to
enforce
that
recommendation
or
not.
Okay,.
A
A
G
A
G
Which
requires
distributing
that
logic
to
all
the
different
things
that
are
being
written
in
so
when
someone
adds
a
new
one,
they
have
to
go
set
that,
whereas
today
and
then
they
could
just
set
them
in
order,
so
it's
kind
of
a
I'm
not
against
it.
I
just
don't
know
that
it's
the
simplest
thing,
that's
that
we
start
with,
because
the
latency
isn't
really
our
problem
or
create
latency.
Isn't
our
problem
today,
okay,.
G
A
A
A
K
Hear
me
yeah,
perfect
and
so
hello.
My
name
is
on
it
I'm
from
Google,
and
this
is
my
first
cig
API
machinery
meeting
and
I'd
like
to
say
that
I
don't
know
much
about
the
API
and
I
think
that
it's
one
of
the
reasons
what
that's,
what
one
of
the
things
that
we're
going
to
talk
about
so
I'm
trying
to
create
a
proposal
for
a
samples
effort
for
the
client,
my
bridge.
K
The
reason
for
that
is
because
I
think
there
we
kind
of
assume
that
people
know
how
to
use
with
your
API
and
people
know
how,
to
you
know,
directly,
take
the
client
libraries
and
do
whatever
they
wanted
to
with
do
with
them.
However,
currently
the
case
of
that,
it
is
not
super
easy
and
we're
not
making
it
super
friendly
to
use
the
Krays
API,
so
I'm
I'm
planning
to
start
a
samples
effort
to
create
a
set
of
samples
that
exists
in
the
official
support,
show,
supported,
client,
libraries,
I
guess:
that's
currently
going
Python.
K
I
Commented
in
that
issue
and
there
I
think
what
you,
what
we
wanted
to
achieve
as
a
as
a
is
it
making
that
learning.
I
K
Purse
can't
click
on
the
link
yeah.
They
were
great.
Thank
you.
So
one
of
the
first
things
is,
you
know
we
got
some
of
the
beginning
of
tasks
already
covered
into
today's
client
libraries
and
that's
great,
but
you
know
we
have
more
advanced
tasks
like
how
do
I
use
the
informer,
so
they
actually
use
the
watch
API
and
how
does
this
work
thing
work
with
possibly
more
explanation
and
then
possibly
I'm
thinking
about
adding
some
scaffolding?
You
know
controllers
operator
schedulers,
like
basically
just
pulled
styles,
that
let
people
just
take
off
from
there.
A
Sounds
great
you
should
think
of
with
filled
with
rock.
If
you
haven't
already
had
yeah,
you
should
also
think
about.
How
are
you
going
to
represent
these
things
and
alternative
languages,
because
we
want
to
end
up
with
documentation
that
shows
you
snippets
for
basically,
every
API
call
in
every
language
that
we
have
a
client
for
yeah.
K
H
A
A
G
So
I
just
put
this
end.
We
put
the
proposal
into
two
parts.
The
underlying
details
and
the
config
slash
invitation,
slash
complicated
stuff
about
is
that
the
core
details
are
fairly
straightforward,
barring
some
tuning
or
tweaking
that
we
might
do
of
making
sure
that
there's
appropriate
room
for
future.
We
are
replacing
crypto
providers
with
different
elbows,
etc,
as
well
as
the
on
disk
format,
and
then
everything
else,
Keane's
cetera,
is
all
too
higher
plane,
so
so
trying
to
get
that
push
through
an
adequate
review
so
that
we
can
merge
the
storage
Pio
for
that.
Okay.
F
A
A
We
encountered
some
a
couple
problems
with
FTD
three
here,
as
it
has
resulted
in.
You
know
digging
attempting
to
dig
through
the
data.
Just
you
know
what
what
is
going
on
is
that
the
key
database,
so
we're
still
trying
to
figure
out
what's
going
on
I'm,
just
a
lot
of
change
stuff,
there's
just
something
else
now.
A
Brendan
senses
it
to
be
no,
it's
like
it's
like.
It
filled
up
with
free
pages,
and
we
don't
understand
why
the
bolts,
yeah
yeah
it
sounds
like
things
are
the
bull
database
I
called
an
issue
with
one
issue
with
both
both
where
the
tan
line
tool
doesn't
parse,
freeze
a
free
page
locks
correctly,
and
some
so
basically,
I
said
this
to
say
something
we're
going
to
need
is
an
analysis
program
that
parses
the
critical
database
and
prints
off.
What
is
there
and
we
need.
We
especially
need
this
after
we
start
storing
proto
yeah.
F
A
A
G
A
G
B
We
figured
that
one
out.
It
was
a
version
you
remember
when
we
had
the
issue
and
and
II
ended
up,
keep
looking
at
finding
a
problem
in
SAT
three
with
regard
to
compaction,
Windows
and
stuff.
What
so
it
looks
like
at
some
level
of
a
CD,
they
fix
it,
and
so
you
need
to
have
like
a
CD,
311
or
something
I
linked
the
issue,
and
we
found
a
lot
of
a
level
we
just
have
to
have
some
way
to
tech
help.
People.